Merge branch 'llm-error-types' into llm-remove-provider-error

Delete the provider-error LLMEvent so streams carry output only and
every provider-reported failure exits through the typed error channel.
Anthropic SSE error events, OpenAI Responses response.failed/error
events, and Bedrock exception frames now fail the stream with an
LLMError from the shared classifier (carrying the provider code, e.g.
rate_limit_exceeded -> LLM.RateLimit).

Core follows: the publisher drops its provider-error case, the runner
drops held-back overflow events (overflow recovery keys off a thrown
LLM.ContextOverflow), compaction and title stop scanning events for
failures, and isContextOverflowFailure is deleted with its last
consumer. V1 packages/opencode gets minimal compile fixes only (dead
switch case, one test stream).

.changeset/calm-services-start.md

@@ -0,0 +1,5 @@
+---
+"@opencode-ai/client": patch
+---
+
+Reuse a same-version background service when a repeated health probe succeeds instead of replacing an endpoint another client may already be using.

CONTEXT.md

@@ -1,245 +0,0 @@
-# OpenCode Session Runtime
-
-OpenCode sessions preserve durable conversational history while assembling the runtime context an agent needs to act correctly in its current environment.
-
-## Language
-
-**Model Context**:
-The complete model-visible input assembled for one **Step**, including system instructions, **Session History**, tool definitions, and step-local additions. **Instructions** are one component of Model Context, not a synonym for it.
-_Avoid_: System Context
-
-**Instructions**:
-The opaque algebra of independently refreshable typed instruction sources that render the durable instruction baseline and chronological updates shown to the model.
-_Avoid_: Model Context, System Context
-
-**Session History**:
-The projected chronological conversation selected for a **Step** after applying the active compaction boundary and interleaving derived **Instruction Updates** from the current **Instruction Epoch**.
-_Avoid_: Session Context
-
-**Instruction Source**:
-One independently read typed value within **Instructions**, represented by a stable namespaced key, canonical JSON codec, pure first/changed renderers, and an optional removal renderer.
-_Avoid_: Prompt fragment
-
-**InstructionEntry**:
-One API-managed, durable, per-Session instruction value. Its slash-free client key maps to the `api/<key>` **Instruction Source** key. Entries deliberately render to the model as mechanism-neutral `<context>` blocks: the model sees session context, not how it was attached.
-
-**InstructionDiscovery**:
-The Location-scoped service that observes ambient global and upward-project `AGENTS.md` files as one ordered aggregate **Instruction Source**.
-
-**Instruction State**:
-The Session-owned projection cache of one instruction log fold: epoch start, values at that start, current values, and the last folded sequence. It is rebuilt from durable events and never authors model-visible facts.
-
-**Instruction Update**:
-A durable `session.instructions.updated` value delta admitted at a **Safe Step Boundary**. Its model-visible System text is rendered from stored values at request assembly and is never persisted verbatim.
-_Avoid_: Correction, stored prose, raw text diff
-
-**Initial Instructions**:
-The deterministic instruction text rendered from values at the current **Instruction Epoch** start and sent as provider-cache prefix state until completed compaction moves the epoch or Session movement or committed revert resets it.
-_Avoid_: Live system prompt
-
-**Instruction Epoch**:
-The span between completed compactions. Its start is the last `session.compaction.ended` sequence, or the initial complete instruction delta when no prior epoch exists.
-
-**Instruction Values**:
-The key-to-hash map produced by folding instruction deltas in durable sequence order. Hash bodies live once in the content-addressed instruction blob store.
-
-**Unavailable Instruction Source**:
-An expected temporary inability to read an **Instruction Source** value; the runtime retains its prior effective value and emits no update, while an unavailable source blocks the initial complete delta.
-
-**Safe Step Boundary**:
-The point during Step preparation, after prior tool settlement and before durable input promotion, where instruction changes may be admitted chronologically.
-
-**Admitted Prompt**:
-A durable user input accepted into the Session inbox but not yet included in **Session History**.
-
-**Prompt Promotion**:
-The durable transition that removes an **Admitted Prompt** from pending input and appends its user message to **Session History**.
-
-**Step**:
-One logical LLM call spanning pre-flight instruction synchronization, input promotion, request build, and compaction check; the provider stream; and tool settlement.
-_Avoid_: provider turn, turn (unqualified)
-
-**Physical Attempt**:
-One actual provider request on the wire in service of a **Step**; most Steps have one Physical Attempt, while overflow-triggered compaction recovery may give one Step two.
-
-**Assistant Turn**:
-A reserved name for the not-yet-modeled unit containing all **Steps** from prompt promotion until the assistant yields the floor; do not reify it until something durable needs it.
-
-**Settlement**:
-The terminal transition for a unit of work: Step and tool settlement are durable, while drain and execution settlement are coordinator-observed.
-
-**Execution**:
-One session-scoped coordinator busy period from first wake until idle. An Execution is process-local coordination rather than a durable domain entity.
-
-**Session Drain**:
-One process-local execution span that promotes eligible input and runs required **Steps** until no immediate continuation remains. A Session Drain has no durable identity or transcript boundary.
-
-**Model Tool Output**:
-The bounded projection of a Core-executed tool result persisted in Session history and replayed to the model. A tool may shape this projection semantically, but the Tool Registry enforces the final size limit.
-
-**Managed Tool Output File**:
-A temporary file created under OpenCode's shared tool-output directory to retain complete output that was too large for Session history.
-
-**Model Request Options**:
-Provider-semantic model settings selected from the Catalog and active Session variant before the LLM protocol adapter encodes them for a provider request.
-_Avoid_: Request body, wire options
-
-**Generation Controls**:
-Provider-neutral sampling and output controls, partitioned from provider semantics and compatibility wire fields when model metadata enters the Catalog.
-
-**Native Continuation Metadata**:
-Opaque protocol-shaped data attached to assistant content and required to continue that content natively with a compatible model, such as a reasoning signature or provider-hosted item identifier.
-
-**PTY Environment**:
-The host-supplied environment overlay applied by the server when creating a PTY, observed for the request Location and resolved PTY working directory.
-
-**OpenCode Client**:
-The generated Promise and Effect APIs derived from the public `HttpApi`; **Embedded OpenCode** shares the Effect API through an in-memory `HttpClient` against the same router and handlers.
-_Avoid_: Remote client
-
-**SDK Contract IR**:
-The runtime-neutral compiled representation of the authoritative `HttpApi`, preserving encoded and decoded type projections plus transport metadata so independent SDK emitters can choose their public value model and runtime interpreter.
-
-**Embedded OpenCode**:
-A scoped in-process host that structurally extends the **OpenCode Client**, supplies an in-memory HTTP transport, and exposes additional same-process capabilities directly.
-_Avoid_: Local implementation
-
-**Page**:
-A bounded ordered result containing `items` and opaque `previous` and `next` cursor links for navigating the same query in either direction.
-_Avoid_: Response envelope
-
-## Relationships
-
-- **Instructions** is an opaque carrier composed from zero or more **Instruction Sources**.
-- **Model Context** is broader than **Instructions**. For each **Step**, the runner assembles the selected agent or provider system text, **Initial Instructions**, **Session History**, available tools, and step-local additions into one model request.
-- **Session History** persists conversational messages. The runner derives model-facing **Instruction Update** messages from value deltas and interleaves them by durable sequence; **Initial Instructions** remain separate provider-request state.
-- The runner explicitly loads and combines instruction built-ins, **InstructionDiscovery**, selected-agent skill guidance, reference guidance, MCP guidance, and **InstructionEntry** values. There is no instruction registry.
-- `Instructions.combine(...)` preserves caller order and rejects duplicate stable namespaced source keys. The runner loads its producers concurrently, then combines them in its fixed declared order.
-- Each **Instruction Source** read returns one coherent typed value, explicit removal, or temporary unavailability. `Instructions.make(...)` hides the value type so differently typed sources compose uniformly; its canonical codec defines storage and hash equivalence, while pure renderers produce first, changed, and optional removal text.
-- `Instructions.read(...)` reads every composed source concurrently and exactly once at the boundary. `Instructions.diff(...)` compares encoded-value hashes with current **Instruction Values** and returns one delta plus new blob bodies.
-- `Instructions.renderInitial(...)` renders values at the **Instruction Epoch** start. `Instructions.renderUpdate(...)` renders one hydrated delta against the values immediately before it.
-- A changed **Instruction Source** contributes its hash to one **Instruction Update**; explicit removal contributes the `"removed"` sentinel.
-- An **Instruction Update** persists only its value delta. Rendered text is derived during request assembly and excluded from compaction summaries.
-- The instruction blob insert, durable delta, and **Instruction State** advance commit atomically.
-- Changes from multiple **Instruction Sources** admitted at one safe boundary combine into one **Instruction Update**.
-- Instruction changes are sampled and admitted lazily at a **Safe Step Boundary**, never pushed asynchronously when their source changes.
-- At a **Safe Step Boundary**, prior tool results are already settled; instruction preparation completes before newly admitted user input promotes.
-- An **Admitted Prompt** is replayable pending input, not yet model-visible **Session History**.
-- **Prompt Promotion** atomically consumes the pending inbox entry and appends its model-visible user message.
-- Steering prompts promote at the next **Safe Step Boundary** while the current **Session Drain** still requires continuation. Promoting any newly admitted user input resets the selected agent's step allowance; multiple prompts promoted at one boundary reset it once.
-- A queued prompt does not promote while the current **Session Drain** requires continuation. The runner promotes one queued prompt when the Session would otherwise become idle, then reevaluates continuation before promoting another.
-- A **Session Drain** is process-local coordination rather than a durable domain entity. Durable recovery must reason from prompts, projected history, physical attempts, and tool state rather than inventing an enclosing execution identity.
-- An **Execution** contains one or more **Session Drains**; a **Session Drain** contains one reserved assistant-turn span at a time; that span contains **Steps**; and each **Step** contains one or more **Physical Attempts** plus any tool calls it requires.
-- A **Step** record covers only the model-visible span from first assistant output through tool settlement; pre-flight leaves no record, and one Step settles at most one record.
-- The first **Step** admits one complete delta and renders **Initial Instructions** without narrating that delta in history; an unavailable initial source blocks the Step instead of persisting incomplete values.
-- Instruction preparation precedes durable input promotion on every Step so an unavailable first baseline leaves pending input untouched and later updates enter history before newly promoted input.
-- Completed compaction moves the **Instruction Epoch** to the exact `session.compaction.ended` sequence and copies current hashes to the epoch's initial values. Earlier updates leave active model history while durable deltas remain.
-- A newly composed **Instruction Source** absent from current **Instruction Values** emits its first rendering once at the next **Safe Step Boundary**.
-- **Unavailable Instruction Source** uses stale-while-revalidate semantics and is distinct from a successfully loaded absence, which may emit removal text.
-- **InstructionDiscovery** observes ambient instructions as one ordered aggregate **Instruction Source**.
-- Ambient discovery reads global and upward-project `AGENTS.md` files and honors `OPENCODE_DISABLE_PROJECT_CONFIG` for project files.
-- After a successful internal file or directory read, nearby `AGENTS.md` files toward the Location root are injected once per Session as durable synthetic instruction messages.
-- **InstructionEntry** stores API-managed per-Session JSON values. Each entry contributes one `api/<key>` **Instruction Source**, so adding, replacing, or removing an entry is reconciled at the next **Safe Step Boundary**.
-- Location-scoped instruction producers naturally re-resolve when a moved Session next runs in its destination Location.
-- Moving a Session clears its **Instruction State**, so the destination must admit a complete delta before another prompt can promote. Committed revert does the same; replay derives both resets from their durable events.
-- Selected-agent available-skill guidance is an **Instruction Source** composed explicitly by the runner. It lists only names and descriptions permitted for that agent; skill bodies and locations are exposed only through the permission-checked `skill` tool.
-- The selected agent and model are sampled when a **Step** starts. Changes admitted after that boundary apply to the next Step and do not restart the current Step.
-- An agent switch that changes selected-agent guidance produces an **Instruction Update** while preserving the current baseline.
-- Local tool authorization and pending permission requests retain the effective agent of the **Step** that issued the call; a later agent switch cannot change that call's policy.
-- Instruction source changes never wake idle Sessions; the next naturally scheduled **Safe Step Boundary** loads and compares current values lazily.
-- Once admitted, an **Instruction Update** remains durable even if the following **Physical Attempt** fails and is replayed unchanged on retry.
-- **Instruction Updates** remain durable value history but are not `session_message` rows. Clients display changed keys rather than model-facing prose.
-- The date **Instruction Source** initially preserves host-local calendar-date behavior; a configured user timezone may replace that default later.
-- **Initial Instructions** are recomputed deterministically from durable values for every request; rendered bytes are not stored.
-- A model/provider switch preserves current **Instruction Values**, the **Instruction Epoch**, and chronological conversation history; the new selection applies to the next **Step**.
-- **Native Continuation Metadata** remains in durable history. Step projection includes it only for a successful exact originating provider/model match; failed Steps and incompatible models omit opaque metadata, while non-empty visible reasoning lowers to ordinary assistant text after a model switch. This conservative relation may widen only when recorded provider tests establish compatibility.
-- **Model Request Options** remain provider-semantic through Catalog resolution. The Session runner maps them into the LLM package's provider-option namespace; the selected protocol adapter alone owns provider wire encoding.
-- **Generation Controls**, protocol-semantic **Model Request Options**, and compatibility request body fields are separate Catalog domains. A shared ingestion adapter partitions legacy and models.dev AI-SDK-shaped options before routing.
-- The **PTY Environment** is a server concern rather than a Core PTY concern. PTY creation merges caller values, then the host overlay, then Core-forced terminal invariants such as `TERM` and `OPENCODE_TERMINAL`.
-- Networked and **Embedded OpenCode** use the same **OpenCode Client** and preserve the full HTTP encoding, routing, middleware, and decoding boundary; only the `HttpClient` transport differs.
-- The Effect-native network constructor obtains `HttpClient.HttpClient` from its environment so callers own transport selection, recording, tracing, retries, and tests. Convenience runtimes may provide a fetch transport separately.
-- Creating **Embedded OpenCode** is scoped. Closing its owning Scope releases the in-process server resources, database resources, registrations, and fibers.
-- **Embedded OpenCode** exposes shared client capabilities and embedded-only capabilities on one object; consumers do not navigate through a nested `.client` property.
-- The beta **OpenCode Client** currently uses plural consumer-facing capability groups such as `sessions`; whether the stable Session namespace should instead be singular `session` must be settled before stabilization. Internal server identifiers do not implicitly define public client names.
-- Server's concrete `HttpApi` is authoritative for shared **OpenCode Client** capabilities. Codegen compiles its Session group directly; the Effect runtime uses an equivalent Protocol-only projection so generated artifacts remain independent of Core and Server.
-- SDK generation reflects the public `HttpApi` once into an **SDK Contract IR**. Promise and Effect emitters share endpoint structure and transport metadata without being required to expose identical public values: an emitter may select encoded wire types, decoded domain types, compile-time brands, runtime validation, and its own execution abstraction independently.
-- The first Effect emitter is the rich projection: it exposes decoded Effect-native values, preserves brands and schema transformations, performs runtime schema decoding, and delegates transport interpretation to `HttpApiClient`. Lighter wire-shaped Effect output remains possible through another emitter policy rather than constraining the shared IR.
-- The rich Effect emitter regenerates private executable schemas when the **SDK Contract IR** proves that their transport semantics can be reproduced exactly. Contracts with authoritative custom transformations use the import-based Effect emitter against a Protocol-only client projection whose generated transport output is tested against Server's concrete API; the Promise emitter still derives zero-Effect structural wire types from the same IR.
-- `@opencode-ai/protocol` owns Session endpoint construction and middleware placement. Server supplies concrete middleware keys to produce the authoritative build-time API; the client projection supplies transport-only keys without importing Core or Server at runtime.
-- The first Promise emitter targets the same clean domain-oriented method organization rather than Hey API source compatibility. It returns unwrapped values directly, rejects declared and infrastructure failures, and begins with minimal client-level transport configuration; result wrappers, interceptors, and legacy generated signatures are outside the initial surface.
-- The first Promise emitter parses response syntax and trusts its generated structural types; it does not perform runtime structural validation. Malformed payload syntax fails, while a syntactically valid shape mismatch is not detected at the SDK boundary. Standalone validator generation remains an optional future emitter policy.
-- Declared Promise-client failures retain their tagged structural wire values and have generated type guards. Consumers do not depend on generated `Error` subclass identity, preserving discrimination across package copies and realms while remaining structurally aligned with Effect domain errors.
-- Promise-client infrastructure failures use one generated `ClientError` class with a structured reason such as transport failure, unexpected status, unsupported content type, or malformed response. Promise methods reject with either a tagged declared domain failure or `ClientError`, matching the Effect client's conceptual domain/infrastructure error division.
-- Promise methods accept a separate optional per-call transport-options argument containing `AbortSignal` and header overrides. Cancellation and transport metadata do not enter the domain input object; broader interceptor and response-mode APIs remain deferred.
-- Promise streaming methods return a lazy `AsyncIterable` directly rather than a Promise-wrapped stream object. Iteration opens the connection, `AbortSignal` cancels it, and ending iteration closes the underlying request; the Effect emitter analogously returns `Stream` directly.
-- Promise SSE connection establishment, declared HTTP failures, and infrastructure failures occur during `AsyncIterable` iteration, beginning with its first `next()` call, rather than during synchronous method construction.
-- Neither generated streaming runtime automatically reconnects after disconnection. Promise `AsyncIterable` and Effect `Stream` fail explicitly; live consumers refresh and resubscribe, while durable sequence-based resume remains explicit composition above the generated client.
-- Promise client construction is synchronous and network-free. It requires `baseUrl`, defaults to `globalThis.fetch`, accepts client-level headers, and merges them with per-call header overrides.
-- Effect client construction accepts an explicit `baseUrl` and obtains `HttpClient.HttpClient` from the Effect environment. It does not install fetch or duplicate per-call transport policy; callers transform/provide the client for headers, tracing, retries, recording, and tests, while fiber interruption owns cancellation.
-- Promise and Effect emitters each own their generated public type modules. The **SDK Contract IR**, not a physically shared generated type package, is the common source; this permits zero-Effect wire types and rich decoded Effect types to evolve independently.
-- Promise and Effect network clients ship from `@opencode-ai/client` behind isolated root and `/effect` exports. The root has no runtime path to Effect; `/effect` imports only Effect, Schema, and Protocol.
-- The Effect-native scoped host belongs to `@opencode-ai/sdk-next`, which will assume the existing `@opencode-ai/sdk` name after legacy consumers migrate. Client remains network-only and SDK depends one-way on Client.
-- SDK executes Server's assembled `HttpRouter` in memory. It opens no listener and performs no network I/O, while preserving Server routing, middleware, codecs, handlers, and errors.
-- The Effect Client and SDK re-export their decoded datatype facade from Schema so callers do not depend on internal package locations or Core's versioned names.
-- A capability intended for both networked and **Embedded OpenCode** belongs in the authoritative public `HttpApi`; embedded-only same-process capabilities extend **Embedded OpenCode** separately.
-- `sessions.log({ sessionID, after, follow })` is the public durable Session event stream. It verifies the Session, replays durable events after the optional aggregate sequence, optionally continues with newly committed durable events, excludes live-only fragments, and is transported as SSE in both networked and embedded modes.
-- `events.subscribe()` is a distinct public instance-wide live stream for Session and non-Session activity. It has no replay guarantee and includes connection, heartbeat, and instance-disposal lifecycle events; consumers recover from disconnection by refreshing authoritative state.
-- A Session ID is not an optional filter on `events.subscribe()`: instance-wide live events and durable Session events have different schemas, replay guarantees, cursors, lifecycle events, and failure behavior.
-- The initial common OpenCode Client does not expose server-global event aggregation. `events.subscribe()` is bounded to the connected OpenCode instance or workspace; any future cross-instance administrative stream requires a separately designed API.
-- `events.subscribe()` does not automatically reconnect after transport loss. The live-only stream fails with `ClientError`; consumers refresh authoritative state before explicitly opening a new subscription because events missed during disconnection cannot be replayed.
-- `sessions.log({ sessionID, after, follow })` returns the generated HTTP client's cold durable event stream and does not build reconnection policy into the endpoint or client constructor. Transport loss fails the stream with `ClientError`. Callers may compose an explicit resuming stream above it by retaining the last observed durable sequence and opening a new subscription with `after`; any reusable resume helper remains a separate API design question.
-- The stable `sessions.list(...)` design returns a **Page** in both networked and **Embedded OpenCode**; embedded execution does not define a separate unbounded array-returning list operation. The beta client currently preserves the existing HTTP `{ data, cursor }` envelope until emitter-level Page projection is implemented.
-- Session list cursors are opaque branded values carrying continuation query and ordering state. Consumers pass them back unchanged and do not inspect storage anchors or encoded filter fields.
-- A Session list continuation accepts only its opaque cursor. Scope, filters, ordering, and page size are fixed by the initial query and carried by that cursor.
-- `sessions.messages(...)` returns a **Page** and uses the same cursor discipline as `sessions.list(...)`: the initial request supplies `sessionID`, ordering, and page size; continuation supplies `sessionID` plus only an opaque branded message cursor carrying ordering, page size, direction, and message anchor. Using a cursor with another Session is invalid.
-- `sessions.message({ sessionID, messageID })` is a required resource lookup. An unknown Session fails with `SessionNotFoundError`; a known Session with an absent or differently owned message fails with `MessageNotFoundError` without disclosing cross-Session ownership. Absence is not represented as `undefined` across the public HTTP boundary.
-- `sessions.interrupt({ sessionID })` first verifies that the durable Session exists, failing with `SessionNotFoundError` otherwise. For a known Session, interruption is idempotent: idle, already-settled, or locally unowned execution is a no-op.
-- `sessions.active()` snapshots the current process's foreground Session drain registry as a record of Session IDs to `{ type: "running" }`. Missing IDs are inactive; background subagents and tasks do not make their parent Session active, and process restart clears the registry.
-- `sessions.context({ sessionID })` preserves the existing message-only operation. It returns projected **Session History**; it does not include or represent the complete **Model Context**, whose agent system text, **Initial Instructions**, tools, and step-local additions remain separate.
-- **Open question**: Should a future, separately named operation expose complete **Model Context**, including the instruction baseline, applied instruction metadata, tools, and step-local additions?
-- `sessions.prompt(...)` exposes `resume?: boolean`. Omitting it preserves durable admission followed by an advisory execution wake; `resume: false` requests durable admit-only behavior.
-- The public operation remains `sessions.prompt(...)`; `SessionPending.admit` is the internal primitive, while the public `Admission` result and `resume` option express its durable admission semantics.
-- `sessions.create(...)` accepts an optional `location`. Omission resolves through the connected OpenCode instance's default or current location; an explicit value selects a known location. Networked and embedded transports use the same handler semantics.
-- `sessions.switchAgent({ sessionID, agent })` is part of the common client alongside `sessions.switchModel(...)`. It affects subsequent Session activity and fails with `SessionNotFoundError` for an unknown Session.
-- The **Embedded OpenCode** Layer delegates to the same scoped creation path; it does not define a second implementation.
-- A **PTY Environment** adapter observes plugins in the request Location while passing the resolved PTY working directory to the hook; standalone servers use an empty adapter.
-- An **Instruction Update** lowers to the provider's native chronological instruction role when supported and to a wrapped chronological fallback otherwise.
-- When the aggregate discovered instruction set changes, its **Instruction Update** includes the complete current ordered set and supersedes the prior aggregate value; when no discovered instructions remain, the message states that previously loaded instructions no longer apply.
-- Ambient project instruction discovery honors `OPENCODE_DISABLE_PROJECT_CONFIG`; global instructions remain eligible.
-- Oversized textual **Model Tool Output** retains a bounded preview in Session history while its complete text moves to managed tool-output storage. Arbitrary structured-result size is a separate concern.
-- One tool settlement receives one aggregate textual limit, using the configured maximum lines or UTF-8 bytes, whichever is reached first. The limit is provider-independent; token pressure belongs to context assembly and compaction.
-- Generic truncation preserves the beginning and end of textual output. Tools may apply a more meaningful strategy before the Tool Registry enforces the final limit.
-- A truncated **Model Tool Output** identifies its complete text in the bounded model-visible preview. The Tool Registry also supplies managed paths as internal metadata to tool hooks; Session events do not expose a typed `outputPaths` field.
-- A **Managed Tool Output File** is temporary and may expire after its retention period. The bounded **Model Tool Output**, not the file, is the durable replayable record.
-- Failure to retain a **Managed Tool Output File** fails settlement operationally. The Session never publishes a successful result whose complete output was lost during generic bounding.
-- Once a tool operation succeeds, bounding its **Model Tool Output** and publishing its one durable settlement form an interruption-safe completion region. Raw oversized success is never published before a later correction.
-- When a structured-only result would exceed the **Model Tool Output** limit, its validated structured value remains unchanged for Session consumers while model replay uses a bounded textual JSON preview and optional managed output path.
-- Existing tool-managed output paths survive generic bounding. A fallback file retains exactly the complete projected text received by the Tool Registry and never claims to reconstruct output already discarded by tool-specific shaping.
-- **Managed Tool Output Files** use globally unique names in one shared flat directory. They receive no special filesystem authority; each tool applies its ordinary external-path policy.
-- Provider-executed tool results remain provider-native transcript facts outside generic Tool Registry bounding. Their context control requires provider-aware pruning or compaction because some providers require exact structured round-trip payloads.
-
-## Client contract architecture
-
-Semantic values that mean the same thing internally and publicly live in the lightweight Schema leaf. Core consumes Schema for domain behavior; Protocol composes Schema values into paths, payloads, envelopes, errors, cursors, and streams; Server imports both, hosts Protocol's exact groups, and owns protocol/domain adaptation. The root Promise client remains zero-Effect, `/effect` depends on Effect plus Schema and Protocol, and `@opencode-ai/sdk-next` composes the scoped in-process host above Client, Core, and Server.
-
-Shared public records are plain objects declared with `Schema.Struct`. A same-name inferred interface gives object records readable TypeScript signatures without constructors, prototypes, or nominal identity; unions retain explicit type aliases.
-
-Before stabilizing the client API:
-
-- Keep additional public schemas in Schema and additional network groups in Protocol; neither package may transitively load databases, Drizzle, Session execution, providers, watchers, native modules, or WASM.
-- Keep concrete Location middleware keys in Server while Protocol owns their placement. Client projections may supply transport-only keys, but must prove generated equivalence with Server's concrete API.
-- Project the existing list response envelope to the stable client **Page** shape and enforce separate initial-query and cursor-continuation inputs without changing the hosted V2 wire contract.
-- Settle the stable consumer namespace (`session` versus the current beta `sessions`) and use an explicit codegen annotation if the consumer name should differ from the server group identifier.
-- Preserve V2 route paths, operation IDs, codecs, errors, middleware behavior, and OpenAPI output while making this change.
-- Preserve browser-safe `@opencode-ai/client` and `@opencode-ai/client/effect` bundles through import-boundary tests.
-- Define embedded-host placement before supporting multiple hosts over one database. Hosts that share durable Session storage must also share process-local Session execution coordination, or each host must receive isolated storage explicitly.
-- Keep an embedded request scope alive until any streamed response body finishes. The initial non-streaming Session surface does not exercise this lifetime boundary; Session and instance event streams must do so before joining the embedded client.
-
-## Example dialogue
-
-> **Dev:** "The date changed while the session was active. Should the **Instruction Update** say what the old date was?"
-> **Domain expert:** "No. Emit the newly effective date so the agent can act on the current instructions."
-
-## Flagged ambiguities
-
-- Legacy `experimental.chat.system.transform` can mutate assembled system text arbitrarily, but V2 plugins do not yet expose an equivalent hook. Decide separately whether to port it, model dynamic uses as explicit **Instruction Sources**, or narrow its semantics.

package.json

@@ -12,6 +12,7 @@
     "dev:web": "bun --cwd packages/app dev",
     "dev:console": "ulimit -n 10240 2>/dev/null; bun run --cwd packages/console/app dev",
     "dev:stats": "bun sst shell --stage=production -- bun run --cwd packages/stats/app dev",
+    "dev:www": "bun run --cwd packages/www dev",
     "dev:storybook": "bun --cwd packages/storybook storybook",
     "lint": "oxlint",
     "lint:effect-patterns": "ast-grep scan -c script/ast-grep/sgconfig.yml packages/core/src packages/server/src packages/protocol/src packages/cli/src",
@@ -102,6 +103,8 @@
   "devDependencies": {
     "@actions/artifact": "5.0.1",
     "@ast-grep/cli": "0.44.0",
+    "@types/react": "19.2.17",
+    "@types/react-dom": "19.2.3",
     "@tsconfig/bun": "catalog:",
     "@types/mime-types": "3.0.1",
     "@typescript/native-preview": "catalog:",

packages/app/src/i18n/en.ts

@@ -65,8 +65,8 @@ export const dict = {
   "command.message.next.description": "Go to the next user message",
   "command.model.choose": "Choose model",
   "command.model.choose.description": "Select a different model",
-  "command.mcp.toggle": "Toggle MCPs",
-  "command.mcp.toggle.description": "Toggle MCPs",
+  "command.mcp.toggle": "Manage MCP servers",
+  "command.mcp.toggle.description": "Enable or disable MCP servers",
   "command.agent.cycle": "Cycle agent",
   "command.agent.cycle.description": "Switch to the next agent",
   "command.agent.cycle.reverse": "Cycle agent backwards",
@@ -307,9 +307,9 @@ export const dict = {
   "prompt.toast.promptSendFailed.title": "Failed to send prompt",
   "prompt.toast.promptSendFailed.description": "Unable to retrieve session",
 
-  "dialog.mcp.title": "MCPs",
+  "dialog.mcp.title": "MCP servers",
   "dialog.mcp.description": "{{enabled}} of {{total}} enabled",
-  "dialog.mcp.empty": "No MCPs configured",
+  "dialog.mcp.empty": "No MCP servers configured",
 
   "dialog.lsp.empty": "LSPs auto-detected from file types",
   "dialog.plugins.empty": "Plugins configured in opencode.json",

packages/cli/package.json

@@ -45,6 +45,7 @@
     "@parcel/watcher": "2.5.1",
     "effect": "catalog:",
     "fuzzysort": "catalog:",
+    "immer": "11.1.4",
     "jsonc-parser": "3.3.1",
     "open": "10.1.2",
     "opentui-spinner": "catalog:",

packages/cli/src/commands/handlers/default.ts

@@ -4,7 +4,7 @@ import { run } from "@opencode-ai/tui"
 import { loadBuiltinPlugins } from "@opencode-ai/tui/builtins"
 import { Commands } from "../commands"
 import { Runtime } from "../../framework/runtime"
-import { TuiConfig } from "../../tui-config"
+import { Config } from "../../config"
 import { Effect, Option } from "effect"
 import { Server } from "../../services/server"
 import { Updater } from "../../services/updater"
@@ -35,13 +35,18 @@ export default Runtime.handler(Commands, (input) =>
       ),
     )
     preflight.loading()
-    const config = yield* TuiConfig.load()
+    const config = yield* Config.Service
     let disposeSlots: (() => void) | undefined
-    const runFork = Effect.runForkWith(yield* Effect.context())
+    const context = yield* Effect.context()
+    const runFork = Effect.runForkWith(context)
+    const runPromise = Effect.runPromiseWith(context)
     yield* run({
       server,
       args: { continue: input.continue, sessionID: Option.getOrUndefined(input.session) },
-      config,
+      config: {
+        get: () => runPromise(config.get()),
+        update: (update) => runPromise(config.update(update)),
+      },
       terminalHandoff: () => preflight.finish(),
       log: (level, message, tags) => {
         const effect =

packages/cli/src/config/config.ts

@@ -0,0 +1,109 @@
+export * as Config from "./config"
+
+import { Global } from "@opencode-ai/core/global"
+import { Context, Effect, FileSystem, Layer, Option, Schema, Semaphore } from "effect"
+import { produce, type Draft } from "immer"
+import { applyEdits, modify, parse, type ParseError } from "jsonc-parser"
+import path from "path"
+import { ConfigMigration } from "./migrate"
+import { Info } from "./schema"
+
+export * from "./schema"
+
+export interface Interface {
+  readonly path: string
+  readonly get: () => Effect.Effect<Info>
+  readonly update: (update: (draft: Draft<Info>) => void) => Effect.Effect<Info, Error>
+}
+
+export class Service extends Context.Service<Service, Interface>()("@opencode/cli/config/Config") {}
+
+const decode = Schema.decodeUnknownOption(Info)
+const decodeRecord = Schema.decodeUnknownOption(Schema.Record(Schema.String, Schema.Any))
+const empty: Info = {}
+
+export const layer = Layer.effect(
+  Service,
+  Effect.gen(function* () {
+    const fs = yield* FileSystem.FileSystem
+    const global = yield* Global.Service
+    const file = path.join(global.config, "cli.json")
+    const lock = yield* Semaphore.make(1)
+
+    const readJson = Effect.fnUntraced(function* () {
+      const text = yield* fs.readFileString(file).pipe(Effect.catch(() => Effect.succeed(undefined)))
+      if (text === undefined) return undefined
+      const errors: ParseError[] = []
+      const value: any = parse(text, errors, { allowTrailingComma: true })
+      if (errors.length) return undefined
+      return Option.getOrUndefined(decodeRecord(value))
+    })
+
+    const write = Effect.fnUntraced(function* (text: string) {
+      const temp = file + ".tmp"
+      yield* fs.makeDirectory(path.dirname(file), { recursive: true })
+      yield* fs.writeFileString(temp, text, { mode: 0o600 })
+      yield* fs.rename(temp, file)
+    })
+
+    const migrate = ConfigMigration.run({ file, config: global.config, state: global.state }).pipe(
+      Effect.provideService(FileSystem.FileSystem, fs),
+    )
+
+    const get = Effect.fn("cli.config.get")(function* () {
+      yield* migrate.pipe(Effect.catchCause((cause) => Effect.logWarning("failed to migrate cli config", { cause })))
+      return Option.getOrElse(decode(yield* readJson()), () => empty)
+    })
+
+    const update = Effect.fn("cli.config.update")((update: (draft: Draft<Info>) => void) =>
+      lock
+        .withPermits(1)(
+          Effect.gen(function* () {
+            yield* migrate
+            const current = Option.getOrElse(decode(yield* readJson()), () => empty)
+            const next = produce(current, update)
+            const edits = changes(current, next)
+            if (!edits.length) return current
+            const text = yield* fs.readFileString(file).pipe(Effect.catch(() => Effect.succeed("{}")))
+            const updated = edits.reduce(
+              (text, edit) =>
+                applyEdits(
+                  text,
+                  modify(text, edit.path, edit.value, { formattingOptions: { tabSize: 2, insertSpaces: true } }),
+                ),
+              text,
+            )
+            const errors: ParseError[] = []
+            const config = Option.getOrUndefined(decode(parse(updated, errors, { allowTrailingComma: true })))
+            if (errors.length || config === undefined) return yield* Effect.fail(new Error("Invalid CLI config update"))
+            yield* write(updated.endsWith("\n") ? updated : updated + "\n")
+            return config
+          }),
+        )
+        .pipe(Effect.mapError((cause) => new Error("Failed to update CLI config", { cause }))),
+    )
+
+    return Service.of({ path: file, get, update })
+  }),
+)
+
+type Edit = { readonly path: (string | number)[]; readonly value: any }
+
+function changes(before: any, after: any, path: (string | number)[] = []): Edit[] {
+  if (Object.is(before, after)) return []
+  if (
+    before !== null &&
+    after !== null &&
+    typeof before === "object" &&
+    typeof after === "object" &&
+    !Array.isArray(before) &&
+    !Array.isArray(after)
+  ) {
+    return [...new Set([...Object.keys(before), ...Object.keys(after)])].flatMap((key) => {
+      if (!(key in after)) return [{ path: [...path, key], value: undefined }]
+      if (!(key in before)) return [{ path: [...path, key], value: after[key] }]
+      return changes(before[key], after[key], [...path, key])
+    })
+  }
+  return [{ path, value: after }]
+}

packages/cli/src/config/index.ts

@@ -0,0 +1 @@
+export * as Config from "./config"

packages/cli/src/config/migrate.ts

@@ -0,0 +1,142 @@
+export * as ConfigMigration from "./migrate"
+
+import { TuiConfigV1 } from "@opencode-ai/tui/config/v1"
+import { Effect, FileSystem, Option, Schema } from "effect"
+import { parse, type ParseError } from "jsonc-parser"
+import path from "path"
+import type { Info } from "./schema"
+
+const decodeV1 = Schema.decodeUnknownOption(TuiConfigV1.Info)
+const decodeRecord = Schema.decodeUnknownOption(Schema.Record(Schema.String, Schema.Any))
+
+export const run = Effect.fn("cli.config.migrate")(function* (input: {
+  readonly file: string
+  readonly config: string
+  readonly state: string
+}) {
+  const fs = yield* FileSystem.FileSystem
+  if (yield* fs.exists(input.file).pipe(Effect.orElseSucceed(() => false))) return
+
+  const legacyValue = yield* readJson(path.join(input.config, "tui.json"))
+  const legacy = Option.getOrUndefined(decodeV1(legacyValue))
+  const kv = yield* readJson(path.join(input.state, "kv.json"))
+  const migrated = migrateV1(legacy, kv ?? {})
+  if (!Object.keys(migrated).length) return
+
+  const temp = input.file + ".tmp"
+  yield* fs.makeDirectory(path.dirname(input.file), { recursive: true })
+  yield* fs.writeFileString(temp, JSON.stringify(migrated, null, 2) + "\n", { mode: 0o600 })
+  yield* fs.rename(temp, input.file)
+  yield* Effect.logInfo("migrated cli config", {
+    from: [
+      legacyValue === undefined ? undefined : path.join(input.config, "tui.json"),
+      kv === undefined ? undefined : path.join(input.state, "kv.json"),
+    ].filter(Boolean),
+    to: input.file,
+  })
+})
+
+export function migrateV1(legacy: TuiConfigV1.Info | undefined, kv: Record<string, any>): Info {
+  const plugins = [
+    ...(legacy?.plugin?.map((plugin) =>
+      typeof plugin === "string" ? plugin : { package: plugin[0], options: plugin[1] },
+    ) ?? []),
+    ...Object.entries(legacy?.plugin_enabled ?? {}).map(([id, enabled]) => (enabled ? id : `-${id}`)),
+  ]
+  const themeName = legacy?.theme ?? kv.theme
+  const themeMode = kv.theme_mode_lock
+  const attentionSoundPack = kv.attention_sound_pack
+  const diffView = kv.diff_viewer_view ?? (legacy?.diff_style === "stacked" ? "unified" : undefined)
+  const thinking =
+    kv.thinking_mode ??
+    (kv.thinking_visibility === undefined ? undefined : kv.thinking_visibility ? "show" : "hide")
+
+  return {
+    ...(themeName !== undefined || themeMode !== undefined
+      ? { theme: { ...(themeName === undefined ? {} : { name: themeName }), ...(themeMode === undefined ? {} : { mode: themeMode }) } }
+      : {}),
+    ...(legacy?.keybinds === undefined ? {} : { keybinds: legacy.keybinds }),
+    ...(plugins.length ? { plugins } : {}),
+    ...(legacy?.leader_timeout === undefined ? {} : { leader: { timeout: legacy.leader_timeout } }),
+    ...(legacy?.scroll_speed === undefined && legacy?.scroll_acceleration?.enabled === undefined
+      ? {}
+      : {
+          scroll: {
+            ...(legacy.scroll_speed === undefined ? {} : { speed: legacy.scroll_speed }),
+            ...(legacy.scroll_acceleration?.enabled === undefined
+              ? {}
+              : { acceleration: legacy.scroll_acceleration.enabled }),
+          },
+        }),
+    ...(legacy?.attention === undefined && attentionSoundPack === undefined
+      ? {}
+      : {
+          attention: {
+            ...legacy?.attention,
+            ...(attentionSoundPack === undefined ? {} : { sound_pack: attentionSoundPack }),
+          },
+        }),
+    ...(legacy?.diff_style === undefined &&
+    kv.diff_wrap_mode === undefined &&
+    kv.diff_viewer_show_file_tree === undefined &&
+    kv.diff_viewer_single_patch === undefined &&
+    diffView === undefined
+      ? {}
+      : {
+          diffs: {
+            ...(kv.diff_wrap_mode === undefined ? {} : { wrap: kv.diff_wrap_mode }),
+            ...(kv.diff_viewer_show_file_tree === undefined ? {} : { tree: kv.diff_viewer_show_file_tree }),
+            ...(kv.diff_viewer_single_patch === undefined ? {} : { single: kv.diff_viewer_single_patch }),
+            ...(diffView === undefined ? {} : { view: diffView }),
+          },
+        }),
+    ...(kv.terminal_title_enabled === undefined ? {} : { terminal: { title: kv.terminal_title_enabled } }),
+    ...(kv.file_context_enabled === undefined && kv.paste_summary_enabled === undefined
+      ? {}
+      : {
+          prompt: {
+            ...(kv.file_context_enabled === undefined ? {} : { editor: kv.file_context_enabled }),
+            ...(kv.paste_summary_enabled === undefined
+              ? {}
+              : { paste: kv.paste_summary_enabled ? ("compact" as const) : ("full" as const) }),
+          },
+        }),
+    ...(kv.sidebar === undefined &&
+    kv.scrollbar_visible === undefined &&
+    thinking === undefined &&
+    kv.exploration_grouping === undefined
+      ? {}
+      : {
+          session: {
+            ...(kv.sidebar === undefined ? {} : { sidebar: kv.sidebar }),
+            ...(kv.scrollbar_visible === undefined ? {} : { scrollbar: kv.scrollbar_visible }),
+            ...(thinking === undefined ? {} : { thinking }),
+            ...(kv.exploration_grouping === undefined
+              ? {}
+              : { grouping: kv.exploration_grouping ? ("auto" as const) : ("none" as const) }),
+          },
+        }),
+    ...(kv.tips_hidden === undefined && kv.dismissed_getting_started === undefined
+      ? {}
+      : {
+          hints: {
+            ...(kv.tips_hidden === undefined ? {} : { tips: !kv.tips_hidden }),
+            ...(kv.dismissed_getting_started === undefined
+              ? {}
+              : { onboarding: !kv.dismissed_getting_started }),
+          },
+        }),
+    ...(kv.animations_enabled === undefined ? {} : { animations: kv.animations_enabled }),
+    ...(legacy?.mouse === undefined ? {} : { mouse: legacy.mouse }),
+  }
+}
+
+const readJson = Effect.fnUntraced(function* (target: string) {
+  const fs = yield* FileSystem.FileSystem
+  const text = yield* fs.readFileString(target).pipe(Effect.catch(() => Effect.succeed(undefined)))
+  if (text === undefined) return undefined
+  const errors: ParseError[] = []
+  const value: any = parse(text, errors, { allowTrailingComma: true })
+  if (errors.length) return undefined
+  return Option.getOrUndefined(decodeRecord(value))
+})

packages/cli/src/config/schema.ts

@@ -0,0 +1,5 @@
+import { Config } from "@opencode-ai/tui/config"
+import { Schema } from "effect"
+
+export const Info = Schema.Struct({ ...Config.Info.fields })
+export type Info = Schema.Schema.Type<typeof Info>

packages/cli/src/framework/runtime.ts

@@ -3,6 +3,7 @@ import { Command } from "effect/unstable/cli"
 import { Spec } from "./spec"
 import { Global } from "@opencode-ai/core/global"
 import { Updater } from "../services/updater"
+import { Config } from "../config"
 
 export type Input<Value> =
   Value extends Spec.Node<infer _Name, infer Command, infer _Commands>
@@ -13,18 +14,26 @@ export type Input<Value> =
 
 type RuntimeHandler = (
   input: unknown,
-) => Effect.Effect<void, unknown, FileSystem.FileSystem | Global.Service | Updater.Service | Scope.Scope>
+) => Effect.Effect<
+  void,
+  unknown,
+  FileSystem.FileSystem | Global.Service | Updater.Service | Config.Service | Scope.Scope
+>
 type Loader<Node extends Spec.Any> = () => Promise<{
   default: (
     input: Input<Node>,
-  ) => Effect.Effect<void, any, FileSystem.FileSystem | Global.Service | Updater.Service | Scope.Scope>
+  ) => Effect.Effect<
+    void,
+    any,
+    FileSystem.FileSystem | Global.Service | Updater.Service | Config.Service | Scope.Scope
+  >
 }>
 type ProvidedCommand = Command.Command<
   string,
   unknown,
   unknown,
   unknown,
-  FileSystem.FileSystem | Global.Service | Updater.Service | Scope.Scope
+  FileSystem.FileSystem | Global.Service | Updater.Service | Config.Service | Scope.Scope
 >
 
 export type Handlers<Node extends Spec.Any> = keyof Node["commands"] extends never

packages/cli/src/index.ts

@@ -11,6 +11,7 @@ import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
 import { Global } from "@opencode-ai/core/global"
 import { AppProcess } from "@opencode-ai/core/process"
+import { Config } from "./config"
 
 const Handlers = Runtime.handlers(Commands, {
   $: () => import("./commands/handlers/default"),
@@ -51,6 +52,7 @@ Effect.logInfo("cli starting", {
 }).pipe(
   Effect.flatMap(() => Runtime.run(Commands, Handlers, { version: InstallationVersion })),
   Effect.annotateLogs({ role: "cli" }),
+  Effect.provide(Config.layer),
   Effect.provide(Updater.layer),
   Effect.provide(AppNodeBuilder.build(LayerNode.group([Global.node, AppProcess.node]))),
   Effect.provide(Observability.layer),

packages/cli/src/mini/footer.ts

@@ -627,7 +627,6 @@ export class RunFooter implements FooterApi {
 
     this.themes.splice(index, 1)
     theme.block.syntax?.destroy()
-    theme.block.subtleSyntax?.destroy()
   }
 
   public close(): void {
@@ -1023,7 +1022,6 @@ export class RunFooter implements FooterApi {
     void resolveRunTheme(this.renderer).then((theme) => {
       if (this.isGone) {
         theme.block.syntax?.destroy()
-        theme.block.subtleSyntax?.destroy()
         return
       }
 

packages/cli/src/mini/scrollback.shared.ts

@@ -6,11 +6,7 @@ function syntax(style?: SyntaxStyle): SyntaxStyle {
   return style ?? SyntaxStyle.fromTheme([])
 }
 
-export function entrySyntax(commit: StreamCommit, theme: RunTheme): SyntaxStyle {
-  if (commit.kind === "reasoning") {
-    return syntax(theme.block.subtleSyntax ?? theme.block.syntax)
-  }
-
+export function entrySyntax(theme: RunTheme): SyntaxStyle {
   return syntax(theme.block.syntax)
 }
 

packages/cli/src/mini/scrollback.surface.ts

@@ -143,7 +143,7 @@ export class RunScrollbackStream {
     }
 
     active.renderable.fg = entryColor(active.commit, theme)
-    active.renderable.syntaxStyle = entrySyntax(active.commit, theme)
+    active.renderable.syntaxStyle = entrySyntax(theme)
   }
 
   private createEntry(commit: StreamCommit, body: ActiveBody): ActiveEntry {
@@ -165,7 +165,7 @@ export class RunScrollbackStream {
           ? new CodeRenderable(surface.renderContext, {
               content: "",
               filetype: body.filetype,
-              syntaxStyle: entrySyntax(commit, this.theme),
+              syntaxStyle: entrySyntax(this.theme),
               width: "100%",
               wrapMode: "word",
               drawUnstyledText: false,
@@ -175,7 +175,7 @@ export class RunScrollbackStream {
             })
           : new MarkdownRenderable(surface.renderContext, {
               content: "",
-              syntaxStyle: entrySyntax(commit, this.theme),
+              syntaxStyle: entrySyntax(this.theme),
               width: "100%",
               streaming: true,
               internalBlockMode: "top-level",

packages/cli/src/mini/scrollback.writer.tsx

@@ -84,7 +84,7 @@ export function RunEntryContent(props: {
   const theme = createMemo(() => props.theme ?? RUN_THEME_FALLBACK)
   const body = createMemo(() => props.body ?? entryBody(props.commit))
   const style = createMemo(() => entryLook(props.commit, theme().entry))
-  const syntax = createMemo(() => entrySyntax(props.commit, theme()))
+  const syntax = createMemo(() => entrySyntax(theme()))
   const color = createMemo(() => entryColor(props.commit, theme()))
   const suppressBackgrounds = createMemo(() => props.opts?.suppressBackgrounds === true)
   const diffBg = (color: ColorInput) => (suppressBackgrounds() ? transparent : color)

packages/cli/src/mini/theme.ts

@@ -47,7 +47,6 @@ export type RunBlockTheme = {
   text: ColorInput
   muted: ColorInput
   syntax?: SyntaxStyle
-  subtleSyntax?: SyntaxStyle
   diffAdded: ColorInput
   diffRemoved: ColorInput
   diffAddedBg: ColorInput
@@ -172,42 +171,10 @@ function tint(base: RGBA, overlay: RGBA, value: number): RGBA {
   )
 }
 
-function blend(color: RGBA, bg: RGBA): RGBA {
-  if (color.a >= 1) {
-    return color
-  }
-
-  return RGBA.fromValues(
-    bg.r + (color.r - bg.r) * color.a,
-    bg.g + (color.g - bg.g) * color.a,
-    bg.b + (color.b - bg.b) * color.a,
-    1,
-  )
-}
-
 function chroma(color: RGBA) {
   return Math.max(color.r, color.g, color.b) - Math.min(color.r, color.g, color.b)
 }
 
-function opaqueSyntaxStyle(style: SyntaxStyle | undefined, bg: RGBA): SyntaxStyle | undefined {
-  if (!style) {
-    return undefined
-  }
-
-  return SyntaxStyle.fromStyles(
-    Object.fromEntries(
-      [...style.getAllStyles()].map(([name, value]) => [
-        name,
-        {
-          ...value,
-          fg: value.fg ? blend(value.fg, bg) : value.fg,
-          bg: value.bg ? blend(value.bg, bg) : value.bg,
-        },
-      ]),
-    ),
-  )
-}
-
 function indexedPalette(colors: TerminalColors, size: number = Math.max(colors.palette.length, 16)): RGBA[] {
   return Array.from({ length: size }, (_, index) => {
     const value = colors.palette[index]
@@ -502,10 +469,7 @@ function map(
   scrollbackTheme: TuiThemeCurrent,
   splash: RunSplashTheme,
   syntax?: SyntaxStyle,
-  subtleSyntax?: SyntaxStyle,
 ): RunTheme {
-  const opaqueSubtleSyntax = opaqueSyntaxStyle(subtleSyntax, scrollbackTheme.background)
-  subtleSyntax?.destroy()
   const footerBackground = alpha(footerTheme.background, 1)
   const footerMode = mode(footerBackground)
   const shade = fade(footerTheme.backgroundMenu, footerTheme.background, 0.12, 0.56, 0.72)
@@ -566,7 +530,6 @@ function map(
       text: scrollbackTheme.text,
       muted: scrollbackTheme.textMuted,
       syntax,
-      subtleSyntax: opaqueSubtleSyntax,
       diffAdded: scrollbackTheme.diffAdded,
       diffRemoved: scrollbackTheme.diffRemoved,
       diffAddedBg: transparent,
@@ -677,13 +640,7 @@ export async function resolveRunTheme(renderer: CliRenderer): Promise<RunTheme>
       _hasSelectedListItemText: true,
     }
     const syntax = shared.generateSyntax(syntaxTheme)
-    return map(
-      footerTheme,
-      scrollbackTheme,
-      splashTheme(scrollbackTheme, indexed),
-      syntax,
-      shared.generateSubtleSyntax(syntaxTheme),
-    )
+    return map(footerTheme, scrollbackTheme, splashTheme(scrollbackTheme, indexed), syntax)
   } catch {
     return RUN_THEME_FALLBACK
   }

packages/cli/src/tui-config.ts

@@ -1,24 +0,0 @@
-export * as TuiConfig from "./tui-config"
-
-import { Global } from "@opencode-ai/core/global"
-import { TuiConfig } from "@opencode-ai/tui/config/v1"
-import { Effect, FileSystem, Option, Schema } from "effect"
-import { parse, type ParseError } from "jsonc-parser"
-import path from "path"
-
-export const load = Effect.fn("TuiConfig.load")(function* () {
-  const fs = yield* FileSystem.FileSystem
-  const global = yield* Global.Service
-  const filepath = path.join(global.config, "tui.json")
-  const text = yield* fs.readFileString(filepath).pipe(Effect.catch(() => Effect.succeed(undefined)))
-  if (!text) return TuiConfig.resolve({}, { terminalSuspend: process.platform !== "win32" })
-
-  const errors: ParseError[] = []
-  const input: unknown = parse(text, errors, { allowTrailingComma: true })
-  if (errors.length) return TuiConfig.resolve({}, { terminalSuspend: process.platform !== "win32" })
-
-  return TuiConfig.resolve(
-    Option.getOrElse(Schema.decodeUnknownOption(TuiConfig.Info)(input), () => ({})),
-    { terminalSuspend: process.platform !== "win32" },
-  )
-})

packages/cli/test/config.test.ts

@@ -0,0 +1,144 @@
+import { NodeFileSystem } from "@effect/platform-node"
+import { Global } from "@opencode-ai/core/global"
+import { Effect } from "effect"
+import { expect, test } from "bun:test"
+import path from "path"
+import { Config } from "../src/config"
+
+function run<A, E>(directory: string, effect: Effect.Effect<A, E, Config.Service>) {
+  return Effect.runPromise(
+    effect.pipe(
+      Effect.provide(Config.layer),
+      Effect.provide(Global.layerWith({ config: directory, state: directory })),
+      Effect.provide(NodeFileSystem.layer),
+    ),
+  )
+}
+
+test("migrates tui and kv config into cli.json", async () => {
+  const directory = await Bun.$`mktemp -d`.text().then((value) => value.trim())
+  await Bun.write(
+    path.join(directory, "tui.json"),
+    JSON.stringify({
+      theme: "legacy",
+      keybinds: { leader: "ctrl+o" },
+      plugin: [["example", { mode: "safe" }]],
+      plugin_enabled: { disabled: false },
+      leader_timeout: 500,
+      scroll_speed: 2,
+      scroll_acceleration: { enabled: true },
+      diff_style: "stacked",
+      mouse: false,
+    }),
+  )
+  await Bun.write(
+    path.join(directory, "kv.json"),
+    JSON.stringify({
+      theme_mode_lock: "light",
+      attention_sound_pack: "custom.pack",
+      diff_wrap_mode: "none",
+      diff_viewer_show_file_tree: false,
+      diff_viewer_single_patch: true,
+      diff_viewer_view: "split",
+      terminal_title_enabled: false,
+      file_context_enabled: false,
+      paste_summary_enabled: false,
+      sidebar: "hide",
+      scrollbar_visible: true,
+      thinking_mode: "show",
+      exploration_grouping: false,
+      tips_hidden: true,
+      dismissed_getting_started: true,
+      animations_enabled: false,
+      skipped_version: "9.9.9",
+      which_key_layout: "overlay",
+    }),
+  )
+
+  try {
+    const config = await run(
+      directory,
+      Effect.gen(function* () {
+        const service = yield* Config.Service
+        return yield* service.get()
+      }),
+    )
+
+    expect(config).toMatchObject({
+      theme: { name: "legacy", mode: "light" },
+      keybinds: { leader: "ctrl+o" },
+      plugins: [{ package: "example", options: { mode: "safe" } }, "-disabled"],
+      leader: { timeout: 500 },
+      scroll: { speed: 2, acceleration: true },
+      attention: { sound_pack: "custom.pack" },
+      diffs: { wrap: "none", tree: false, single: true, view: "split" },
+      terminal: { title: false },
+      prompt: { editor: false, paste: "full" },
+      session: { sidebar: "hide", scrollbar: true, thinking: "show", grouping: "none" },
+      hints: { tips: false, onboarding: false },
+      animations: false,
+      mouse: false,
+    })
+    expect(config).not.toHaveProperty("skipped_version")
+    expect(config).not.toHaveProperty("which_key")
+    expect((await Bun.file(path.join(directory, "cli.json")).json()).keybinds).toEqual({ leader: "ctrl+o" })
+    expect(await Bun.file(path.join(directory, "cli.json")).exists()).toBe(true)
+    expect(await Bun.file(path.join(directory, "tui.json")).exists()).toBe(true)
+    expect(await Bun.file(path.join(directory, "kv.json")).exists()).toBe(true)
+  } finally {
+    await Bun.$`rm -rf ${directory}`
+  }
+})
+
+test("migrates before the first update and does not remigrate afterward", async () => {
+  const directory = await Bun.$`mktemp -d`.text().then((value) => value.trim())
+  await Bun.write(path.join(directory, "tui.json"), JSON.stringify({ theme: "legacy" }))
+
+  try {
+    const config = await run(
+      directory,
+      Effect.gen(function* () {
+        const service = yield* Config.Service
+        yield* service.update((draft) => {
+          draft.animations = false
+          draft.mouse = false
+        })
+        yield* Effect.promise(() =>
+          Bun.write(path.join(directory, "tui.json"), JSON.stringify({ theme: "changed" })),
+        )
+        return yield* service.get()
+      }),
+    )
+
+    expect(config).toEqual({ theme: { name: "legacy" }, animations: false, mouse: false })
+    expect(await Bun.file(path.join(directory, "cli.json")).json()).toEqual({
+      theme: { name: "legacy" },
+      animations: false,
+      mouse: false,
+    })
+  } finally {
+    await Bun.$`rm -rf ${directory}`
+  }
+})
+
+test("updates a config draft while preserving JSONC comments", async () => {
+  const directory = await Bun.$`mktemp -d`.text().then((value) => value.trim())
+  await Bun.write(path.join(directory, "cli.json"), "{\n  // Keep this comment\n  \"animations\": true\n}\n")
+
+  try {
+    const config = await run(
+      directory,
+      Effect.gen(function* () {
+        const service = yield* Config.Service
+        return yield* service.update((draft) => {
+          draft.prompt = { paste: "compact" }
+        })
+      }),
+    )
+
+    expect(config).toEqual({ animations: true, prompt: { paste: "compact" } })
+    expect(await Bun.file(path.join(directory, "cli.json")).text()).toContain("// Keep this comment")
+  } finally {
+    await Bun.$`rm -rf ${directory}`
+  }
+})

packages/cli/test/tui-config.test.ts

@@ -1,25 +0,0 @@
-import { NodeFileSystem } from "@effect/platform-node"
-import { Global } from "@opencode-ai/core/global"
-import { Effect } from "effect"
-import { expect, test } from "bun:test"
-import path from "path"
-import { TuiConfig } from "../src/tui-config"
-
-test("loads the global tui config", async () => {
-  const directory = await Bun.$`mktemp -d`.text().then((value) => value.trim())
-  await Bun.write(path.join(directory, "tui.json"), JSON.stringify({ keybinds: { leader: "ctrl+o" } }))
-
-  try {
-    const config = await Effect.runPromise(
-      TuiConfig.load().pipe(
-        Effect.provide(Global.layerWith({ config: directory })),
-        Effect.provide(NodeFileSystem.layer),
-      ),
-    )
-
-    expect(config.keybinds.get("leader")?.[0]?.key).toBe("ctrl+o")
-    expect(config.keybinds.get("session.new")?.[0]?.key).toBe("<leader>n")
-  } finally {
-    await Bun.$`rm -rf ${directory}`
-  }
-})

packages/client/src/effect/service.ts

@@ -59,11 +59,11 @@ const discoverLocal = Effect.fnUntraced(function* (options: Options) {
 export const start = Effect.fn("service.start")(function* (options: StartOptions = {}) {
   const compatible = yield* discover(options)
   if (compatible !== undefined) return compatible
-  const mismatched = yield* find(options)
-  yield* Effect.sync(() =>
-    options.onStart?.(mismatched === undefined ? "missing" : "version-mismatch", mismatched?.info),
-  )
-  if (mismatched !== undefined) yield* kill(mismatched.info, options).pipe(Effect.ignore)
+  const existing = yield* find(options)
+  if (existing?.version !== undefined && (options.version === undefined || existing.version === options.version))
+    return existing.endpoint
+  yield* Effect.sync(() => options.onStart?.(existing === undefined ? "missing" : "version-mismatch", existing?.info))
+  if (existing !== undefined) yield* kill(existing.info, options).pipe(Effect.ignore)
 
   const [command, ...args] = options.command ?? ["opencode", "serve", "--service"]
   if (command === undefined) return yield* Effect.fail(new Error("Missing service command"))
@@ -138,6 +138,7 @@ const read = Effect.fnUntraced(function* (file?: string) {
 type LocalService = {
   readonly info: Info
   readonly endpoint: Endpoint
+  readonly version?: string
 }
 
 const probe = Effect.fnUntraced(function* (info: Info, version?: string, allowLegacy = false) {
@@ -161,7 +162,7 @@ const probe = Effect.fnUntraced(function* (info: Info, version?: string, allowLe
     if (health.value.pid !== info.pid) return undefined
     if (info.version !== undefined && health.value.version !== info.version) return undefined
     if (version !== undefined && health.value.version !== version) return undefined
-    return { info, endpoint } satisfies LocalService
+    return { info, endpoint, version: health.value.version } satisfies LocalService
   }
   if (
     !allowLegacy ||

packages/client/test/fixture/service.ts

@@ -0,0 +1,39 @@
+import { rename, writeFile } from "node:fs/promises"
+
+const [registration, mode] = process.argv.slice(2)
+if (registration === undefined || mode === undefined) throw new Error("Missing service fixture arguments")
+
+let requests = 0
+const server = Bun.serve({
+  port: 0,
+  async fetch(request) {
+    if (new URL(request.url).pathname !== "/api/health") return new Response(null, { status: 404 })
+    requests += 1
+    if (mode === "modern" && requests === 1) {
+      await writeFile(registration + ".first-request", "")
+      while (!(await Bun.file(registration + ".release").exists())) await Bun.sleep(5)
+      return new Response(null, { status: 503 })
+    }
+    if (mode === "legacy") return Response.json({ healthy: true })
+    return Response.json({ healthy: true, version: "test", pid: process.pid })
+  },
+})
+
+await writeFile(
+  registration + ".tmp",
+  JSON.stringify({
+    id: crypto.randomUUID(),
+    version: mode === "legacy" ? undefined : "test",
+    url: server.url.toString(),
+    pid: process.pid,
+  }),
+  { mode: 0o600 },
+)
+await rename(registration + ".tmp", registration)
+
+const shutdown = () => {
+  server.stop(true)
+  process.exit()
+}
+process.on("SIGTERM", shutdown)
+process.on("SIGINT", shutdown)

packages/client/test/service.test.ts

@@ -0,0 +1,91 @@
+import { NodeFileSystem } from "@effect/platform-node"
+import { afterEach, expect, test } from "bun:test"
+import { Effect } from "effect"
+import { mkdtemp, rm, writeFile } from "node:fs/promises"
+import { tmpdir } from "node:os"
+import { join } from "node:path"
+import { Service } from "../src/effect/index"
+
+const fixture = join(import.meta.dir, "fixture/service.ts")
+const processes: Bun.Subprocess[] = []
+const directories: string[] = []
+
+afterEach(async () => {
+  processes.forEach((process) => process.kill("SIGTERM"))
+  await Promise.all(processes.splice(0).map((process) => process.exited))
+  await Promise.all(directories.splice(0).map((directory) => rm(directory, { recursive: true, force: true })))
+})
+
+test("a concurrent same-version start cannot invalidate a resolved endpoint", async () => {
+  const directory = await temp()
+  const registration = join(directory, "service.json")
+  spawn(registration, "modern")
+  await waitForFile(registration)
+  const original = await Bun.file(registration).json()
+
+  const starts: Service.StartReason[] = []
+  const first = run(
+    Service.start({
+      file: registration,
+      version: "test",
+      command: [],
+      onStart: (reason) => starts.push(reason),
+    }),
+  )
+  await waitForFile(registration + ".first-request")
+
+  const resolved = await run(Service.start({ file: registration, version: "test" }))
+  expect(resolved.url).toBe(original.url)
+
+  await writeFile(registration + ".release", "")
+  await first
+
+  expect(starts).toEqual([])
+  expect(await Bun.file(registration).json()).toEqual(original)
+  expect(await health(resolved.url)).toEqual({ healthy: true, version: "test", pid: original.pid })
+})
+
+test("a legacy health response is still replaced", async () => {
+  const directory = await temp()
+  const registration = join(directory, "service.json")
+  const existing = spawn(registration, "legacy")
+  await waitForFile(registration)
+
+  const starts: Service.StartReason[] = []
+  const result = run(Service.start({ file: registration, command: [], onStart: (reason) => starts.push(reason) }))
+
+  await expect(result).rejects.toThrow("Missing service command")
+  expect(starts).toEqual(["version-mismatch"])
+  await existing.exited
+})
+
+function run<A, E>(effect: Effect.Effect<A, E, never>) {
+  return Effect.runPromise(effect.pipe(Effect.provide(NodeFileSystem.layer)))
+}
+
+function spawn(registration: string, mode: string, ...args: string[]) {
+  const subprocess = Bun.spawn([process.execPath, fixture, registration, mode, ...args], {
+    stdout: "ignore",
+    stderr: "inherit",
+  })
+  processes.push(subprocess)
+  return subprocess
+}
+
+async function temp() {
+  const directory = await mkdtemp(join(tmpdir(), "opencode-client-service-"))
+  directories.push(directory)
+  return directory
+}
+
+async function waitForFile(file: string) {
+  for (let attempt = 0; attempt < 600; attempt++) {
+    if (await Bun.file(file).exists()) return
+    await Bun.sleep(5)
+  }
+  throw new Error(`Timed out waiting for ${file}`)
+}
+
+async function health(url: string) {
+  return fetch(new URL("/api/health", url), { signal: AbortSignal.timeout(1_000) }).then((response) => response.json())
+}

packages/codemode/AGENTS.md

@@ -4,7 +4,7 @@
 - Do not add a speculative generic permission or approval policy. A host omits tools it does not expose and enforces domain authorization inside each provided tool.
 - Keep Code Mode unaware of host session, channel, and conversation models. The hosting application supplies trusted execution scope around it.
 - Tool schemas are the model-facing Interface. Keep arguments minimal and natural to the operation; never add unrelated IDs as ambient capability tokens.
-- When interpreter behavior or support changes, update `interpreter-support.md` and direct tests in the same PR. Update `codemode.md` when the package design, integration status, or rationale changes.
+- When interpreter behavior or support changes, update `interpreter-support.md` and direct tests in the same PR.
 
 ## OpenAPI
 

packages/codemode/README.md

@@ -177,8 +177,8 @@ No limit has a default, on purpose: execution budgets are host policy. A host wi
 interruption should set `maxOutputBytes` and `timeoutMs`. Limits are safe integers; invalid configuration throws a
 `RangeError` at construction. Exceeding `maxOutputBytes` never fails the execution - oversized output is truncated
 with an in-band marker. The timeout interrupts in-flight tool fibers and pure busy loops alike; a value the program
-already returned survives a cleanup timeout as a success with a `TimeoutExceeded` warning. Two internals are fixed
-constants, not knobs: at most 8 concurrent tool calls, and 32 levels of data nesting at boundaries.
+already returned survives a cleanup timeout as a success with a `TimeoutExceeded` warning. CodeMode does not limit
+tool-call concurrency. Data nesting at boundaries is limited to 32 levels.
 
 ## Boundaries and Non-Goals
 

packages/codemode/codemode.md

@@ -1,164 +0,0 @@
-# CodeMode Design and Status
-
-This is the living design and status document for `@opencode-ai/codemode` and its existing V2 OpenCode adapter.
-It records current behavior, intentional boundaries, durable rationale, and material remaining work.
-
-Completed implementation history, branch names, test counts, and closed findings belong in git, not here. Remove
-completed work instead of preserving checked-off chronology.
-
-Detailed package API documentation lives in [README.md](./README.md), and the checkable language/runtime matrix lives
-in [interpreter-support.md](./interpreter-support.md). OpenAPI-specific follow-ups live in
-[src/openapi/TODO.md](./src/openapi/TODO.md).
-
-## How CodeMode Works
-
-### Purpose
-
-CodeMode gives a model one `execute` tool backed by a confined JavaScript interpreter. Inside the program, the model
-can call an explicit tree of schema-described tools, sequence dependent work, run independent calls concurrently,
-and filter or aggregate results before returning them to the agent loop.
-
-The goals are:
-
-- Reduce model context consumed by large tool catalogs.
-- Avoid an agent round-trip between every dependent tool call.
-- Keep large intermediate results inside the program instead of sending them through model context.
-- Give generated code only the authority explicitly supplied by the host.
-
-CodeMode is an orchestration language, not a general JavaScript runtime or an application authorization system.
-
-### Runtime
-
-The generic runtime lives in `packages/codemode` and is host-neutral:
-
-1. The host builds a tree of `Tool.make(...)` definitions and calls `CodeMode.make(...)` or `CodeMode.execute(...)`.
-2. CodeMode generates model instructions, a budgeted inline catalog, and the global `search(...)` built-in.
-3. TypeScript syntax is transpiled away, Acorn parses the resulting JavaScript, and an owned tree-walking interpreter
-   executes it without `eval`.
-4. Tool inputs and outputs cross schema and plain-data boundaries before they become visible on either side.
-5. Execution returns `CodeMode.Result`. Expected program and tool failures are diagnostic data; host interruption
-   remains Effect interruption.
-
-Effect Schemas validate and transform tool inputs and outputs. JSON Schemas render model-facing signatures but do not
-validate values; adapter-provided values still cross the plain-data boundary. A tool without an output schema is
-advertised as `Promise<unknown>`.
-
-### Discovery and model workflow
-
-The model sees a token-budgeted catalog. Every namespace remains visible, and complete signatures are selected
-round-robin across namespaces so one large namespace cannot starve the others. The global `search(...)` built-in is
-always callable - synchronously, counted as an admitted tool call - and is advertised when the inline catalog is
-partial.
-
-The intended workflow is:
-
-1. Pick an exact signature from the inline catalog, or return `search(...)` results and use a selected path in the
-   next execution.
-2. Call the exact returned path without guessing or normalizing segments.
-3. Narrow `Promise<unknown>` results before reading fields.
-4. Start independent calls together and await them with `Promise.all`.
-5. Filter and aggregate inside the program, then return only the data needed by the model.
-
-Search returns directly usable JavaScript paths, descriptions, and complete TypeScript signatures. It supports exact
-path lookup, namespace browsing, deterministic ranking, and pagination.
-
-### Tool execution
-
-Every sandbox promise starts eagerly on a run-once fiber owned by the whole CodeMode execution, including tool calls,
-async functions, chained `.then`/`.catch`/`.finally` reactions, `new Promise(executor)` constructions, and the
-`Promise.all`/`allSettled`/`race`/`any`/`resolve`/`reject` statics. Nested functions therefore cannot end the lifetime
-of work they started.
-Independent aggregate batches overlap, and rejection is observed at the eventual `await` or chained rejection handler.
-`Promise.race` and `Promise.any` use native non-cancelling settlement semantics: the deciding member wins while losers
-continue running, and an all-rejected `Promise.any` rejects with an `AggregateError`. `new Promise(...)` hands the
-executor first-class resolve/reject callables that may escape and settle the promise later, exactly once.
-Reaction ordering matches what V8 makes observable - handlers and await continuations are deferred and run in attach
-order, and a combinator settles one reaction turn after its deciding member - without promising exact microtask-count
-parity beyond that. At normal completion CodeMode interrupts everything still running - race losers,
-fail-fast `Promise.all` stragglers, and fire-and-forget calls alike: the program has returned, so no future await can
-exist, and work whose completion matters must be awaited by the program. Waiting for any class of leftover instead
-would let it hold the execution open, or deadlock it when queued work needs tool-call permits the leftovers occupy.
-Rejections that settled un-awaited before the return become `Success.warnings` diagnostics. A fatal program failure or
-host interruption closes the execution promise scope and interrupts its active fibers instead. A timeout does the
-same, except that a value the program already returned is preserved alongside a `TimeoutExceeded` warning rather than
-discarded. At most eight tool calls execute concurrently.
-
-The public execution-policy knobs are `timeoutMs`, `maxToolCalls`, and `maxOutputBytes`. The package supplies no
-defaults because budgets are host policy. The interpreter also enforces fixed internal boundaries for tool-call
-concurrency and data nesting depth. `maxOutputBytes` bounds retained payload bytes, not the complete rendered message;
-warning diagnostics have an equal separate budget so a large value cannot starve them, and fixed truncation notices and
-host-added framing are intentionally outside the budgets.
-
-### Data, files, and failures
-
-Program results and tool arguments are JSON-like data. Dates become ISO strings at host boundaries; RegExp, Map, and
-Set values become `{}` as they do under JSON serialization. Promise and runtime reference values cannot cross the
-boundary.
-
-Unknown host failures and invalid outputs are sanitized. `ToolError` is the explicit channel for a safe message that a
-tool wants the model to see. Diagnostic categories distinguish parsing, unsupported syntax, unknown tools, invalid
-data, tool failures, limits, timeouts, execution failures, and warning truncation.
-
-Files and other attachment content stay outside the interpreter. A host may collect them while child tools execute and
-attach them to the outer result, but the program receives only the structured tool output.
-
-### V2 OpenCode adapter
-
-CodeMode is integrated into V2 through `packages/core/src/tool/registry.ts` and
-`packages/core/src/tool/execute.ts`:
-
-- Core has one canonical `Tool` representation. Location-scoped producers register direct or deferred tools through
-  `Tools.Service`.
-- Each model step snapshots effective registrations, applies catalog visibility filtering, and exposes direct tools
-  normally.
-- When visible deferred tools exist, Core reserves and materializes one `execute` tool. Grouped deferred tools become
-  CodeMode namespaces instead of flattened model-facing names.
-- Nested calls execute the registered `Tool` values captured for the model request; later registrations affect later
-  requests.
-- Authorization and side-effect ordering remain responsibilities of the leaf tool. Catalog visibility is not execution
-  authorization.
-- Structured child output enters the interpreter. File parts are collected host-side and attached to the outer result.
-- Nested call statuses are returned as final `execute` metadata for the TUI.
-- `execute` is the one model-facing tool invocation. Nested calls reuse its invocation context and do not independently
-  run registry hooks or model-output bounding; this keeps complete intermediate structured values available for
-  in-program filtering. The outer `execute` settlement is the single model-output bounding boundary.
-- Core supplies no CodeMode timeout or tool-call limit. User cancellation interrupts the outer invocation and its
-  supervised children; the outer settlement applies Core's normal output-retention policy.
-
-MCP tools use this canonical path: they register as grouped tools and are deferred while CodeMode is enabled. Existing
-output schemas are preserved in generated signatures. Direct Core tools remain direct and are not ambient globals
-inside CodeMode.
-
-## Intentionally Unsupported
-
-These are product boundaries rather than DSL backlog:
-
-- Ambient filesystem, process, environment, network, credential, or application access. External work must go through
-  supplied tools.
-- Modules, imports, dynamic imports, `eval`, arbitrary host globals, npm packages, and prototype mutation.
-- Generic permission prompts, authorization policy, durable pause/resume, replay, storage, or exactly-once external
-  side effects. Hosts and tools own those concerns.
-- Heuristic parsing of text tool results as JSON. A result should not silently change type based on its contents.
-
-The OpenAPI adapter may gain more transports and encodings, but it must continue skipping operations it cannot
-represent accurately rather than guessing semantics.
-
-## Decisions and Rationale
-
-| Decision                                                     | Rationale                                                                                                                                                                                                                |
-| ------------------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
-| Keep an owned tree-walking interpreter.                      | The product need is bounded tool orchestration, not arbitrary JavaScript. Owning the language surface keeps authority and behavior explicit.                                                                             |
-| Treat schemas as the model-facing interface.                 | Signatures drive correct calls; Effect Schema also provides the runtime validation boundary, while JSON Schema supports adapter interoperability.                                                                        |
-| Keep authority host-owned.                                   | CodeMode can only confine programs to supplied tools. The host chooses those tools, and each tool enforces its own authorization and side-effect policy.                                                                 |
-| Use progressive catalog disclosure plus search.              | Large tool sets should not consume the prompt, but every namespace must remain discoverable and speculative search calls should remain valid.                                                                            |
-| Start promises eagerly and supervise them for the execution. | This preserves normal call-time parallelism and run-once settlement while allowing pending work to be interrupted when the program returns.                                                                              |
-| Keep files outside the sandbox value space.                  | Models should compose structured data without routing binary payloads through generated code or context.                                                                                                                 |
-| Treat `execute` as the model-facing invocation boundary.     | Nested calls are implementation details of one orchestration program. Reusing the outer context and bounding only the final result preserves complete intermediate data without inventing durable child-call identities. |
-| Return expected failures as data.                            | Models need actionable diagnostics without exposing private host causes; host interruption and defects must still propagate correctly.                                                                                   |
-| Leave execution-limit defaults to hosts.                     | Appropriate budgets depend on the surrounding product and its own cancellation, retention, and output-bounding policies.                                                                                                 |
-| Skip unsupported OpenAPI operations.                         | Incorrect parameter encoding, authentication, or transport behavior is worse than a precise `skipped` reason.                                                                                                            |
-
-## Remaining Work
-
-The [interpreter support checklist](./interpreter-support.md) owns concrete DSL, standard-library, semantic-correctness,
-diagnostic, and data-boundary work. OpenAPI adapter work remains in [src/openapi/TODO.md](./src/openapi/TODO.md).

packages/codemode/interpreter-support.md

@@ -24,7 +24,7 @@ ultimate source of truth.
 - [x] Tool calls through the host-provided `tools` tree only.
 - [x] The global `search(...)` built-in: synchronous tool discovery that counts as an admitted tool call and is
       shadowable by program declarations like other globals.
-- [x] Cooperative timeout, tool-call accounting, output bounding, and a maximum of eight concurrent tool calls.
+- [x] Cooperative timeout, an optional total tool-call limit, output bounding, and unrestricted tool-call concurrency.
 - [ ] Full JavaScript or TypeScript compatibility. CodeMode is a bounded orchestration language.
 
 ## Values and literals
@@ -171,14 +171,13 @@ ultimate source of truth.
 - [ ] `Array.prototype.toSpliced`.
 - [ ] Canonical index handling: a key such as `"01"` must not alias index `1`.
 - [ ] Complete sparse-array parity. Promise combinators do consume holes as `undefined` members, as in JS.
-- [ ] Correct `findLast` return behavior when its predicate mutates the examined element.
 
 ## Strings
 
 - [x] Case/normalization: `toLowerCase`, `toUpperCase`, `normalize`.
-- [x] Trimming: `trim`, `trimStart`, `trimEnd`, `trimLeft`, and `trimRight`.
+- [x] Trimming: `trim`, `trimStart`, and `trimEnd`.
 - [x] Searching/tests: `includes`, `startsWith`, `endsWith`, `indexOf`, `lastIndexOf`, and `search`.
-- [x] Slicing/access: `slice`, `substring`, `substr`, `at`, `charAt`, `charCodeAt`, and `codePointAt`.
+- [x] Slicing/access: `slice`, `substring`, `at`, `charAt`, `charCodeAt`, and `codePointAt`.
 - [x] Construction/transformation: `split`, `concat`, `repeat`, `padStart`, `padEnd`, `replace`, and `replaceAll`.
 - [x] Regular-expression integration: `match`, materialized `matchAll`, `replace`, `replaceAll`, `split`, and `search`.
 - [x] `localeCompare`; locale and options arguments are currently ignored.
@@ -291,7 +290,6 @@ ultimate source of truth.
 These are actionable implementation items. Check them off only when behavior and direct tests land.
 
 - [x] Return real promises from `Promise.all`, `Promise.allSettled`, and `Promise.race`.
-- [ ] Bound pending tool-call admission/allocation in addition to execution concurrency.
 - [ ] Guarantee every advertised tool path is executable, including dotted and blocked path segments.
 - [ ] Define safe outbound handling for non-finite numbers and `undefined` so invalid values cannot silently become
       `null` in render-only or OpenAPI tool calls.

packages/codemode/src/codemode.ts

@@ -1,5 +1,5 @@
 import { Effect, Schema } from "effect"
-import { executeWithLimits } from "./interpreter/runtime.js"
+import { executeWithLimits } from "./interpreter/execute.js"
 import { type HostTools, type Services, type ToolDescription, ToolRuntime } from "./tool-runtime.js"
 import type { Definition } from "./tool.js"
 
@@ -9,15 +9,15 @@ export type { ToolCall, ToolCallEnded, ToolCallHooks, ToolCallStarted, ToolDescr
 /** Resource budgets enforced independently during each CodeMode program execution. */
 export type ExecutionLimits = {
   /**
-   * Wall-clock milliseconds before execution is interrupted; result delivery additionally
-   * waits for tool interruption cleanup. No default: absent means no timeout.
+   * Wall-clock milliseconds before interruption. Result delivery waits for tool cleanup.
+   * No default: absent means no timeout.
    */
   readonly timeoutMs?: number
   /** Maximum number of tool calls admitted by the runtime. No default: absent means unlimited. */
   readonly maxToolCalls?: number
   /**
-   * Maximum UTF-8 bytes retained from the result value and logs; warnings have a separate
-   * budget of the same size. Fixed truncation notices and host formatting are additional.
+   * Maximum UTF-8 bytes retained from the result and logs. Warnings have a separate equal budget;
+   * truncation notices and host formatting are additional.
    */
   readonly maxOutputBytes?: number
 }
@@ -94,7 +94,6 @@ const ToolCallSchema = Schema.Struct({ name: Schema.String })
 export const Success = Schema.Struct({
   ok: Schema.Literal(true),
   value: Schema.Json,
-  // Runtime-authored non-fatal diagnostics; program console output stays in `logs`.
   warnings: Schema.optionalKey(Schema.Array(Diagnostic)),
   logs: Schema.optionalKey(Schema.Array(Schema.String)),
   truncated: Schema.optionalKey(Schema.Boolean),

packages/codemode/src/interpreter/errors.ts

@@ -0,0 +1,93 @@
+import type { Diagnostic } from "../codemode.js"
+import { ToolError } from "../tool-error.js"
+import { copyOut, ToolRuntimeError, type SafeObject } from "../tool-runtime.js"
+import { type AstNode, formatLocation, InterpreterRuntimeError, ProgramThrow, sourceLocation } from "./model.js"
+import { containsRuntimeReference } from "./references.js"
+import { spreadItems } from "../stdlib/collections.js"
+import { coerceToString, createAggregateErrorValue, createErrorValue, errorConstructors } from "../stdlib/value.js"
+
+export const normalizeError = (error: unknown): Diagnostic => {
+  if (error instanceof InterpreterRuntimeError) {
+    return {
+      kind: error.kind,
+      message: `${error.message}${formatLocation(error.node)}`,
+      ...(error.node?.loc ? { location: sourceLocation(error.node) } : {}),
+      ...(error.suggestions ? { suggestions: error.suggestions } : {}),
+    }
+  }
+
+  if (error instanceof ToolRuntimeError) {
+    return {
+      kind: error.kind,
+      message: error.message,
+      ...(error.suggestions.length > 0 ? { suggestions: error.suggestions } : {}),
+    }
+  }
+
+  if (error instanceof ToolError) {
+    return { kind: "ToolFailure", message: error.message }
+  }
+
+  if (error instanceof ProgramThrow) {
+    const value = error.value
+    let message: string
+    if (containsRuntimeReference(value)) {
+      // Never expose runtime reference internals through thrown values.
+      message = "a non-data value"
+    } else if (typeof value === "string") {
+      message = value
+    } else if (
+      value !== null &&
+      typeof value === "object" &&
+      typeof (value as { message?: unknown }).message === "string"
+    ) {
+      message = (value as { message: string }).message
+    } else {
+      try {
+        message = JSON.stringify(copyOut(value)) ?? String(value)
+      } catch {
+        message = String(value)
+      }
+    }
+    return { kind: "ExecutionFailure", message: `Uncaught: ${message}` }
+  }
+
+  if (error instanceof RangeError && /call stack|recursion/i.test(error.message)) {
+    return {
+      kind: "ExecutionFailure",
+      message: "Execution exceeded the maximum nesting depth.",
+    }
+  }
+
+  if (error instanceof Error) {
+    return {
+      kind: error.name === "SyntaxError" ? "ParseError" : "ExecutionFailure",
+      message: error.message,
+    }
+  }
+
+  return {
+    kind: "ExecutionFailure",
+    message: String(error),
+  }
+}
+
+export const caughtErrorValue = (thrown: unknown): unknown => {
+  if (thrown instanceof ProgramThrow) return thrown.value
+  if (thrown instanceof InterpreterRuntimeError) return createErrorValue(thrown.errorName, thrown.message)
+  const name = thrown instanceof Error && errorConstructors.has(thrown.name) ? thrown.name : "Error"
+  return createErrorValue(name, normalizeError(thrown).message)
+}
+
+export const constructErrorValue = (name: string, args: Array<unknown>, node: AstNode): SafeObject => {
+  if (name !== "AggregateError") return createErrorValue(name, args[0] === undefined ? "" : coerceToString(args[0]))
+  const errors = spreadItems(args[0])
+  if (errors === undefined) {
+    throw new InterpreterRuntimeError(
+      "new AggregateError(...) expects an array of errors (e.g. new AggregateError(errors, message?)).",
+      node,
+    ).as("TypeError")
+  }
+  // Error values must not alias caller-owned arrays.
+  return createAggregateErrorValue([...errors], args[1] === undefined ? "" : coerceToString(args[1]))
+}

packages/codemode/src/interpreter/execute.ts

@@ -0,0 +1,224 @@
+import { parse } from "acorn"
+import { Cause, Effect, Scope } from "effect"
+import { DiagnosticCategory, ModuleKind, ScriptTarget, flattenDiagnosticMessageText, transpileModule } from "typescript"
+import type { DataValue, Diagnostic, ExecuteOptions, ResolvedExecutionLimits, Result } from "../codemode.js"
+import { copyIn, copyOut, ToolRuntime, type HostTools, type Services } from "../tool-runtime.js"
+import { normalizeError } from "./errors.js"
+import { InterpreterRuntimeError, isRecord, type ProgramNode } from "./model.js"
+import { PromiseRuntime } from "./promises.js"
+import { Interpreter } from "./runtime.js"
+
+export const executeWithLimits = <const Tools extends Record<string, unknown>>(
+  options: ExecuteOptions<Tools>,
+  limits: ResolvedExecutionLimits,
+  searchIndex: ToolRuntime.DiscoveryPlan["searchIndex"],
+): Effect.Effect<Result, never, Services<Tools>> => {
+  if (options.code.trim().length === 0) {
+    return Effect.succeed({
+      ok: false,
+      error: { kind: "ParseError", message: "Code cannot be empty." },
+      toolCalls: [],
+    })
+  }
+
+  // Allocate execution state inside suspension so reused Effects never share it.
+  return Effect.suspend(() => {
+    const tools = ToolRuntime.make(
+      (options.tools ?? {}) as HostTools<Services<Tools>>,
+      limits.maxToolCalls,
+      searchIndex,
+      {
+        onToolCallStart: options.onToolCallStart,
+        onToolCallEnd: options.onToolCallEnd,
+      },
+    )
+    const logs: Array<string> = []
+    const logged = () => (logs.length > 0 ? { logs: [...logs] } : {})
+    // Set only after copy-out so timeouts cannot report invalid values as completed.
+    let returned: { value: DataValue; promises: PromiseRuntime<Services<Tools>> } | undefined
+
+    const base = Effect.acquireUseRelease(
+      Scope.make("parallel"),
+      (scope) =>
+        Effect.gen(function* () {
+          const program = parseProgram(options.code)
+          const promises = new PromiseRuntime<Services<Tools>>(scope)
+          const interpreter = new Interpreter<Services<Tools>>(tools.invoke, tools.search, tools.keys, promises, logs)
+          const value = yield* interpreter.run(program)
+          const result = copyOut(copyIn(value, "Execution result"), true) as DataValue
+          returned = { value: result, promises }
+          const warnings = yield* promises.interrupt()
+          return {
+            ok: true,
+            value: result,
+            ...(warnings.length > 0 ? { warnings } : {}),
+            ...logged(),
+            toolCalls: tools.calls,
+          } satisfies Result
+        }),
+      (scope, exit) => Scope.close(scope, exit),
+    )
+    const timeoutMs = limits.timeoutMs
+    const operation =
+      timeoutMs === undefined
+        ? base
+        : base.pipe(
+            Effect.timeoutOrElse({
+              duration: timeoutMs,
+              orElse: () =>
+                Effect.sync(() => {
+                  if (returned === undefined) {
+                    return {
+                      ok: false,
+                      error: { kind: "TimeoutExceeded", message: `Execution timed out after ${timeoutMs}ms.` },
+                      ...logged(),
+                      toolCalls: tools.calls,
+                    } satisfies Result
+                  }
+                  // Keep the timeout warning first so truncation preserves it.
+                  return {
+                    ok: true,
+                    value: returned.value,
+                    warnings: [
+                      {
+                        kind: "TimeoutExceeded",
+                        message: `The program returned, but background work was still running at the ${timeoutMs}ms timeout and was interrupted. Await all started promises.`,
+                      },
+                      ...returned.promises.diagnostics(),
+                    ],
+                    ...logged(),
+                    toolCalls: tools.calls,
+                  } satisfies Result
+                }),
+            }),
+          )
+
+    return operation.pipe(
+      Effect.catchCause((cause) =>
+        Cause.hasInterruptsOnly(cause)
+          ? Effect.interrupt
+          : Effect.succeed({
+              ok: false,
+              error: normalizeError(Cause.squash(cause)),
+              ...logged(),
+              toolCalls: tools.calls,
+            } satisfies Result),
+      ),
+      Effect.map((result) =>
+        limits.maxOutputBytes === undefined ? result : boundOutput(result, limits.maxOutputBytes),
+      ),
+    )
+  })
+}
+
+const parseProgram = (code: string): ProgramNode => {
+  const transpiled = transpileModule(`async function __codemode__() {\n${code}\n}`, {
+    reportDiagnostics: true,
+    compilerOptions: {
+      target: ScriptTarget.ESNext,
+      module: ModuleKind.ESNext,
+    },
+  })
+  const diagnostic = transpiled.diagnostics?.find((item) => item.category === DiagnosticCategory.Error)
+
+  if (diagnostic) {
+    throw new InterpreterRuntimeError(
+      `Failed to parse TypeScript: ${flattenDiagnosticMessageText(diagnostic.messageText, "\n")}`,
+      undefined,
+      "ParseError",
+    )
+  }
+
+  const bodyStart = transpiled.outputText.indexOf("{") + 1
+  const bodyEnd = transpiled.outputText.lastIndexOf("}")
+  const executableCode = transpiled.outputText.slice(bodyStart, bodyEnd)
+  const parsed = parse(executableCode, {
+    ecmaVersion: "latest",
+    sourceType: "script",
+    allowReturnOutsideFunction: true,
+    allowAwaitOutsideFunction: true,
+    locations: true,
+  }) as unknown
+
+  if (!isRecord(parsed) || parsed.type !== "Program" || !Array.isArray(parsed.body)) {
+    throw new InterpreterRuntimeError("Failed to parse script as a Program node.")
+  }
+
+  return parsed as ProgramNode
+}
+
+const utf8ByteLength = (value: string): number => new TextEncoder().encode(value).byteLength
+
+// Drop a replacement character produced by truncating inside a UTF-8 sequence.
+const utf8Truncate = (value: string, maxBytes: number): string => {
+  const bytes = new TextEncoder().encode(value)
+  if (bytes.byteLength <= maxBytes) return value
+  const text = new TextDecoder("utf-8").decode(bytes.slice(0, Math.max(0, maxBytes)))
+  return text.endsWith("\uFFFD") ? text.slice(0, -1) : text
+}
+
+// Warnings have a separate budget so result data cannot starve diagnostics.
+const boundOutput = (result: Result, maxOutputBytes: number): Result => {
+  let truncated = false
+
+  let value: DataValue = null
+  let valueBytes = 0
+  if (result.ok) {
+    const serialized = JSON.stringify(result.value) ?? "null"
+    const bytes = utf8ByteLength(serialized)
+    if (bytes > maxOutputBytes) {
+      truncated = true
+      value = `${utf8Truncate(serialized, maxOutputBytes)} [result truncated: ${bytes} bytes exceeds the ${maxOutputBytes}-byte output limit; return a smaller value]`
+      valueBytes = maxOutputBytes
+    } else {
+      value = result.value
+      valueBytes = bytes
+    }
+  }
+
+  const warnings = result.ok ? (result.warnings ?? []) : []
+  const keptWarnings: Array<Diagnostic> = []
+  let warningBytes = 0
+  for (const warning of warnings) {
+    const bytes = utf8ByteLength(JSON.stringify(warning)) + 1
+    if (warningBytes + bytes > maxOutputBytes) break
+    warningBytes += bytes
+    keptWarnings.push(warning)
+  }
+  if (keptWarnings.length < warnings.length) {
+    truncated = true
+    keptWarnings.push({
+      kind: "Truncated",
+      message: `${warnings.length - keptWarnings.length} additional warnings omitted by the output limit.`,
+    })
+  }
+
+  const logs = result.logs ?? []
+  const kept: Array<string> = []
+  const logBudget = Math.max(0, maxOutputBytes - valueBytes)
+  let logBytes = 0
+  for (const line of logs) {
+    const lineBytes = utf8ByteLength(line) + 1
+    if (logBytes + lineBytes > logBudget) break
+    logBytes += lineBytes
+    kept.push(line)
+  }
+  if (kept.length < logs.length) {
+    truncated = true
+    kept.push(`[logs truncated: showing ${kept.length} of ${logs.length} lines]`)
+  }
+
+  if (!truncated) return result
+  const warningsPart = keptWarnings.length > 0 ? { warnings: keptWarnings } : {}
+  const logsPart = kept.length > 0 ? { logs: kept } : {}
+  return result.ok
+    ? {
+        ok: true,
+        value,
+        ...warningsPart,
+        ...logsPart,
+        truncated: true,
+        toolCalls: result.toolCalls,
+      }
+    : { ok: false, error: result.error, ...logsPart, truncated: true, toolCalls: result.toolCalls }
+}

packages/codemode/src/interpreter/methods.ts

@@ -0,0 +1,820 @@
+import { Effect } from "effect"
+import {
+  type AstNode,
+  CodeModeFunction,
+  CoercionFunction,
+  GlobalMethodReference,
+  IntrinsicReference,
+  InterpreterRuntimeError,
+  PromiseCapabilityFunction,
+  supportedSyntaxMessage,
+  UriFunction,
+} from "./model.js"
+import { rejectCircularInsertion } from "./references.js"
+import { isBlockedMember, type SafeObject } from "../tool-runtime.js"
+import {
+  SandboxDate,
+  SandboxMap,
+  SandboxPromise,
+  SandboxRegExp,
+  SandboxSet,
+  SandboxURL,
+  SandboxURLSearchParams,
+} from "../values.js"
+import { invokeDateMethod, invokeDateStatic } from "../stdlib/date.js"
+import { invokeJsonMethod } from "../stdlib/json.js"
+import { invokeMathMethod } from "../stdlib/math.js"
+import { invokeNumberMethod, invokeNumberStatic } from "../stdlib/number.js"
+import { invokeObjectMethod } from "../stdlib/object.js"
+import { invokeRegExpMethod, matchToValue, toHostRegex } from "../stdlib/regexp.js"
+import { invokeStringStatic } from "../stdlib/string.js"
+import { invokeUriFunction, invokeURLMethod, invokeURLStatic, uriArgument } from "../stdlib/url.js"
+import { boundedData, coerceToNumber, coerceToString, invokeCoercion } from "../stdlib/value.js"
+
+export type CallbackRunner<R> = {
+  readonly invokeFunction: (fn: CodeModeFunction, args: Array<unknown>) => Effect.Effect<unknown, unknown, R>
+  readonly settlePromise: (promise: SandboxPromise) => Effect.Effect<unknown, unknown, never>
+}
+
+export const invokeIntrinsic = <R>(
+  runner: CallbackRunner<R>,
+  ref: IntrinsicReference,
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<unknown, unknown, R> => {
+  if (typeof ref.receiver === "string") {
+    if (
+      (ref.name === "replace" || ref.name === "replaceAll") &&
+      (args[1] instanceof CodeModeFunction || args[1] instanceof CoercionFunction || args[1] instanceof UriFunction)
+    ) {
+      return invokeStringReplacer(runner, ref.receiver, ref.name, args, node)
+    }
+    return Effect.succeed(invokeStringMethod(ref.receiver, ref.name, args, node))
+  }
+  if (typeof ref.receiver === "number") {
+    return Effect.succeed(invokeNumberMethod(ref.receiver, ref.name, args, node))
+  }
+  if (Array.isArray(ref.receiver)) {
+    return invokeArrayMethod(runner, ref.receiver, ref.name, args, node)
+  }
+  if (ref.receiver instanceof SandboxDate) {
+    return Effect.succeed(invokeDateMethod(ref.receiver, ref.name, node))
+  }
+  if (ref.receiver instanceof SandboxRegExp) {
+    return Effect.succeed(invokeRegExpMethod(ref.receiver, ref.name, args, node))
+  }
+  if (ref.receiver instanceof SandboxMap) {
+    return invokeMapMethod(runner, ref.receiver, ref.name, args, node)
+  }
+  if (ref.receiver instanceof SandboxSet) {
+    return invokeSetMethod(runner, ref.receiver, ref.name, args, node)
+  }
+  if (ref.receiver instanceof SandboxURL) {
+    return Effect.succeed(invokeURLMethod(ref.receiver, ref.name, node))
+  }
+  if (ref.receiver instanceof SandboxURLSearchParams) {
+    return invokeURLSearchParamsMethod(runner, ref.receiver, ref.name, args, node)
+  }
+  throw new InterpreterRuntimeError(`Method '${ref.name}' is not available in CodeMode.`, node)
+}
+
+export const invokeGlobalMethod = (ref: GlobalMethodReference, args: Array<unknown>, node: AstNode): unknown => {
+  if (ref.namespace === "console")
+    throw new InterpreterRuntimeError(`console.${ref.name} is not available in CodeMode.`, node)
+  if (ref.namespace === "Object") return invokeObjectMethod(ref.name, args, node)
+  if (ref.namespace === "Math") return invokeMathMethod(ref.name, args, node)
+  if (ref.namespace === "Array") return invokeArrayStatic(ref.name, args, node)
+  if (ref.namespace === "Number") return invokeNumberStatic(ref.name, args, node)
+  if (ref.namespace === "String") return invokeStringStatic(ref.name, args, node)
+  if (ref.namespace === "URL") return invokeURLStatic(ref.name, args, node)
+  if (ref.namespace === "Date") return invokeDateStatic(ref.name, args, node)
+  if (
+    ref.namespace === "RegExp" ||
+    ref.namespace === "Map" ||
+    ref.namespace === "Set" ||
+    ref.namespace === "URLSearchParams"
+  ) {
+    throw new InterpreterRuntimeError(`${ref.namespace}.${ref.name} is not available in CodeMode.`, node)
+  }
+  return invokeJsonMethod(ref.name, args, node)
+}
+
+const invokeStringMethod = (value: string, name: string, args: Array<unknown>, node: AstNode): unknown => {
+  const str = (index: number): string => {
+    const arg = args[index]
+    if (typeof arg !== "string")
+      throw new InterpreterRuntimeError(`String.${name} expects argument ${index + 1} to be a string.`, node)
+    return arg
+  }
+  const num = (index: number): number => {
+    const arg = args[index]
+    if (typeof arg !== "number")
+      throw new InterpreterRuntimeError(`String.${name} expects argument ${index + 1} to be a number.`, node)
+    return arg
+  }
+  const optNum = (index: number): number | undefined => (args[index] === undefined ? undefined : num(index))
+  const optStr = (index: number): string | undefined => (args[index] === undefined ? undefined : str(index))
+
+  let result: unknown
+  switch (name) {
+    case "toLowerCase":
+      result = value.toLowerCase()
+      break
+    case "toUpperCase":
+      result = value.toUpperCase()
+      break
+    case "trim":
+      result = value.trim()
+      break
+    case "trimStart":
+      result = value.trimStart()
+      break
+    case "trimEnd":
+      result = value.trimEnd()
+      break
+    // Locale/options are deliberately unsupported; comparison uses the host default locale.
+    case "localeCompare":
+      result = value.localeCompare(str(0))
+      break
+    case "normalize": {
+      const form = optStr(0)
+      try {
+        result = value.normalize(form)
+      } catch {
+        throw new InterpreterRuntimeError(
+          `String.normalize expects the form "NFC", "NFD", "NFKC", or "NFKD" (got ${JSON.stringify(form)}).`,
+          node,
+        ).as("RangeError")
+      }
+      break
+    }
+    case "split": {
+      if (args.length === 0) {
+        result = [value]
+        break
+      }
+      if (args[0] instanceof SandboxRegExp) {
+        result = value.split(args[0].regex, optNum(1))
+        break
+      }
+      const requestedLimit = optNum(1)
+      result = value.split(str(0), requestedLimit === undefined ? undefined : requestedLimit >>> 0)
+      break
+    }
+    case "slice":
+      result = value.slice(optNum(0), optNum(1))
+      break
+    case "includes":
+      result = value.includes(str(0), optNum(1))
+      break
+    case "startsWith":
+      result = value.startsWith(str(0), optNum(1))
+      break
+    case "endsWith":
+      result = value.endsWith(str(0), optNum(1))
+      break
+    case "indexOf":
+      result = value.indexOf(str(0), optNum(1))
+      break
+    case "lastIndexOf":
+      result = value.lastIndexOf(str(0), optNum(1))
+      break
+    case "replace":
+    case "replaceAll": {
+      if (args[0] instanceof SandboxRegExp) {
+        const pattern = args[0].regex
+        const replacement = str(1)
+        if (name === "replaceAll" && !pattern.global) {
+          throw new InterpreterRuntimeError(
+            `String.replaceAll requires a regular expression with the global (g) flag: write /${pattern.source}/${pattern.flags}g, or use String.replace to replace only the first match.`,
+            node,
+          )
+        }
+        result = name === "replace" ? value.replace(pattern, replacement) : value.replaceAll(pattern, replacement)
+        break
+      }
+      if (name === "replace") {
+        result = value.replace(str(0), str(1))
+        break
+      }
+      result = value.replaceAll(str(0), str(1))
+      break
+    }
+    case "match": {
+      const pattern = toHostRegex(args[0], name, node)
+      const matched = value.match(pattern)
+      if (matched === null) return null
+      // Preserve the own `index` and `groups` properties on non-global matches.
+      if (pattern.global) return boundedData(matched, "String.match result")
+      return matchToValue(matched)
+    }
+    case "matchAll": {
+      const pattern = toHostRegex(args[0], name, node, "g")
+      if (!pattern.global) {
+        throw new InterpreterRuntimeError(
+          `String.matchAll requires a regular expression with the global (g) flag: write /${pattern.source}/${pattern.flags}g, or use String.match for a single match.`,
+          node,
+        )
+      }
+      return Array.from(value.matchAll(pattern), matchToValue)
+    }
+    case "search": {
+      result = value.search(toHostRegex(args[0], name, node))
+      break
+    }
+    case "repeat": {
+      const count = num(0)
+      if (!Number.isFinite(count) || count < 0)
+        throw new InterpreterRuntimeError("String.repeat expects a finite non-negative count.", node)
+      result = value.repeat(count)
+      break
+    }
+    case "padStart":
+      result = value.padStart(num(0), optStr(1))
+      break
+    case "padEnd":
+      result = value.padEnd(num(0), optStr(1))
+      break
+    case "charAt":
+      result = value.charAt(optNum(0) ?? 0)
+      break
+    case "at":
+      result = value.at(optNum(0) ?? 0)
+      break
+    case "substring":
+      result = value.substring(optNum(0) ?? 0, optNum(1))
+      break
+    case "charCodeAt":
+      result = value.charCodeAt(optNum(0) ?? 0)
+      break
+    case "codePointAt":
+      result = value.codePointAt(optNum(0) ?? 0)
+      break
+    case "toString":
+      result = value
+      break
+    case "concat": {
+      result = value.concat(...args.map((_, index) => str(index)))
+      break
+    }
+    default:
+      throw new InterpreterRuntimeError(`String method '${name}' is not available in CodeMode.`, node)
+  }
+  return boundedData(result, `String.${name} result`)
+}
+
+const invokeArrayStatic = (name: string, args: Array<unknown>, node: AstNode): unknown => {
+  switch (name) {
+    case "isArray":
+      return Array.isArray(args[0])
+    case "of":
+      return [...args]
+    case "from": {
+      if (args.length > 1) {
+        throw new InterpreterRuntimeError(
+          "Array.from(...) does not support a map function in CodeMode; call .map() on the result instead.",
+          node,
+          "UnsupportedSyntax",
+          [supportedSyntaxMessage],
+        )
+      }
+      if (args[0] instanceof SandboxMap) return Array.from(args[0].map.entries(), ([key, item]) => [key, item])
+      if (args[0] instanceof SandboxSet) return Array.from(args[0].set.values())
+      if (args[0] instanceof SandboxURLSearchParams) {
+        return Array.from(args[0].params.entries(), ([key, value]) => [key, value])
+      }
+      const source = args[0]
+      if (source instanceof SandboxPromise) {
+        throw new InterpreterRuntimeError(
+          "Array.from received an un-awaited Promise; await it before creating the array.",
+          node,
+          "InvalidDataValue",
+        )
+      }
+      if (typeof source === "string") return Array.from(source)
+      if (Array.isArray(source)) return [...source]
+      if (
+        source !== null &&
+        typeof source === "object" &&
+        (Object.getPrototypeOf(source) === Object.prototype || Object.getPrototypeOf(source) === null) &&
+        typeof (source as { length?: unknown }).length === "number"
+      ) {
+        return Array.from(source as ArrayLike<unknown>)
+      }
+      throw new InterpreterRuntimeError(
+        "Array.from expects an array, string, Map, Set, or array-like value.",
+        node,
+        "InvalidDataValue",
+      )
+    }
+    default:
+      throw new InterpreterRuntimeError(`Array.${name} is not available in CodeMode.`, node)
+  }
+}
+
+const invokeStringReplacer = <R>(
+  runner: CallbackRunner<R>,
+  value: string,
+  name: "replace" | "replaceAll",
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<unknown, unknown, R> => {
+  const apply = applyCollectionCallback(runner, args[1], `String.${name}`, node)
+  const matches: Array<{ readonly match: string; readonly offset: number; readonly args: Array<unknown> }> = []
+  const collect = (...callbackArgs: Array<unknown>): string => {
+    const match = callbackArgs[0]
+    const groups = callbackArgs[callbackArgs.length - 1]
+    const hasGroups = groups !== null && typeof groups === "object"
+    const offset = callbackArgs[callbackArgs.length - (hasGroups ? 3 : 2)]
+    if (typeof match !== "string" || typeof offset !== "number") {
+      throw new InterpreterRuntimeError(`String.${name} produced an invalid replacement match.`, node)
+    }
+    if (hasGroups) {
+      const safeGroups: SafeObject = Object.create(null) as SafeObject
+      for (const [key, group] of Object.entries(groups)) {
+        if (!isBlockedMember(key)) safeGroups[key] = group
+      }
+      callbackArgs[callbackArgs.length - 1] = safeGroups
+    }
+    matches.push({ match, offset, args: callbackArgs })
+    return match
+  }
+
+  const pattern = args[0]
+  if (pattern instanceof SandboxRegExp) {
+    if (name === "replaceAll" && !pattern.regex.global) {
+      throw new InterpreterRuntimeError(
+        `String.replaceAll requires a regular expression with the global (g) flag: write /${pattern.regex.source}/${pattern.regex.flags}g, or use String.replace to replace only the first match.`,
+        node,
+      )
+    }
+    if (name === "replace") value.replace(pattern.regex, collect)
+    else value.replaceAll(pattern.regex, collect)
+  } else {
+    if (typeof pattern !== "string") {
+      throw new InterpreterRuntimeError(`String.${name} expects argument 1 to be a string.`, node)
+    }
+    if (name === "replace") value.replace(pattern, collect)
+    else value.replaceAll(pattern, collect)
+  }
+
+  return Effect.gen(function* () {
+    const output: Array<string> = []
+    let end = 0
+    for (const match of matches) {
+      const replacement = yield* apply(match.args)
+      const resolved =
+        args[1] instanceof CodeModeFunction && args[1].async && replacement instanceof SandboxPromise
+          ? yield* runner.settlePromise(replacement)
+          : replacement
+      output.push(
+        value.slice(end, match.offset),
+        coerceToString(boundedData(resolved, `String.${name} replacer result`)),
+      )
+      end = match.offset + match.match.length
+    }
+    output.push(value.slice(end))
+    return boundedData(output.join(""), `String.${name} result`)
+  })
+}
+
+export const applyCollectionCallback = <R>(
+  runner: CallbackRunner<R>,
+  callback: unknown,
+  name: string,
+  node: AstNode,
+): ((args: Array<unknown>) => Effect.Effect<unknown, unknown, R>) => {
+  if (
+    !(callback instanceof CodeModeFunction) &&
+    !(callback instanceof CoercionFunction) &&
+    !(callback instanceof UriFunction) &&
+    !(callback instanceof PromiseCapabilityFunction)
+  ) {
+    throw new InterpreterRuntimeError(`${name} expects a function callback.`, node)
+  }
+  return (callbackArgs) =>
+    callback instanceof CoercionFunction
+      ? Effect.succeed(invokeCoercion(callback, callbackArgs, node))
+      : callback instanceof UriFunction
+        ? Effect.succeed(invokeUriFunction(callback, callbackArgs, node))
+        : callback instanceof PromiseCapabilityFunction
+          ? Effect.sync(() => callback.settle(callbackArgs[0]))
+          : runner.invokeFunction(callback, callbackArgs)
+}
+
+const invokeMapMethod = <R>(
+  runner: CallbackRunner<R>,
+  target: SandboxMap,
+  name: string,
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<unknown, unknown, R> => {
+  switch (name) {
+    case "get":
+      return Effect.succeed(target.map.get(args[0]))
+    case "has":
+      return Effect.succeed(target.map.has(args[0]))
+    case "set":
+      return Effect.sync(() => {
+        target.map.set(args[0], args[1])
+        return target
+      })
+    case "delete":
+      return Effect.sync(() => target.map.delete(args[0]))
+    case "clear":
+      return Effect.sync(() => {
+        target.map.clear()
+        return undefined
+      })
+    case "keys":
+      return Effect.sync(() => Array.from(target.map.keys()))
+    case "values":
+      return Effect.sync(() => Array.from(target.map.values()))
+    case "entries":
+      return Effect.sync(() => Array.from(target.map.entries(), ([key, item]): Array<unknown> => [key, item]))
+    case "forEach": {
+      const apply = applyCollectionCallback(runner, args[0], "Map.forEach", node)
+      return Effect.gen(function* () {
+        for (const [key, item] of Array.from(target.map.entries())) yield* apply([item, key, target])
+        return undefined
+      })
+    }
+    default:
+      throw new InterpreterRuntimeError(`Map method '${name}' is not available in CodeMode.`, node)
+  }
+}
+
+const invokeSetMethod = <R>(
+  runner: CallbackRunner<R>,
+  target: SandboxSet,
+  name: string,
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<unknown, unknown, R> => {
+  switch (name) {
+    case "has":
+      return Effect.succeed(target.set.has(args[0]))
+    case "add":
+      return Effect.sync(() => {
+        target.set.add(args[0])
+        return target
+      })
+    case "delete":
+      return Effect.sync(() => target.set.delete(args[0]))
+    case "clear":
+      return Effect.sync(() => {
+        target.set.clear()
+        return undefined
+      })
+    case "keys":
+    case "values":
+      return Effect.sync(() => Array.from(target.set.values()))
+    case "entries":
+      return Effect.sync(() => Array.from(target.set.values(), (item): Array<unknown> => [item, item]))
+    case "forEach": {
+      const apply = applyCollectionCallback(runner, args[0], "Set.forEach", node)
+      return Effect.gen(function* () {
+        for (const item of Array.from(target.set.values())) yield* apply([item, item, target])
+        return undefined
+      })
+    }
+    default:
+      throw new InterpreterRuntimeError(`Set method '${name}' is not available in CodeMode.`, node)
+  }
+}
+
+const invokeURLSearchParamsMethod = <R>(
+  runner: CallbackRunner<R>,
+  target: SandboxURLSearchParams,
+  name: string,
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<unknown, unknown, R> => {
+  const arg = (index: number): string => uriArgument(args[index], `URLSearchParams.${name} argument ${index + 1}`)
+  const requireArgs = (count: number): void => {
+    if (args.length < count) {
+      throw new InterpreterRuntimeError(
+        `URLSearchParams.${name} requires ${count} argument${count === 1 ? "" : "s"}.`,
+        node,
+      ).as("TypeError")
+    }
+  }
+  switch (name) {
+    case "append": {
+      requireArgs(2)
+      return Effect.sync(() => {
+        target.params.append(arg(0), arg(1))
+        return undefined
+      })
+    }
+    case "delete": {
+      requireArgs(1)
+      return Effect.sync(() => {
+        if (args[1] !== undefined) target.params.delete(arg(0), arg(1))
+        else target.params.delete(arg(0))
+        return undefined
+      })
+    }
+    case "get":
+      requireArgs(1)
+      return Effect.sync(() => target.params.get(arg(0)))
+    case "getAll":
+      requireArgs(1)
+      return Effect.sync(() => target.params.getAll(arg(0)))
+    case "has":
+      requireArgs(1)
+      return Effect.sync(() => (args[1] !== undefined ? target.params.has(arg(0), arg(1)) : target.params.has(arg(0))))
+    case "set": {
+      requireArgs(2)
+      return Effect.sync(() => {
+        target.params.set(arg(0), arg(1))
+        return undefined
+      })
+    }
+    case "sort":
+      return Effect.sync(() => {
+        target.params.sort()
+        return undefined
+      })
+    case "keys":
+      return Effect.sync(() => Array.from(target.params.keys()))
+    case "values":
+      return Effect.sync(() => Array.from(target.params.values()))
+    case "entries":
+      return Effect.sync(() => Array.from(target.params.entries(), ([key, value]): Array<unknown> => [key, value]))
+    case "toString":
+      return Effect.sync(() => target.params.toString())
+    case "forEach": {
+      requireArgs(1)
+      const apply = applyCollectionCallback(runner, args[0], "URLSearchParams.forEach", node)
+      return Effect.gen(function* () {
+        for (const [key, value] of Array.from(target.params.entries())) yield* apply([value, key, target])
+        return undefined
+      })
+    }
+    default:
+      throw new InterpreterRuntimeError(`URLSearchParams method '${name}' is not available in CodeMode.`, node)
+  }
+}
+
+const invokeArrayMethod = <R>(
+  runner: CallbackRunner<R>,
+  target: Array<unknown>,
+  name: string,
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<unknown, unknown, R> => {
+  const optNumber = (value: unknown, label: string): number | undefined => {
+    if (value === undefined) return undefined
+    if (typeof value !== "number")
+      throw new InterpreterRuntimeError(`Array.${name} expects ${label} to be a number.`, node)
+    return value
+  }
+  switch (name) {
+    case "join": {
+      if (args.length > 1 || (args.length === 1 && typeof args[0] !== "string")) {
+        throw new InterpreterRuntimeError("Array.join expects zero arguments or one string separator.", node)
+      }
+      const input = boundedData(target, "Array.join input") as Array<unknown>
+      return Effect.succeed(
+        input.map((item) => coerceToString(item ?? "")).join(args.length === 0 ? "," : (args[0] as string)),
+      )
+    }
+    case "includes":
+      if (args.length === 0 || args.length > 2)
+        throw new InterpreterRuntimeError("Array.includes expects a value and optional start index.", node)
+      return Effect.succeed(target.includes(args[0], optNumber(args[1], "start index")))
+    case "indexOf":
+      return Effect.succeed(target.indexOf(args[0], optNumber(args[1], "start index")))
+    case "lastIndexOf":
+      return Effect.succeed(
+        args[1] === undefined
+          ? target.lastIndexOf(args[0])
+          : target.lastIndexOf(args[0], optNumber(args[1], "start index")),
+      )
+    case "at":
+      return Effect.succeed(target.at(optNumber(args[0], "index") ?? 0))
+    case "slice":
+      return Effect.succeed(target.slice(optNumber(args[0], "start"), optNumber(args[1], "end")))
+    case "concat":
+      return Effect.succeed(target.concat(...args))
+    case "flat":
+      return Effect.succeed(target.flat(optNumber(args[0], "depth") ?? 1))
+    case "reverse":
+      return Effect.succeed(target.reverse())
+    case "sort":
+      return Effect.map(sortArray(runner, target, args[0], node), (sorted) => {
+        target.splice(0, target.length, ...sorted)
+        return target
+      })
+    case "toSorted":
+      return sortArray(runner, target, args[0], node)
+    case "toReversed":
+      return Effect.succeed([...target].reverse())
+    case "with": {
+      const index = optNumber(args[0], "index") ?? 0
+      const resolved = index < 0 ? target.length + index : index
+      if (resolved < 0 || resolved >= target.length) {
+        throw new InterpreterRuntimeError("Array.with index is out of range.", node)
+      }
+      const copied = [...target]
+      copied[resolved] = args[1]
+      return Effect.succeed(copied)
+    }
+    case "push": {
+      // Validate all insertions before mutating to avoid partial cyclic updates.
+      for (const item of args) rejectCircularInsertion(target, item, "Array.push result", node)
+      target.push(...args)
+      return Effect.succeed(target.length)
+    }
+    case "unshift": {
+      for (const item of args) rejectCircularInsertion(target, item, "Array.unshift result", node)
+      target.unshift(...args)
+      return Effect.succeed(target.length)
+    }
+    case "pop":
+      return Effect.succeed(target.pop())
+    case "shift":
+      return Effect.succeed(target.shift())
+    case "splice": {
+      if (args.length === 0) return Effect.succeed(target.splice(0, 0))
+      const start = optNumber(args[0], "start") ?? 0
+      if (args.length === 1) return Effect.succeed(target.splice(start))
+      const deleteCount = optNumber(args[1], "delete count") ?? 0
+      const inserted = args.slice(2)
+      for (const item of inserted) rejectCircularInsertion(target, item, "Array.splice result", node)
+      return Effect.succeed(target.splice(start, deleteCount, ...inserted))
+    }
+    case "fill": {
+      rejectCircularInsertion(target, args[0], "Array.fill result", node)
+      return Effect.succeed(target.fill(args[0], optNumber(args[1], "start"), optNumber(args[2], "end")))
+    }
+    case "copyWithin":
+      return Effect.succeed(
+        target.copyWithin(
+          optNumber(args[0], "target index") ?? 0,
+          optNumber(args[1], "start") ?? 0,
+          optNumber(args[2], "end"),
+        ),
+      )
+    case "keys":
+      return Effect.succeed(Array.from(target.keys()))
+    case "values":
+      return Effect.succeed([...target])
+    case "entries":
+      return Effect.succeed(Array.from(target.entries(), ([index, item]): Array<unknown> => [index, item]))
+  }
+
+  const apply = applyCollectionCallback(runner, args[0], `Array.${name}`, node)
+  return Effect.gen(function* () {
+    // Fix iteration length while reading existing elements live.
+    const length = target.length
+    switch (name) {
+      case "map": {
+        const values: Array<unknown> = []
+        values.length = length
+        for (let index = 0; index < length; index += 1) {
+          if (!(index in target)) continue
+          values[index] = yield* apply([target[index], index, target])
+        }
+        return values
+      }
+      case "flatMap": {
+        const values: Array<unknown> = []
+        for (let index = 0; index < length; index += 1) {
+          if (!(index in target)) continue
+          const mapped = yield* apply([target[index], index, target])
+          if (Array.isArray(mapped)) values.push(...mapped)
+          else values.push(mapped)
+        }
+        return values
+      }
+      case "filter": {
+        const values: Array<unknown> = []
+        for (let index = 0; index < length; index += 1) {
+          if (!(index in target)) continue
+          const item = target[index]
+          if (yield* apply([item, index, target])) values.push(item)
+        }
+        return values
+      }
+      case "find":
+        for (let index = 0; index < length; index += 1) {
+          const item = target[index]
+          if (yield* apply([item, index, target])) return item
+        }
+        return undefined
+      case "findIndex":
+        for (let index = 0; index < length; index += 1) {
+          if (yield* apply([target[index], index, target])) return index
+        }
+        return -1
+      case "some":
+        for (let index = 0; index < length; index += 1) {
+          if (!(index in target)) continue
+          if (yield* apply([target[index], index, target])) return true
+        }
+        return false
+      case "every":
+        for (let index = 0; index < length; index += 1) {
+          if (!(index in target)) continue
+          if (!(yield* apply([target[index], index, target]))) return false
+        }
+        return true
+      case "forEach":
+        for (let index = 0; index < length; index += 1) {
+          if (index in target) yield* apply([target[index], index, target])
+        }
+        return undefined
+      case "reduce": {
+        let start = 0
+        let accumulator = args[1]
+        if (args.length < 2) {
+          while (start < length && !(start in target)) start += 1
+          if (start === length)
+            throw new InterpreterRuntimeError("Array.reduce of an empty array with no initial value.", node).as(
+              "TypeError",
+            )
+          accumulator = target[start]
+          start += 1
+        }
+        for (let index = start; index < length; index += 1) {
+          if (!(index in target)) continue
+          accumulator = yield* apply([accumulator, target[index], index, target])
+        }
+        return accumulator
+      }
+      case "reduceRight": {
+        let start = length - 1
+        let accumulator = args[1]
+        if (args.length < 2) {
+          while (start >= 0 && !(start in target)) start -= 1
+          if (start < 0)
+            throw new InterpreterRuntimeError("Array.reduceRight of an empty array with no initial value.", node).as(
+              "TypeError",
+            )
+          accumulator = target[start]
+          start -= 1
+        }
+        for (let index = start; index >= 0; index -= 1) {
+          if (!(index in target)) continue
+          accumulator = yield* apply([accumulator, target[index], index, target])
+        }
+        return accumulator
+      }
+      case "findLast":
+        for (let index = length - 1; index >= 0; index -= 1) {
+          const item = target[index]
+          if (yield* apply([item, index, target])) return item
+        }
+        return undefined
+      case "findLastIndex":
+        for (let index = length - 1; index >= 0; index -= 1) {
+          if (yield* apply([target[index], index, target])) return index
+        }
+        return -1
+    }
+    throw new InterpreterRuntimeError(`Array method '${name}' is not available in CodeMode.`, node)
+  })
+}
+
+const sortArray = <R>(
+  runner: CallbackRunner<R>,
+  target: Array<unknown>,
+  comparator: unknown,
+  node: AstNode,
+): Effect.Effect<Array<unknown>, unknown, R> => {
+  if (comparator !== undefined && !(comparator instanceof CodeModeFunction)) {
+    throw new InterpreterRuntimeError("Array.sort expects an arrow function comparator.", node)
+  }
+  if (!(comparator instanceof CodeModeFunction)) {
+    return Effect.sync(() =>
+      [...target].sort((a, b) => {
+        const left = coerceToString(a)
+        const right = coerceToString(b)
+        return left < right ? -1 : left > right ? 1 : 0
+      }),
+    )
+  }
+  const mergeSort = (items: Array<unknown>): Effect.Effect<Array<unknown>, unknown, R> => {
+    if (items.length <= 1) return Effect.succeed(items)
+    const midpoint = Math.floor(items.length / 2)
+    return Effect.gen(function* () {
+      const left = yield* mergeSort(items.slice(0, midpoint))
+      const right = yield* mergeSort(items.slice(midpoint))
+      const merged: Array<unknown> = []
+      let leftIndex = 0
+      let rightIndex = 0
+      while (leftIndex < left.length && rightIndex < right.length) {
+        // Treat a NaN comparator result as equal to preserve stable ordering.
+        const order = coerceToNumber(yield* runner.invokeFunction(comparator, [left[leftIndex], right[rightIndex]]))
+        if (Number.isNaN(order) || order <= 0) merged.push(left[leftIndex++])
+        else merged.push(right[rightIndex++])
+      }
+      return [...merged, ...left.slice(leftIndex), ...right.slice(rightIndex)]
+    })
+  }
+  const defined = target.filter((item) => item !== undefined)
+  const undefinedCount = target.length - defined.length
+  return Effect.map(mergeSort(defined), (items) => [...items, ...Array(undefinedCount).fill(undefined)])
+}

packages/codemode/src/interpreter/model.ts

@@ -76,8 +76,6 @@ export class PromiseInstanceMethodReference {
   ) {}
 }
 
-// The resolve/reject callables handed to a `new Promise(executor)` executor. `settle` closes
-// over the promise's deferred and is first-settlement-wins; later calls are no-ops, as in JS.
 export class PromiseCapabilityFunction {
   constructor(readonly settle: (value: unknown) => void) {}
 }
@@ -114,8 +112,6 @@ export class UriFunction {
   constructor(readonly name: "encodeURI" | "encodeURIComponent" | "decodeURI" | "decodeURIComponent") {}
 }
 
-// The global `search` built-in: synchronous tool discovery that shares the tool admission
-// pipeline (budget, audit, hooks) without living in the `tools` tree.
 export class SearchFunction {}
 
 export class ProgramThrow {

packages/codemode/src/interpreter/promises.ts

@@ -0,0 +1,336 @@
+import { Cause, Deferred, Effect, Exit, Fiber, Scope } from "effect"
+import type { Diagnostic } from "../codemode.js"
+import type { SafeObject } from "../tool-runtime.js"
+import {
+  type AstNode,
+  CodeModeFunction,
+  CoercionFunction,
+  InterpreterRuntimeError,
+  ProgramThrow,
+  PromiseCapabilityFunction,
+  PromiseInstanceMethodReference,
+  PromiseMethodReference,
+  UriFunction,
+} from "./model.js"
+import { caughtErrorValue, normalizeError } from "./errors.js"
+import { applyCollectionCallback, type CallbackRunner } from "./methods.js"
+import { typeofValue } from "./references.js"
+import { spreadItems } from "../stdlib/collections.js"
+import { createAggregateErrorValue } from "../stdlib/value.js"
+import { SandboxPromise } from "../values.js"
+
+// Observation only controls rejection reporting; program completion interrupts all promise work.
+export class PromiseRuntime<R> {
+  private readonly active = new Set<SandboxPromise>()
+  private readonly ids = new WeakMap<SandboxPromise, number>()
+  private readonly observed = new WeakSet<SandboxPromise>()
+  private readonly failures = new Map<number, Diagnostic>()
+  private nextID = 0
+
+  constructor(private readonly scope: Scope.Scope) {}
+
+  create(effect: Effect.Effect<unknown, unknown, R>): Effect.Effect<SandboxPromise, never, R> {
+    return Effect.suspend(() => {
+      // Allocate before forking so reruns get distinct IDs and diagnostics retain creation order.
+      const id = this.nextID++
+      return Effect.map(Effect.forkIn(effect, this.scope, { startImmediately: true }), (fiber) => {
+        const promise = new SandboxPromise(fiber)
+        this.active.add(promise)
+        this.ids.set(promise, id)
+        fiber.addObserver((exit) => {
+          this.active.delete(promise)
+          if (Exit.isSuccess(exit) || Cause.hasInterruptsOnly(exit.cause) || this.observed.has(promise)) {
+            this.ids.delete(promise)
+            return
+          }
+          const failure = normalizeError(Cause.squash(exit.cause))
+          this.failures.set(id, {
+            ...failure,
+            message: `Unhandled rejection from an un-awaited promise: ${failure.message}`,
+          })
+        })
+        return promise
+      })
+    })
+  }
+
+  // Observation must be recorded when responsibility transfers, before the consumer fiber runs.
+  markObserved(promise: SandboxPromise): void {
+    this.observed.add(promise)
+    const id = this.ids.get(promise)
+    this.ids.delete(promise)
+    if (id !== undefined) this.failures.delete(id)
+  }
+
+  await(promise: SandboxPromise): Effect.Effect<Exit.Exit<unknown, unknown>> {
+    return Fiber.await(promise.fiber)
+  }
+
+  diagnostics(): Array<Diagnostic> {
+    return [...this.failures].sort(([left], [right]) => left - right).map(([, failure]) => failure)
+  }
+
+  // Re-check because a straggler can create promises before its interruption lands.
+  interrupt(): Effect.Effect<Array<Diagnostic>> {
+    const self = this
+    return Effect.gen(function* () {
+      while (self.active.size > 0) {
+        yield* Fiber.interruptAll([...self.active].map((promise) => promise.fiber))
+      }
+      return self.diagnostics()
+    })
+  }
+}
+
+export const selfResolutionError = (node?: AstNode): InterpreterRuntimeError =>
+  new InterpreterRuntimeError("Chaining cycle detected: a promise cannot resolve with itself.", node).as("TypeError")
+
+export const invokePromiseMethod = <R>(
+  runner: CallbackRunner<R>,
+  promises: PromiseRuntime<R>,
+  ref: PromiseMethodReference,
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<unknown, unknown, R> => {
+  if (ref.name === "resolve") {
+    const value = args[0]
+    return value instanceof SandboxPromise ? Effect.succeed(value) : promises.create(Effect.succeed(value))
+  }
+  if (ref.name === "reject") {
+    return promises.create(Effect.fail(new ProgramThrow(args[0])))
+  }
+
+  const spread = spreadItems(args[0])
+  if (spread === undefined) {
+    return promises.create(
+      Effect.fail(
+        new InterpreterRuntimeError(
+          `Promise.${ref.name} expects an array of promises or plain values (e.g. Promise.${ref.name}(items.map((item) => tools.ns.tool(item)))).`,
+          node,
+        ).as("TypeError"),
+      ),
+    )
+  }
+  const items = Array.from(spread)
+
+  for (const item of items) {
+    if (item instanceof SandboxPromise) promises.markObserved(item)
+  }
+
+  switch (ref.name) {
+    case "all": {
+      const observations = items.map((item) =>
+        item instanceof SandboxPromise ? Effect.flatten(promises.await(item)) : Effect.succeed(item),
+      )
+      return promises.create(settleAfterTurn(Effect.all(observations, { concurrency: "unbounded" })))
+    }
+    case "allSettled": {
+      const observations = items.map((item) =>
+        item instanceof SandboxPromise ? promises.await(item) : Effect.succeed(Exit.succeed(item)),
+      )
+      return promises.create(
+        settleAfterTurn(
+          Effect.gen(function* () {
+            const outcomes: Array<unknown> = []
+            for (const observation of observations) {
+              const exit = yield* observation
+              if (Exit.isSuccess(exit)) {
+                outcomes.push(
+                  Object.assign(Object.create(null) as SafeObject, { status: "fulfilled", value: exit.value }),
+                )
+                continue
+              }
+              if (Cause.hasInterruptsOnly(exit.cause)) {
+                // Teardown interruption is not a program-level rejection.
+                return yield* Effect.failCause(exit.cause)
+              }
+              outcomes.push(
+                Object.assign(Object.create(null) as SafeObject, {
+                  status: "rejected",
+                  reason: caughtErrorValue(Cause.squash(exit.cause)),
+                }),
+              )
+            }
+            return outcomes
+          }),
+        ),
+      )
+    }
+    case "race": {
+      if (items.length === 0) {
+        return promises.create(
+          Effect.fail(
+            new InterpreterRuntimeError(
+              "Promise.race([]) would never settle; provide at least one promise or value.",
+              node,
+            ),
+          ),
+        )
+      }
+      const observations = items.map((item) =>
+        item instanceof SandboxPromise ? promises.await(item) : Effect.succeed(Exit.succeed(item)),
+      )
+      return promises.create(settleAfterTurn(Effect.flatten(Effect.raceAll(observations))))
+    }
+    case "any": {
+      const flipped = items.map((item) =>
+        item instanceof SandboxPromise
+          ? Effect.flatMap(promises.await(item), (exit) => {
+              if (Exit.isSuccess(exit)) return Effect.fail(new PromiseAnyFulfilled(exit.value))
+              if (Cause.hasInterruptsOnly(exit.cause)) return Effect.failCause(exit.cause)
+              return Effect.succeed(caughtErrorValue(Cause.squash(exit.cause)))
+            })
+          : Effect.fail(new PromiseAnyFulfilled(item)),
+      )
+      const body = Effect.all(flipped, { concurrency: "unbounded" }).pipe(
+        Effect.flatMap((reasons) =>
+          Effect.fail(new ProgramThrow(createAggregateErrorValue(reasons, "All promises were rejected"))),
+        ),
+        Effect.catch((error) =>
+          error instanceof PromiseAnyFulfilled ? Effect.succeed(error.value) : Effect.fail(error),
+        ),
+      )
+      return promises.create(settleAfterTurn(body))
+    }
+  }
+}
+
+export const invokePromiseInstanceMethod = <R>(
+  runner: CallbackRunner<R>,
+  promises: PromiseRuntime<R>,
+  ref: PromiseInstanceMethodReference,
+  args: Array<unknown>,
+  node: AstNode,
+): Effect.Effect<SandboxPromise, never, R> => {
+  const method = `Promise.prototype.${ref.name}`
+  promises.markObserved(ref.promise)
+  if (ref.name === "finally") {
+    return chainFinally(runner, promises, ref.promise, reactionHandler(args[0], method, node), method, node)
+  }
+  const onFulfilled = ref.name === "then" ? reactionHandler(args[0], method, node) : undefined
+  const onRejected = reactionHandler(ref.name === "then" ? args[1] : args[0], method, node)
+  return chainReaction(runner, promises, ref.promise, onFulfilled, onRejected, method, node)
+}
+
+export const constructPromise = <R>(
+  runner: CallbackRunner<R>,
+  promises: PromiseRuntime<R>,
+  executor: unknown,
+  node: AstNode,
+): Effect.Effect<SandboxPromise, unknown, R> => {
+  if (!(executor instanceof CodeModeFunction)) {
+    throw new InterpreterRuntimeError(
+      "new Promise(...) expects an executor function (e.g. new Promise((resolve, reject) => { ... })).",
+      node,
+    ).as("TypeError")
+  }
+  return Effect.gen(function* () {
+    const deferred = Deferred.makeUnsafe<unknown, unknown>()
+    const box: { own?: SandboxPromise } = {}
+    const promise = yield* promises.create(
+      Effect.flatMap(Deferred.await(deferred), (value) => {
+        if (!(value instanceof SandboxPromise)) return Effect.succeed(value)
+        if (value === box.own) return Effect.fail(selfResolutionError(node))
+        return runner.settlePromise(value)
+      }),
+    )
+    box.own = promise
+    const resolve = new PromiseCapabilityFunction((value) => {
+      Deferred.doneUnsafe(deferred, Exit.succeed(value))
+    })
+    const reject = new PromiseCapabilityFunction((value) => {
+      Deferred.doneUnsafe(deferred, Exit.fail(new ProgramThrow(value)))
+    })
+    const executed = yield* Effect.exit(runner.invokeFunction(executor, [resolve, reject]))
+    if (!Exit.isSuccess(executed)) {
+      if (Cause.hasInterruptsOnly(executed.cause)) return yield* Effect.failCause(executed.cause)
+      Deferred.doneUnsafe(deferred, Exit.fail(Cause.squash(executed.cause)))
+    }
+    return promise
+  })
+}
+
+// Settle one reaction turn after the deciding member, after its existing reactions.
+const settleAfterTurn = <A, E, R>(body: Effect.Effect<A, E, R>): Effect.Effect<A, E, R> =>
+  Effect.flatMap(Effect.exit(body), (exit) => Effect.andThen(Effect.yieldNow, exit))
+
+class PromiseAnyFulfilled {
+  constructor(readonly value: unknown) {}
+}
+
+type ReactionHandler = CodeModeFunction | CoercionFunction | UriFunction | PromiseCapabilityFunction
+
+const reactionHandler = (value: unknown, method: string, node: AstNode): ReactionHandler | undefined => {
+  if (
+    value instanceof CodeModeFunction ||
+    value instanceof CoercionFunction ||
+    value instanceof UriFunction ||
+    value instanceof PromiseCapabilityFunction
+  ) {
+    return value
+  }
+  if (typeofValue(value) === "function") {
+    throw new InterpreterRuntimeError(
+      `${method} handlers must be plain functions; wrap other callables in an arrow function, e.g. (value) => tools.ns.tool(value).`,
+      node,
+    )
+  }
+  return undefined
+}
+
+// Teardown bypasses handlers; settled reactions yield once so handlers never run inline.
+const reactionExit = <R>(
+  promises: PromiseRuntime<R>,
+  source: SandboxPromise,
+): Effect.Effect<Exit.Exit<unknown, unknown>, unknown, R> =>
+  Effect.gen(function* () {
+    const exit = yield* promises.await(source)
+    if (!Exit.isSuccess(exit) && Cause.hasInterruptsOnly(exit.cause)) return yield* Effect.failCause(exit.cause)
+    yield* Effect.yieldNow
+    return exit
+  })
+
+const chainReaction = <R>(
+  runner: CallbackRunner<R>,
+  promises: PromiseRuntime<R>,
+  source: SandboxPromise,
+  onFulfilled: ReactionHandler | undefined,
+  onRejected: ReactionHandler | undefined,
+  method: string,
+  node: AstNode,
+): Effect.Effect<SandboxPromise, never, R> => {
+  const box: { derived?: SandboxPromise } = {}
+  const body = Effect.gen(function* () {
+    const exit = yield* reactionExit(promises, source)
+    const handler = Exit.isSuccess(exit) ? onFulfilled : onRejected
+    if (handler === undefined) return yield* exit
+    const input = Exit.isSuccess(exit) ? exit.value : caughtErrorValue(Cause.squash(exit.cause))
+    const result = yield* applyCollectionCallback(runner, handler, method, node)([input])
+    if (result === box.derived) return yield* Effect.fail(selfResolutionError(node))
+    if (result instanceof SandboxPromise) return yield* runner.settlePromise(result)
+    return result
+  })
+  return Effect.map(promises.create(body), (derived) => {
+    box.derived = derived
+    return derived
+  })
+}
+
+const chainFinally = <R>(
+  runner: CallbackRunner<R>,
+  promises: PromiseRuntime<R>,
+  source: SandboxPromise,
+  cleanup: ReactionHandler | undefined,
+  method: string,
+  node: AstNode,
+): Effect.Effect<SandboxPromise, never, R> =>
+  promises.create(
+    Effect.gen(function* () {
+      const exit = yield* reactionExit(promises, source)
+      if (cleanup !== undefined) {
+        const result = yield* applyCollectionCallback(runner, cleanup, method, node)([])
+        if (result instanceof SandboxPromise) yield* runner.settlePromise(result)
+      }
+      return yield* exit
+    }),
+  )

packages/codemode/src/interpreter/references.ts

@@ -0,0 +1,99 @@
+import {
+  type AstNode,
+  CodeModeFunction,
+  CoercionFunction,
+  ErrorConstructorReference,
+  GlobalMethodReference,
+  GlobalNamespace,
+  InterpreterRuntimeError,
+  IntrinsicReference,
+  PromiseCapabilityFunction,
+  PromiseInstanceMethodReference,
+  PromiseMethodReference,
+  PromiseNamespace,
+  SearchFunction,
+  UriFunction,
+} from "./model.js"
+import { ToolReference } from "../tool-runtime.js"
+import { isSandboxValue, SandboxPromise } from "../values.js"
+
+export const isRuntimeReference = (value: unknown): boolean =>
+  value instanceof CodeModeFunction ||
+  value instanceof ToolReference ||
+  value instanceof IntrinsicReference ||
+  value instanceof GlobalNamespace ||
+  value instanceof GlobalMethodReference ||
+  value instanceof PromiseNamespace ||
+  value instanceof PromiseMethodReference ||
+  value instanceof PromiseInstanceMethodReference ||
+  value instanceof SandboxPromise ||
+  value instanceof CoercionFunction ||
+  value instanceof UriFunction ||
+  value instanceof SearchFunction ||
+  value instanceof PromiseCapabilityFunction ||
+  value instanceof ErrorConstructorReference ||
+  isSandboxValue(value)
+
+export const containsRuntimeReference = (value: unknown, seen = new Set<object>()): boolean => {
+  if (isRuntimeReference(value)) return true
+  if (value === null || typeof value !== "object") return false
+  if (seen.has(value)) return false
+  seen.add(value)
+  const contains = Array.isArray(value)
+    ? value.some((item) => containsRuntimeReference(item, seen))
+    : Object.values(value).some((item) => containsRuntimeReference(item, seen))
+  seen.delete(value)
+  return contains
+}
+
+// Sandbox values are data here, not opaque interpreter references.
+export const containsOpaqueReference = (value: unknown, seen = new Set<object>()): boolean => {
+  if (isSandboxValue(value)) return false
+  if (isRuntimeReference(value)) return true
+  if (value === null || typeof value !== "object") return false
+  if (seen.has(value)) return false
+  seen.add(value)
+  const contains = Array.isArray(value)
+    ? value.some((item) => containsOpaqueReference(item, seen))
+    : Object.values(value).some((item) => containsOpaqueReference(item, seen))
+  seen.delete(value)
+  return contains
+}
+
+// Reject cycles before mutation so later boundary walks remain safe.
+export const rejectCircularInsertion = (
+  container: object,
+  value: unknown,
+  label: string,
+  node: AstNode,
+  seen = new Set<object>(),
+): void => {
+  if (value === container)
+    throw new InterpreterRuntimeError(`${label} contains a circular value.`, node, "InvalidDataValue")
+  if (value === null || typeof value !== "object" || isRuntimeReference(value) || seen.has(value)) return
+  seen.add(value)
+  const items = Array.isArray(value) ? value : Object.values(value)
+  for (const item of items) rejectCircularInsertion(container, item, label, node, seen)
+  seen.delete(value)
+}
+
+export const typeofValue = (value: unknown): string => {
+  if (
+    value instanceof CodeModeFunction ||
+    value instanceof CoercionFunction ||
+    value instanceof IntrinsicReference ||
+    value instanceof GlobalMethodReference ||
+    value instanceof PromiseMethodReference ||
+    value instanceof PromiseInstanceMethodReference ||
+    value instanceof PromiseNamespace ||
+    value instanceof PromiseCapabilityFunction ||
+    value instanceof ErrorConstructorReference
+  )
+    return "function"
+  if (value instanceof UriFunction || value instanceof SearchFunction) return "function"
+  if (value instanceof ToolReference) return value.path.length > 0 ? "function" : "object"
+  if (value instanceof GlobalNamespace) {
+    return value.name === "Math" || value.name === "JSON" || value.name === "console" ? "object" : "function"
+  }
+  return typeof value
+}

packages/codemode/src/interpreter/scope.ts

@@ -0,0 +1,84 @@
+import { type AstNode, type Binding, InterpreterRuntimeError } from "./model.js"
+
+export class ScopeStack {
+  private readonly scopes: Array<Map<string, Binding>>
+
+  constructor(scopes: Array<Map<string, Binding>>) {
+    this.scopes = scopes
+  }
+
+  declare(name: string, value: unknown, mutable: boolean, node: AstNode): void {
+    const scope = this.current()
+
+    const existing = scope.get(name)
+    if (existing && existing.initialized !== false) {
+      throw new InterpreterRuntimeError(`Identifier '${name}' has already been declared.`, node)
+    }
+
+    scope.set(name, { mutable, value, initialized: true })
+  }
+
+  get(name: string, node: AstNode): unknown {
+    const binding = this.resolve(name)
+
+    if (!binding) {
+      throw new InterpreterRuntimeError(`Unknown identifier '${name}'.`, node).as("ReferenceError")
+    }
+
+    if (binding.initialized === false) {
+      throw new InterpreterRuntimeError(`Cannot access '${name}' before initialization.`, node).as("ReferenceError")
+    }
+
+    return binding.value
+  }
+
+  set(name: string, value: unknown, node: AstNode): unknown {
+    const binding = this.resolve(name)
+
+    if (!binding) {
+      throw new InterpreterRuntimeError(`Unknown identifier '${name}'.`, node).as("ReferenceError")
+    }
+
+    if (!binding.mutable) {
+      throw new InterpreterRuntimeError(`Cannot assign to constant '${name}'.`, node).as("TypeError")
+    }
+
+    binding.value = value
+    return value
+  }
+
+  resolve(name: string): Binding | undefined {
+    for (let index = this.scopes.length - 1; index >= 0; index -= 1) {
+      const scope = this.scopes[index]
+      const binding = scope?.get(name)
+
+      if (binding) {
+        return binding
+      }
+    }
+
+    return undefined
+  }
+
+  current(): Map<string, Binding> {
+    const scope = this.scopes[this.scopes.length - 1]
+
+    if (!scope) {
+      throw new InterpreterRuntimeError("Interpreter scope stack is empty.")
+    }
+
+    return scope
+  }
+
+  push(scope: Map<string, Binding> = new Map()): void {
+    this.scopes.push(scope)
+  }
+
+  pop(): void {
+    this.scopes.pop()
+  }
+
+  capture(): Array<Map<string, Binding>> {
+    return this.scopes.slice()
+  }
+}

packages/codemode/src/openapi/index.ts

@@ -31,10 +31,8 @@ export type {
 } from "./types.js"
 
 /**
- * Builds a CodeMode tool subtree from an OpenAPI 3.x document, one tool per
- * operation. Auth is resolved host-side via `auth.resolve` and never
- * model-visible. Tools require `HttpClient.HttpClient`; unrepresentable
- * operations land in `skipped`.
+ * Builds one CodeMode tool per representable OpenAPI 3.x operation. Auth remains host-side,
+ * tools require `HttpClient.HttpClient`, and unrepresentable operations land in `skipped`.
  */
 export const fromSpec = (options: Options): Result => {
   const document = options.spec

packages/codemode/src/openapi/runtime.ts

@@ -56,7 +56,7 @@ const buildRequest = (
   input: Readonly<Record<string, unknown>>,
 ): Effect.Effect<HttpClientRequest.HttpClientRequest, ToolError> =>
   Effect.gen(function* () {
-    // Validate every model-controlled value before auth resolution, which may refresh tokens.
+    // Validate model input before auth resolution can refresh credentials.
     const url = buildUrl(plan, input)
     if (url instanceof ToolError) return yield* Effect.fail(url)
     const missing = plan.fields.find(
@@ -77,7 +77,6 @@ const buildRequest = (
       request = serialized
     }
 
-    // Host headers first, then declared header parameters.
     request = HttpClientRequest.setHeaders(request, plan.headers)
     for (const field of plan.fields) {
       if (field.location !== "header") continue
@@ -169,7 +168,7 @@ const applyCredentials = (
       continue
     }
     if (credential.type === "basic") {
-      // Buffer instead of btoa: btoa throws on non-Latin-1 credentials.
+      // Basic auth credentials are UTF-8; btoa rejects non-Latin-1 input.
       const duplicate = add(
         "header",
         "authorization",
@@ -183,7 +182,6 @@ const applyCredentials = (
       if (duplicate !== undefined) return duplicate
       continue
     }
-    // apiKey: the carrier comes from the scheme declaration.
     if (definition.type !== "apiKey") {
       return toolError(
         `Security scheme '${name}' is not an apiKey scheme; resolve a bearer, basic, or header credential for it.`,
@@ -212,8 +210,7 @@ const buildUrl = (plan: Plan, input: Readonly<Record<string, unknown>>): string
       ),
     )
     if (fieldValue instanceof ToolError) return fieldValue
-    // '.'/'..' survive encoding and URL normalization collapses them, letting a
-    // model-supplied value retarget the request to a different endpoint.
+    // URL normalization collapses encoded `.` and `..`, which could retarget the request.
     if (fieldValue === "" || fieldValue === "." || fieldValue === "..") {
       return toolError(`Invalid path parameter '${field.inputName}'.`)
     }

packages/codemode/src/openapi/spec.ts

@@ -23,8 +23,7 @@ const asArray = (value: unknown): ReadonlyArray<unknown> => (Array.isArray(value
 export const nonEmptyString = (value: unknown): string | undefined =>
   typeof value === "string" && value !== "" ? value : undefined
 
-// Guards record lookups keyed by spec- or model-controlled names against
-// prototype-inherited values (e.g. a parameter named `toString`).
+// Spec- and model-controlled keys must not resolve inherited properties.
 export const own = <T>(record: Readonly<Record<string, T>>, key: string): T | undefined =>
   Object.hasOwn(record, key) ? record[key] : undefined
 
@@ -106,7 +105,7 @@ const operationParameters = (
   pathItem: Record<string, unknown>,
   operation: Record<string, unknown>,
 ): Parsed<ReadonlyArray<PlannedField>> => {
-  // Operation-level parameters override path-level ones sharing (location, name).
+  // OpenAPI operation parameters override path parameters with the same location and name.
   const declared = new Map<
     string,
     { readonly name: string; readonly location: string; readonly parameter: Record<string, unknown> }

packages/codemode/src/openapi/types.ts

@@ -22,9 +22,8 @@ export type SecurityScheme =
   | { readonly type: "openIdConnect" }
 
 /**
- * Credential material returned by a host auth resolver. The carrier for `apiKey`
- * comes from the scheme definition, not the credential. `header` is the escape
- * hatch for nonstandard schemes.
+ * Credential material returned by a host auth resolver. `apiKey` uses the scheme's carrier;
+ * `header` supports nonstandard schemes.
  */
 export type Credential =
   | { readonly type: "bearer"; readonly token: string }
@@ -33,9 +32,7 @@ export type Credential =
   | { readonly type: "header"; readonly name: string; readonly value: string }
 
 /**
- * Resolves credential material for one named security scheme at call time.
- * `undefined` means unavailable, try the next OR alternative; a failure aborts
- * the call rather than falling through.
+ * Resolves credentials at call time. `undefined` tries the next OR alternative; failure aborts.
  */
 export type AuthResolver = (context: {
   readonly name: string
@@ -74,9 +71,7 @@ export type Parsed<T> = { readonly ok: true; readonly value: T } | { readonly ok
 export type InputLocation = "path" | "query" | "header" | "body"
 
 export type InputField = {
-  /** Model-visible field name after cross-location collision handling. */
   readonly inputName: string
-  /** Original parameter or body-property name used on the wire. */
   readonly name: string
   readonly location: InputLocation
   readonly required: boolean
@@ -92,7 +87,6 @@ export type OperationInput = {
   readonly body: Body | undefined
 }
 
-/** One OR alternative: scheme name -> required scopes. Empty object = unauthenticated is acceptable. */
 export type SecurityRequirement = Readonly<Record<string, ReadonlyArray<string>>>
 
 export type Plan = {

packages/codemode/src/stdlib/console.ts

@@ -1,4 +1,122 @@
+import { containsOpaqueReference, containsRuntimeReference, isRuntimeReference } from "../interpreter/references.js"
+import { copyIn, copyOut } from "../tool-runtime.js"
+import {
+  isSandboxValue,
+  SandboxDate,
+  SandboxMap,
+  SandboxPromise,
+  SandboxRegExp,
+  SandboxSet,
+  SandboxURL,
+  SandboxURLSearchParams,
+} from "../values.js"
+import { boundedData, coerceToString } from "./value.js"
+
 export const consoleMethods = new Set(["log", "info", "debug", "warn", "error", "dir", "table"])
 
-/** Console formatting recursion ceiling; deeper values render as "...". */
-export const MAX_CONSOLE_DEPTH = 32
+const MAX_CONSOLE_DEPTH = 32
+
+export const formatConsoleMessage = (name: string, args: Array<unknown>): string => {
+  if (name === "dir") return args.length === 0 ? "undefined" : formatConsoleArgument(args[0])
+  if (name === "table") return formatConsoleTable(args[0], args[1])
+  const prefix = name === "warn" ? "[warn] " : name === "error" ? "[error] " : name === "debug" ? "[debug] " : ""
+  return `${prefix}${args.map((arg) => formatConsoleArgument(arg)).join(" ")}`
+}
+
+const formatConsoleArgument = (value: unknown): string => {
+  if (value === undefined) return "undefined"
+  if (typeof value === "string") return value
+  return formatConsoleValue(value, new Set(), 0)
+}
+
+const formatConsoleValue = (value: unknown, seen: Set<object>, depth: number): string => {
+  if (value === null || value === undefined) return "null"
+  if (typeof value === "string") return JSON.stringify(value)
+  if (typeof value === "number" || typeof value === "boolean") return String(value)
+  if (typeof value !== "object") return String(value)
+  if (value instanceof SandboxPromise) return "[Promise (await it to get its value)]"
+  if (value instanceof SandboxDate) return coerceToString(value)
+  if (value instanceof SandboxRegExp) return coerceToString(value)
+  if (value instanceof SandboxURL) return coerceToString(value)
+  if (value instanceof SandboxURLSearchParams) return coerceToString(value)
+  if (depth > MAX_CONSOLE_DEPTH) return "..."
+  if (seen.has(value)) return "[Circular]"
+  if (value instanceof SandboxMap) {
+    seen.add(value)
+    try {
+      const entries = Array.from(value.map.entries(), ([key, item]): Array<unknown> => [key, item])
+      return `Map(${value.map.size}) ${formatConsoleValue(entries, seen, depth + 1)}`
+    } finally {
+      seen.delete(value)
+    }
+  }
+  if (value instanceof SandboxSet) {
+    seen.add(value)
+    try {
+      return `Set(${value.set.size}) ${formatConsoleValue(Array.from(value.set.values()), seen, depth + 1)}`
+    } finally {
+      seen.delete(value)
+    }
+  }
+  if (isRuntimeReference(value)) return "[CodeMode reference]"
+  seen.add(value)
+  try {
+    if (Array.isArray(value)) {
+      return `[${value.map((item) => formatConsoleValue(item, seen, depth + 1)).join(",")}]`
+    }
+    return `{${Object.entries(value)
+      .map(([key, item]) => `${JSON.stringify(key)}:${formatConsoleValue(item, seen, depth + 1)}`)
+      .join(",")}}`
+  } finally {
+    seen.delete(value)
+  }
+}
+
+const formatConsoleTable = (value: unknown, columnsArgument: unknown): string => {
+  if (value === undefined) return "undefined"
+  if (containsOpaqueReference(value)) return "[CodeMode reference]"
+  const data = boundedData(value, "console.table argument")
+  const columns = consoleTableColumns(columnsArgument)
+  const rows = consoleTableRows(data, columns)
+  const keys = columns ?? Array.from(new Set(rows.flatMap((row) => Object.keys(row.values))))
+  const header = ["(index)", ...keys].join("\t")
+  return [
+    header,
+    ...rows.map((row) => [row.index, ...keys.map((key) => formatConsoleTableCell(row.values[key]))].join("\t")),
+  ].join("\n")
+}
+
+const consoleTableColumns = (value: unknown): ReadonlyArray<string> | undefined => {
+  if (value === undefined) return undefined
+  if (containsRuntimeReference(value)) return undefined
+  const columns = copyOut(copyIn(value, "console.table columns"), true)
+  return Array.isArray(columns) ? columns.map((column) => String(column)) : undefined
+}
+
+const consoleTableRows = (
+  data: unknown,
+  columns: ReadonlyArray<string> | undefined,
+): Array<{ readonly index: string; readonly values: Record<string, unknown> }> => {
+  if (Array.isArray(data)) {
+    return data.map((item, index) => ({ index: String(index), values: consoleTableValues(item, columns) }))
+  }
+  if (data !== null && typeof data === "object" && !isSandboxValue(data)) {
+    return Object.entries(data).map(([index, item]) => ({ index, values: consoleTableValues(item, columns) }))
+  }
+  return [{ index: "0", values: { Value: data } }]
+}
+
+const consoleTableValues = (value: unknown, columns: ReadonlyArray<string> | undefined): Record<string, unknown> => {
+  if (value !== null && typeof value === "object" && !Array.isArray(value) && !isSandboxValue(value)) {
+    const source = value as Record<string, unknown>
+    if (columns !== undefined) return Object.fromEntries(columns.map((column) => [column, source[column]]))
+    return Object.fromEntries(Object.entries(source))
+  }
+  return { Value: value }
+}
+
+const formatConsoleTableCell = (value: unknown): string => {
+  if (value === undefined) return ""
+  if (typeof value === "string") return value
+  return formatConsoleValue(value, new Set(), 0)
+}

packages/codemode/src/stdlib/promise.ts

@@ -1,6 +1,3 @@
 import type { PromiseMethodName } from "../interpreter/model.js"
 
 export const promiseStatics = new Set<PromiseMethodName>(["all", "allSettled", "race", "any", "resolve", "reject"])
-
-/** Maximum number of eagerly forked tool calls that may run concurrently. */
-export const TOOL_CALL_CONCURRENCY = 8

packages/codemode/src/stdlib/string.ts

@@ -4,12 +4,9 @@ export const stringMethods = new Set([
   "trim",
   "trimStart",
   "trimEnd",
-  "trimLeft",
-  "trimRight",
   "split",
   "slice",
   "substring",
-  "substr",
   "includes",
   "startsWith",
   "endsWith",

packages/codemode/src/tool-runtime.ts

@@ -44,36 +44,30 @@ type ServicesOf<Tools, Depth extends ReadonlyArray<unknown>> = Depth["length"] e
           : ServicesOf<Tools[keyof Tools], [...Depth, unknown]>
         : never
 
-/** Minimal audit record retained for each admitted tool call. */
 export type ToolCall = {
   readonly name: string
 }
 
-/** Decoded tool call observed immediately before tool execution. */
 export type ToolCallStarted = {
   readonly index: number
   readonly name: string
   readonly input: unknown
 }
 
-/** Completed tool call observed immediately after tool execution settles. */
 export type ToolCallEnded = {
   readonly index: number
   readonly name: string
   readonly input: unknown
   readonly durationMs: number
   readonly outcome: "success" | "failure"
-  /** Model-safe failure message; present only when `outcome` is `"failure"`. */
   readonly message?: string
 }
 
-/** Non-throwing observation hooks fired around each admitted tool call. */
 export type ToolCallHooks<R = never> = {
   readonly onToolCallStart?: ((call: ToolCallStarted) => Effect.Effect<void, never, R>) | undefined
   readonly onToolCallEnd?: ((call: ToolCallEnded) => Effect.Effect<void, never, R>) | undefined
 }
 
-/** Model-visible description of one schema-backed tool. */
 export type ToolDescription = {
   readonly path: string
   readonly description: string
@@ -113,11 +107,6 @@ export class ToolReference {
   constructor(readonly path: ReadonlyArray<string>) {}
 }
 
-/**
- * Maximum nesting depth for values crossing a data boundary. Fixed (not a configurable
- * limit) purely because it produces a clearer diagnostic than a native stack-overflow
- * RangeError would.
- */
 const MAX_VALUE_DEPTH = 32
 
 export class ToolRuntimeError extends Error {
@@ -152,21 +141,7 @@ const blockedMemberNames = new Set(["__proto__", "constructor", "prototype"])
 
 export const isBlockedMember = (name: string): boolean => blockedMemberNames.has(name)
 
-/**
- * Validates and copies a value against the plain-data contract (depth, circularity, plain
- * objects only, blocked properties, data-only leaves).
- *
- * Two modes share the walk:
- * - **Boundary** (`preserveSandboxValues` false, the default): the host<->sandbox boundary -
- *   final results, tool-call arguments, `JSON.stringify`. Sandbox value types serialize
- *   exactly as JSON.stringify would: Date/URL -> strings, the remaining value types -> {}.
- * - **Intra-sandbox checkpoint** (`preserveSandboxValues` true; see `boundedData` in
- *   codemode.ts): standard-library value instances pass through untouched (treated as leaves,
- *   contents not walked), so values flowing through `Object.*` helpers, coercion inputs, and
- *   other in-sandbox checkpoints stay fully usable (`.getTime()`, `.has()`, ...).
- *
- * Both modes reject un-awaited promises with an await-hinting diagnostic.
- */
+// Checkpoint mode preserves sandbox values; boundary mode JSON-normalizes them.
 export const copyIn = (value: unknown, label: string, preserveSandboxValues = false): unknown =>
   copyBounded(value, label, 0, new Set(), preserveSandboxValues)
 
@@ -185,10 +160,6 @@ const copyBounded = (
     value === undefined ||
     typeof value === "string" ||
     typeof value === "boolean" ||
-    // NaN/Infinity are allowed to exist as in-sandbox intermediates (matching real JS and a real
-    // engine) so defensive guards like `Number.isNaN(x)` / `parseInt(x) || 0` can run. They are
-    // normalized to `null` when the value leaves the sandbox - see copyOut - exactly as
-    // JSON.stringify already does at any tool boundary.
     typeof value === "number"
   ) {
     return value
@@ -198,8 +169,6 @@ const copyBounded = (
     throw new ToolRuntimeError("InvalidDataValue", `${label} must contain data only.`)
   }
 
-  // An un-awaited promise never crosses a data checkpoint as `{}`; the diagnostic tells the
-  // model exactly how to fix the program instead.
   if (value instanceof SandboxPromise) {
     throw new ToolRuntimeError(
       "InvalidDataValue",
@@ -208,9 +177,6 @@ const copyBounded = (
   }
 
   if (preserveSandboxValues) {
-    // Intra-sandbox checkpoints keep sandbox value instances alive as leaves; their contents
-    // are never walked here (Map/Set members are validated where mutation happens, and the
-    // real boundary still serializes them below).
     if (
       value instanceof SandboxDate ||
       value instanceof SandboxRegExp ||
@@ -221,8 +187,6 @@ const copyBounded = (
     ) {
       return value
     }
-    // Host instances cannot normally reach an intra-sandbox checkpoint (tool results cross
-    // the boundary first), but wrap them defensively rather than degrading to JSON forms.
     if (value instanceof Date) return new SandboxDate(value.getTime())
     if (value instanceof RegExp) return new SandboxRegExp(value.source, value.flags)
     if (value instanceof Map) {
@@ -241,9 +205,6 @@ const copyBounded = (
     if (value instanceof URLSearchParams) return new SandboxURLSearchParams(new URLSearchParams(value))
   }
 
-  // Sandbox value types (and their host counterparts, which a host tool may legitimately
-  // return) serialize exactly as JSON.stringify would at the data boundary: Date/URL use
-  // toJSON(), while RegExp/Map/Set/URLSearchParams have no JSON form beyond {}.
   if (value instanceof SandboxDate) {
     return Number.isFinite(value.time) ? new Date(value.time).toISOString() : null
   }
@@ -274,7 +235,7 @@ const copyBounded = (
   if (Array.isArray(value)) {
     const copied = value.map((item) => copyBounded(item, label, depth + 1, seen, preserveSandboxValues))
     if (preserveSandboxValues) {
-      // Array metadata is not serialized, but intra-sandbox copies must retain it.
+      // Checkpoint copies retain array metadata that boundary copies omit.
       for (const [key, item] of Object.entries(value)) {
         if (Object.hasOwn(copied, key)) continue
         if (isBlockedMember(key)) {
@@ -305,9 +266,6 @@ const copyBounded = (
 
 export const copyOut = (value: unknown, undefinedAsNull = false): unknown => {
   if (value === undefined && undefinedAsNull) return null
-  // Normalize non-finite numbers to null as the value crosses out of the sandbox (final return
-  // and tool-call arguments both funnel through here), matching JSON semantics - NaN/Infinity
-  // have no JSON representation, so JSON.stringify would produce null anyway.
   if (typeof value === "number" && !Number.isFinite(value)) {
     return null
   }
@@ -353,18 +311,10 @@ export type DiscoveryPlan = {
 
 export type SearchEntry = {
   readonly description: ToolDescription
-  /** Top-level namespace (first path segment), matched by the search `namespace` option. */
   readonly namespace: string
-  /** Lowercased path + description + input property names/descriptions, for substring matching. */
   readonly searchText: string
 }
 
-/**
- * Split a query into lowercased search terms. camelCase boundaries are split
- * (`resolveLibrary` -> `resolve library`) and every non-alphanumeric character is a
- * separator, so `resolve-library-id`, `resolveLibraryId`, and `resolve library id` all
- * tokenize alike. Empties and the `*` wildcard are dropped.
- */
 const tokenize = (query: string): Array<string> =>
   query
     .replace(/([a-z0-9])([A-Z])/g, "$1 $2")
@@ -372,13 +322,6 @@ const tokenize = (query: string): Array<string> =>
     .split(/[^a-z0-9]+/)
     .filter((term) => term.length > 0 && term !== "*")
 
-/**
- * A term plus its naive singular variants (trailing "s"/"es" stripped), so a plural
- * query term ("issues") still matches indexed text that only carries the singular
- * ("issue"). Matching is one-directional substring containment, so the variants are
- * needed only on the query side; scoring weights are unchanged - each field check
- * passes when ANY form matches.
- */
 const termForms = (term: string): Array<string> => {
   const forms = [term]
   if (term.endsWith("es") && term.length > 3) forms.push(term.slice(0, -2))
@@ -400,8 +343,6 @@ const makeSearchTool = (searchIndex: ReadonlyArray<SearchEntry>): Definition =>
         request.namespace === undefined
           ? searchIndex
           : searchIndex.filter((entry) => entry.namespace === request.namespace)
-      // A query that names one tool path exactly (canonical path or rendered JavaScript
-      // expression) is a lookup, not a search: return that tool alone.
       const trimmed = query.trim()
       const pathQuery = trimmed.startsWith("tools.") ? trimmed.slice("tools.".length) : trimmed
       const exact =
@@ -411,9 +352,6 @@ const makeSearchTool = (searchIndex: ReadonlyArray<SearchEntry>): Definition =>
               (entry) => entry.description.path === pathQuery || toolExpression(entry.description.path) === trimmed,
             )
       const terms = tokenize(query).map(termForms)
-      // Additive field-weighted scoring, summed across terms: exact path or path segment
-      // (20) > path substring (8) > description substring (4) > any searchable text,
-      // including input parameter names and descriptions (2).
       const ranked =
         exact !== undefined
           ? [exact]
@@ -451,15 +389,12 @@ const makeSearchTool = (searchIndex: ReadonlyArray<SearchEntry>): Definition =>
     }),
 })
 
-// The built-in `search` is a synchronous global function, not a tool-tree entry, so its
-// advertised signature is rendered by hand instead of through `describeDefinition`.
 const searchSignature = (() => {
   const definition = makeSearchTool([])
   return `search(input: ${inputTypeScript(definition, true)}): ${outputTypeScript(definition, true)}`
 })()
 
 const catalogLine = (tool: ToolDescription) => {
-  // Keep the tool description concise; the full schema documentation remains in the signature.
   const line = tool.description.split("\n", 1)[0]!.trim()
   const description = line.length > 120 ? line.slice(0, 119) + "..." : line
   return description === "" ? `  - ${tool.signature}` : `  - ${tool.signature} // ${description}`
@@ -479,21 +414,10 @@ const toSearchEntry = <R>(path: string, definition: Definition<R>, description:
     .toLowerCase(),
 })
 
-/** The runtime search index over every described tool. Search is always registered. */
 export const searchIndex = <R>(tools: HostTools<R>): ReadonlyArray<SearchEntry> =>
   visibleDefinitions(tools).map(({ path, definition, description }) => toSearchEntry(path, definition, description))
 
-/**
- * Budgeted catalog: every namespace is always listed with its tool count; full call
- * signatures are inlined against the `catalogBudget` (estimated tokens,
- * chars/4) round-robin across namespaces - in each round (namespaces alphabetical), every
- * namespace still holding un-inlined tools attempts to place its next-cheapest line, and
- * a namespace whose next line does not fit is done while the others keep going - so every
- * namespace gets some representation before any namespace gets everything. The section
- * states exactly how comprehensive it is - overall (COMPLETE vs PARTIAL) and per
- * namespace. Namespace stub lines are never budgeted: every namespace appears with its
- * tool count even at budget 0.
- */
+// Budget signatures round-robin so every namespace remains visible.
 export const prepare = <R>(tools: HostTools<R>, catalogBudget = defaultCatalogBudget): DiscoveryPlan => {
   if (!Number.isSafeInteger(catalogBudget) || catalogBudget < 0) {
     throw new RangeError("discovery.catalogBudget must be a non-negative safe integer")
@@ -510,12 +434,6 @@ export const prepare = <R>(tools: HostTools<R>, catalogBudget = defaultCatalogBu
   }
   const ordered = [...namespaces].sort(([left], [right]) => left.localeCompare(right))
 
-  // Select which signatures fit the budget before emitting, so the list can state
-  // exactly how comprehensive it is. Round-robin fairness: in each round (namespaces
-  // alphabetical), every namespace still holding un-inlined tools tries to place its
-  // next-cheapest line against the shared budget; a namespace whose next line does not
-  // fit is done - the others keep going - so every namespace gets some representation
-  // before any namespace gets everything.
   const selections = ordered.map(([namespace, group]) => ({
     namespace,
     picked: new Set<ToolDescription>(),
@@ -547,10 +465,6 @@ export const prepare = <R>(tools: HostTools<R>, catalogBudget = defaultCatalogBu
 
   const empty = described.length === 0
 
-  // Section order is deliberate: workflow first (the top is the least likely part of a long
-  // description to be truncated or skimmed away), then rules, then syntax, with the budgeted
-  // catalog at the bottom. Example call forms use placeholders - never a real or fabricated
-  // tool name - and show both dot and bracket notation so non-identifier names are not normalized.
   const intro = [
     empty
       ? "This is a restricted JavaScript language for calling tools, not a general-purpose runtime."
@@ -562,8 +476,6 @@ export const prepare = <R>(tools: HostTools<R>, catalogBudget = defaultCatalogBu
       : ["Do not infer or normalize tool names; use only exact signatures shown below or returned by search."]),
   ]
 
-  // The search step exists only when search is advertised (PARTIAL catalog); a COMPLETE
-  // catalog already shows every signature, so step 1 picks from the list instead.
   const workflow = empty
     ? []
     : [
@@ -627,8 +539,6 @@ export const prepare = <R>(tools: HostTools<R>, catalogBudget = defaultCatalogBu
     for (const [namespace, group] of ordered) {
       const picked = shown.get(namespace)!
       const count = `${group.length} tool${group.length === 1 ? "" : "s"}`
-      // Annotate only when a namespace is not fully shown, so a comprehensive
-      // namespace reads cleanly and a truncated one is unambiguous.
       const label =
         picked.size === group.length
           ? count
@@ -651,13 +561,6 @@ export const prepare = <R>(tools: HostTools<R>, catalogBudget = defaultCatalogBu
   }
 }
 
-/**
- * The enumerable names at one node of the callable tool tree - namespace names at the root,
- * tool/namespace names below - powering `Object.keys(tools)` and `for...in` over tool
- * references. A callable tool is a leaf and enumerates as `[]` (like `Object.keys` of a
- * function in JS). An unknown path is an `UnknownTool` error pointing at the working
- * discovery idioms, mirroring how calling an unknown tool fails.
- */
 const namespaceKeys = <R>(tools: HostTools<R>, path: ReadonlyArray<string>): ReadonlyArray<string> => {
   let value: HostTool<R> | Definition<R> | HostTools<R> = tools
   for (const segment of path) {
@@ -705,18 +608,12 @@ export type ToolRuntime<R = never> = {
   readonly root: ToolReference
   readonly calls: Array<ToolCall>
   readonly invoke: (path: ReadonlyArray<string>, args: Array<unknown>) => Effect.Effect<unknown, unknown, R>
-  /**
-   * The built-in `search` global: a synchronous discovery call that shares the tool
-   * admission pipeline (budget, audit, hooks) without living in the `tools` tree.
-   */
   readonly search: (args: Array<unknown>) => Effect.Effect<unknown, unknown, R>
-  /** Enumerable namespace/tool names at one node of the callable tool tree; see `namespaceKeys`. */
   readonly keys: (path: ReadonlyArray<string>) => ReadonlyArray<string>
 }
 
 export const make = <R>(
   tools: HostTools<R>,
-  /** Undefined means unlimited tool calls. */
   maxToolCalls: number | undefined,
   searchIndex: ReadonlyArray<SearchEntry>,
   hooks?: ToolCallHooks<R>,
@@ -724,8 +621,7 @@ export const make = <R>(
   const calls: Array<ToolCall> = []
   const searchTool = makeSearchTool(searchIndex)
 
-  // Wraps the settling portion of a tool call so onToolCallEnd observes success and failure
-  // symmetrically. Interruption (e.g. the execution timeout) fires neither outcome.
+  // End hooks observe settled success or failure; interruption emits neither outcome.
   const observeEnd = <A, E>(effect: Effect.Effect<A, E, R>, call: ToolCallStarted): Effect.Effect<A, E, R> => {
     const onEnd = hooks?.onToolCallEnd
     if (onEnd === undefined) return effect

packages/codemode/src/tool-schema.ts

@@ -5,13 +5,8 @@ const isEffectSchema = (schema: SchemaType): schema is Schema.Decoder<unknown> &
 
 const renderLiteral = (value: unknown): string => JSON.stringify(value) ?? "unknown"
 
-/**
- * Bare TypeScript identifier - usable unquoted as an object key (and, in the tool runtime,
- * with dot access as a tool-path segment). Anything else must be quoted/bracketed.
- */
 export const identifierSegment = /^[A-Za-z_$][A-Za-z0-9_$]*$/
 
-/** Renders a property name as a valid TS object key: bare when an identifier, quoted otherwise. */
 const renderKey = (name: string): string => (identifierSegment.test(name) ? name : JSON.stringify(name))
 
 const effectNumberSentinel = (schema: JsonSchema) =>
@@ -27,16 +22,10 @@ const intersection = (members: ReadonlyArray<string>): string => {
   return concrete.map((member) => (member.includes(" | ") ? `(${member})` : member)).join(" & ")
 }
 
-/**
- * Recursion ceiling for schema rendering. Object, array, and union recursion all increment
- * depth, so this bounds every recursion path - pathological or structurally cyclic schemas
- * degrade to `unknown` instead of overflowing the stack (rendering must never throw).
- */
 const MAX_RENDER_DEPTH = 8
 
 type RenderContext = {
   readonly definitions: Readonly<Record<string, JsonSchema>>
-  /** Indented, JSDoc-annotated multiline rendering (search results); compact single line otherwise. */
   readonly pretty: boolean
 }
 
@@ -64,10 +53,6 @@ const hasUnresolvedRef = (
   ].some((item) => hasUnresolvedRef(item, definitions, seen, nextVisited))
 }
 
-/**
- * Schema constraints a TypeScript type cannot express natively but a model benefits from,
- * surfaced as JSDoc tags (`@deprecated`, `@default`, `@format`, `@minItems`, `@maxItems`).
- */
 const docTags = (schema: JsonSchema): Array<string> => {
   const tags: Array<string> = []
   if (schema.deprecated === true) tags.push("@deprecated")
@@ -75,9 +60,7 @@ const docTags = (schema: JsonSchema): Array<string> => {
     try {
       const rendered = JSON.stringify(schema.default)
       if (rendered !== undefined) tags.push(`@default ${rendered}`)
-    } catch {
-      // unserializable default: skip rather than emit a broken tag
-    }
+    } catch {}
   }
   if (typeof schema.format === "string") tags.push(`@format ${schema.format}`)
   if (typeof schema.minItems === "number") tags.push(`@minItems ${schema.minItems}`)
@@ -85,13 +68,7 @@ const docTags = (schema: JsonSchema): Array<string> => {
   return tags
 }
 
-/**
- * Format a schema `description` plus `tags` as a JSDoc comment at the given indent,
- * preserving multi-line text (a single line stays `/** ... *\/`; multiple lines become a
- * `*`-prefixed block). `*\/` is neutralized so nothing can close the comment early, and
- * blank leading/trailing lines are trimmed. Returns "" (else a trailing newline) so
- * callers can prepend it directly to the field line.
- */
+// Neutralize `*\/` so model-provided schema text cannot terminate generated documentation.
 const jsdoc = (description: string | undefined, tags: ReadonlyArray<string>, pad: string): string => {
   const lines = [...(description === undefined ? [] : description.split("\n")), ...tags].map((line) =>
     line.replaceAll("*/", "* /").replace(/\s+$/, ""),
@@ -128,17 +105,11 @@ const renderSchema = (
   if (schema.enum) return schema.enum.map(renderLiteral).join(" | ")
   const alternatives = schema.anyOf ?? schema.oneOf
   if (alternatives) {
-    // Effect's number schema emits `anyOf: [{ type: "number" }, { const: "NaN" },
-    // { const: "Infinity" }, { const: "-Infinity" }]`. Collapse only that artifact;
-    // real JSON Schema unions such as `string | number` or `number | null` must keep
-    // every branch.
     if (
       alternatives.some((item) => item.type === "number") &&
       alternatives.every((item) => item.type === "number" || effectNumberSentinel(item))
     )
       return "number"
-    // An empty Schema.Struct({}) emits `anyOf: [{ type: "object" }, { type: "array" }]`
-    // (no properties/items); render the bare shape as {} instead of `{} | Array<unknown>`.
     if (
       alternatives.length === 2 &&
       alternatives[0]?.type === "object" &&
@@ -183,7 +154,6 @@ const renderSchema = (
       return fields.length === 0 ? "{}" : `{ ${fields.join("; ")} }`
     }
 
-    // Pretty: an indented block, each described field preceded by its JSDoc comment.
     if (properties.length === 0 && indexType === undefined) return "{}"
     const pad = "  ".repeat(depth + 1)
     const lines = properties.map(
@@ -208,7 +178,6 @@ export const toTypeScript = (schema: Schema.Top, decoded = false, pretty = false
   }
 }
 
-/** Renders a raw JSON Schema document as a TypeScript type string. */
 export const jsonSchemaToTypeScript = (schema: JsonSchema, pretty = false): string => {
   try {
     return renderSchema(schema, { definitions: { ...(schema.definitions ?? {}), ...(schema.$defs ?? {}) }, pretty })
@@ -217,20 +186,12 @@ export const jsonSchemaToTypeScript = (schema: JsonSchema, pretty = false): stri
   }
 }
 
-/** One input property of a tool, extracted best-effort from its input schema. */
 export type InputProperty = {
   readonly name: string
   readonly description: string | undefined
   readonly required: boolean
 }
 
-/**
- * The property names, descriptions, and required flags of a tool's input schema - the raw
- * material for search text. Best-effort: Effect Schemas go through their
- * JSON Schema document (the same emission signature rendering uses); JSON Schemas are read
- * directly, resolving a trivial top-level `$ref` into `$defs`/`definitions` when present.
- * Anything unresolvable yields `[]` (search falls back to path + description).
- */
 export const inputProperties = <R>(definition: Definition<R>): Array<InputProperty> => {
   try {
     const document = isEffectSchema(definition.input)
@@ -262,20 +223,11 @@ export const inputProperties = <R>(definition: Definition<R>): Array<InputProper
   }
 }
 
-/**
- * The model-visible TypeScript type of a tool's input. `pretty` renders an indented
- * multiline block with schema descriptions and constraints as JSDoc comments on the
- * fields; the default stays the compact single-line form.
- */
 export const inputTypeScript = <R>(definition: Definition<R>, pretty = false): string =>
   isEffectSchema(definition.input)
     ? toTypeScript(definition.input, false, pretty)
     : jsonSchemaToTypeScript(definition.input, pretty)
 
-/**
- * The model-visible TypeScript type of a tool's result; tools without an output schema
- * return `unknown`. `pretty` renders the JSDoc-annotated multiline form, as for inputs.
- */
 export const outputTypeScript = <R>(definition: Definition<R>, pretty = false): string =>
   definition.output === undefined
     ? "unknown"
@@ -283,18 +235,9 @@ export const outputTypeScript = <R>(definition: Definition<R>, pretty = false):
       ? toTypeScript(definition.output, true, pretty)
       : jsonSchemaToTypeScript(definition.output, pretty)
 
-/**
- * Decodes tool input before `run` is invoked. Effect Schemas validate (throwing on failure);
- * JSON-Schema-described inputs pass through unvalidated (render-only).
- */
 export const decodeInput = <R>(definition: Definition<R>, value: unknown): unknown =>
   isEffectSchema(definition.input) ? Schema.decodeUnknownSync(definition.input)(value) : value
 
-/**
- * Decodes a tool result before it is exposed to the program. Effect Schemas validate and
- * transform (throwing on failure); JSON Schema outputs and tools without an output schema pass
- * the host value through unchanged.
- */
 export const decodeOutput = <R>(definition: Definition<R>, value: unknown): unknown =>
   definition.output !== undefined && isEffectSchema(definition.output)
     ? Schema.decodeUnknownSync(definition.output)(value)

packages/codemode/src/tool.ts

@@ -1,11 +1,8 @@
 import { Effect, Schema } from "effect"
 
 /**
- * JSON Schema subset accepted for render-only tool schemas.
- *
- * A JSON-Schema-described side of a tool is used to generate the model-visible TypeScript
- * signature only - CodeMode performs no validation against it. This is the natural shape for
- * adapter-provided tools (e.g. MCP definitions) whose schemas arrive as JSON Schema documents.
+ * JSON Schema subset for model-visible signatures. CodeMode does not validate values against
+ * these schemas.
  */
 export type JsonSchema = {
   readonly type?: string | ReadonlyArray<string>
@@ -41,10 +38,8 @@ export type Definition<R = never> = {
   readonly run: (input: unknown) => Effect.Effect<unknown, unknown, R>
 }
 
-/** The value `run` receives: the decoded type for Effect Schemas, `unknown` for JSON Schemas. */
 type InputType<S> = S extends Schema.Decoder<unknown> ? S["Type"] : unknown
 
-/** The value `run` returns: the encoded type for Effect Schemas, `unknown` otherwise. */
 type ResultType<S> = S extends Schema.Decoder<unknown> ? S["Encoded"] : unknown
 
 /** Options for defining one CodeMode tool. */
@@ -61,29 +56,9 @@ export const isDefinition = <R = never>(value: unknown): value is Definition<R>
 /**
  * Defines one schema-described tool available to a CodeMode program through `tools.*`.
  *
- * `input` and `output` each accept a validating Effect Schema or a render-only JSON Schema
- * document. Effect Schema input is decoded before `run` is invoked, and `run` returns the
- * encoded representation of an Effect Schema `output`, which CodeMode decodes before returning
- * it to the program. JSON Schemas only shape the model-visible signature; values pass through
- * unvalidated. `output` is optional - without it the signature advertises `unknown` and the
- * host result is exposed as-is. The host tool remains responsible for authorization and
- * durable side-effect handling.
- *
- * @example
- * ```ts
- * const lookup = Tool.make({
- *   description: "Look up an order",
- *   input: Schema.Struct({ id: Schema.String }),
- *   output: Schema.Struct({ status: Schema.String }),
- *   run: ({ id }) => Effect.succeed({ status: "open" }),
- * })
- *
- * const fromJsonSchema = Tool.make({
- *   description: "Call an adapter-described tool",
- *   input: { type: "object", properties: { id: { type: "string" } }, required: ["id"] },
- *   run: (input) => callHost(input),
- * })
- * ```
+ * Effect Schemas validate values; JSON Schemas only shape the model-visible signature.
+ * Without `output`, results are exposed as `unknown`. Hosts remain responsible for authorization
+ * and durable side effects.
  */
 export const make = <I extends SchemaType, const O extends SchemaType | undefined = undefined, R = never>(
   options: Options<I, O, R>,

packages/codemode/test/array-callbacks-test262.test.ts

@@ -26,9 +26,11 @@
  * - test/built-ins/Array/prototype/forEach/15.4.4.18-7-1.js
  * - test/built-ins/Array/prototype/forEach/15.4.4.18-7-2.js
  * - test/built-ins/Array/prototype/reduce/15.4.4.21-9-5.js
+ * - test/built-ins/Array/prototype/reduce/15.4.4.21-9-c-ii-20.js
  * - test/built-ins/Array/prototype/reduce/15.4.4.21-9-1.js
  * - test/built-ins/Array/prototype/reduce/15.4.4.21-10-1.js
  * - test/built-ins/Array/prototype/reduceRight/15.4.4.22-9-5.js
+ * - test/built-ins/Array/prototype/reduceRight/15.4.4.22-9-c-ii-20.js
  * - test/built-ins/Array/prototype/reduceRight/15.4.4.22-9-1.js
  * - test/built-ins/Array/prototype/reduceRight/15.4.4.22-10-1.js
  * - test/built-ins/Array/prototype/flatMap/depth-always-one.js
@@ -210,6 +212,11 @@ const cases = [
     code: `let calls = 0; const result = [1].reduce(() => { calls += 1; return 2 }); return [result, calls]`,
     expected: [1, 0],
   },
+  {
+    path: "test/built-ins/Array/prototype/reduce/15.4.4.21-9-c-ii-20.js",
+    code: `let accessed = false; const result = [11].reduce((previous) => { accessed = true; return previous === undefined }, undefined); return [result, accessed]`,
+    expected: [true, true],
+  },
   {
     path: "test/built-ins/Array/prototype/reduce/15.4.4.21-10-1.js",
     code: `const input = [1, 2, 3, 4, 5]; input.reduce(() => 1); return input`,
@@ -225,6 +232,11 @@ const cases = [
     code: `let calls = 0; const result = [1].reduceRight(() => { calls += 1; return 2 }); return [result, calls]`,
     expected: [1, 0],
   },
+  {
+    path: "test/built-ins/Array/prototype/reduceRight/15.4.4.22-9-c-ii-20.js",
+    code: `let accessed = false; const result = [11].reduceRight((previous) => { accessed = true; return previous === undefined }, undefined); return [result, accessed]`,
+    expected: [true, true],
+  },
   {
     path: "test/built-ins/Array/prototype/reduceRight/15.4.4.22-10-1.js",
     code: `const input = [1, 2, 3, 4, 5]; input.reduceRight(() => 1); return input`,
@@ -323,3 +335,46 @@ describe("Test262 Array callback adaptations", () => {
     })
   }
 })
+
+describe("Array callback regressions", () => {
+  test("reduce and reduceRight find the first present element", async () => {
+    expect(
+      await value(`
+        const left = []
+        left[2] = 3
+        const right = []
+        right[0] = 4
+        right[3] = 1
+        right.pop()
+        return [left.reduce((a, b) => a + b), right.reduceRight((a, b) => a + b)]
+      `),
+    ).toEqual([3, 4])
+  })
+
+  test("reduce and reduceRight reject arrays containing only holes", async () => {
+    expect(
+      await value(`
+        const values = []
+        values[2] = 1
+        values.pop()
+        let left
+        let right
+        try { values.reduce((a, b) => a + b) } catch (error) { left = error.name }
+        try { values.reduceRight((a, b) => a + b) } catch (error) { right = error.name }
+        return [left, right]
+      `),
+    ).toEqual(["TypeError", "TypeError"])
+  })
+
+  test("findLast returns the value observed before predicate mutation", async () => {
+    expect(
+      await value(`
+        const values = [1]
+        return values.findLast((item, index, array) => {
+          array[index] = 2
+          return true
+        })
+      `),
+    ).toBe(1)
+  })
+})

packages/codemode/test/codemode.test.ts

@@ -26,6 +26,22 @@ describe("CodeMode host failure boundary", () => {
     })
   })
 
+  test("does not rewrite explicit safe tool failures", async () => {
+    const result = await run(
+      Tool.make({
+        description: "Fail safely",
+        input: Schema.Struct({}),
+        output: Schema.String,
+        run: () => Effect.fail(toolError("File not found: /tmp/report.json")),
+      }),
+    )
+
+    expect(result.ok ? undefined : result.error).toStrictEqual({
+      kind: "ToolFailure",
+      message: "File not found: /tmp/report.json",
+    })
+  })
+
   test("sanitizes unknown host failures and defects", async () => {
     for (const failure of [
       Effect.fail(new UnsafeHostError({ reason: "Authorization: Bearer typed-secret" })),

packages/codemode/test/parity.test.ts

@@ -302,9 +302,12 @@ describe("CodeMode-specific string behavior", () => {
     expect(await value(`try { "x".normalize("nope"); return "no" } catch (e) { return e.message }`)).toContain('"NFC"')
   })
 
-  test("trimLeft/trimRight alias trimStart/trimEnd", async () => {
-    expect(await value(`return "  x ".trimLeft()`)).toBe("x ")
-    expect(await value(`return "  x ".trimRight()`)).toBe("  x")
+  test("does not expose obsolete string aliases", async () => {
+    expect(await value(`return [typeof "x".trimLeft, typeof "x".trimRight, typeof "x".substr]`)).toEqual([
+      "undefined",
+      "undefined",
+      "undefined",
+    ])
   })
 })
 

packages/codemode/test/promise.test.ts

@@ -560,7 +560,7 @@ describe("Promise.all over arbitrary arrays", () => {
     expect(trace.maxActive).toBeGreaterThan(1)
   })
 
-  test("caps live tool-call concurrency at the fixed internal constant (8)", async () => {
+  test("does not cap live tool-call concurrency", async () => {
     const trace = makeTrace()
     const result = await value(
       `
@@ -572,8 +572,7 @@ describe("Promise.all over arbitrary arrays", () => {
       { trace },
     )
     expect(result).toBe(20)
-    expect(trace.maxActive).toBeGreaterThan(1)
-    expect(trace.maxActive).toBeLessThanOrEqual(8)
+    expect(trace.maxActive).toBe(20)
   })
 
   test("resolves the empty array", async () => {

packages/codemode/test/string-search-test262.test.ts

@@ -49,12 +49,6 @@
  * - test/built-ins/String/prototype/substring/S15.5.4.15_A2_T8.js
  * - test/built-ins/String/prototype/substring/S15.5.4.15_A2_T9.js
  * - test/built-ins/String/prototype/substring/S15.5.4.15_A2_T10.js
- * - test/annexB/built-ins/String/prototype/substr/start-negative.js
- * - test/annexB/built-ins/String/prototype/substr/length-negative.js
- * - test/annexB/built-ins/String/prototype/substr/length-positive.js
- * - test/annexB/built-ins/String/prototype/substr/length-falsey.js
- * - test/annexB/built-ins/String/prototype/substr/length-undef.js
- * - test/annexB/built-ins/String/prototype/substr/surrogate-pairs.js
  * - test/built-ins/String/prototype/includes/String.prototype.includes_FailMissingLetter.js
  * - test/built-ins/String/prototype/includes/String.prototype.includes_SuccessNoLocation.js
  * - test/built-ins/String/prototype/includes/String.prototype.includes_FailBadLocation.js
@@ -460,67 +454,6 @@ const cases = [
     expected: ["this_is_"],
     labels: ['#1: __string.substring(0,8) === "this_is_"'],
   },
-  {
-    path: "test/annexB/built-ins/String/prototype/substr/start-negative.js",
-    code: `return ["abc".substr(-1), "abc".substr(-2), "abc".substr(-3), "abc".substr(-4), "abc".substr(-1.1)]`,
-    expected: ["c", "bc", "abc", "abc", "c"],
-    labels: ["-1", "-2", "-3", "size + intStart < 0", "floating point rounding semantics"],
-  },
-  {
-    path: "test/annexB/built-ins/String/prototype/substr/length-negative.js",
-    code: `return [
-      "abc".substr(0, -1), "abc".substr(0, -2), "abc".substr(0, -3), "abc".substr(0, -4),
-      "abc".substr(1, -1), "abc".substr(1, -2), "abc".substr(1, -3), "abc".substr(1, -4),
-      "abc".substr(2, -1), "abc".substr(2, -2), "abc".substr(2, -3), "abc".substr(2, -4),
-      "abc".substr(3, -1), "abc".substr(3, -2), "abc".substr(3, -3), "abc".substr(3, -4),
-    ]`,
-    expected: ["", "", "", "", "", "", "", "", "", "", "", "", "", "", "", ""],
-    labels: [
-      "0, -1", "0, -2", "0, -3", "0, -4", "1, -1", "1, -2", "1, -3", "1, -4",
-      "2, -1", "2, -2", "2, -3", "2, -4", "3, -1", "3, -2", "3, -3", "3, -4",
-    ],
-  },
-  {
-    path: "test/annexB/built-ins/String/prototype/substr/length-positive.js",
-    code: `return [
-      "abc".substr(0, 1), "abc".substr(0, 2), "abc".substr(0, 3), "abc".substr(0, 4),
-      "abc".substr(1, 1), "abc".substr(1, 2), "abc".substr(1, 3), "abc".substr(1, 4),
-      "abc".substr(2, 1), "abc".substr(2, 2), "abc".substr(2, 3), "abc".substr(2, 4),
-      "abc".substr(3, 1), "abc".substr(3, 2), "abc".substr(3, 3), "abc".substr(3, 4),
-    ]`,
-    expected: ["a", "ab", "abc", "abc", "b", "bc", "bc", "bc", "c", "c", "c", "c", "", "", "", ""],
-    labels: [
-      "0, 1", "0, 1", "0, 1", "0, 1", "1, 1", "1, 1", "1, 1", "1, 1",
-      "2, 1", "2, 1", "2, 1", "2, 1", "3, 1", "3, 1", "3, 1", "3, 1",
-    ],
-  },
-  {
-    path: "test/annexB/built-ins/String/prototype/substr/length-falsey.js",
-    code: `return ["abc".substr(0, NaN), "abc".substr(1, NaN), "abc".substr(2, NaN), "abc".substr(3, NaN)]`,
-    expected: ["", "", "", ""],
-    labels: ["start: 0, length: NaN", "start: 1, length: NaN", "start: 2, length: NaN", "start: 3, length: NaN"],
-  },
-  {
-    path: "test/annexB/built-ins/String/prototype/substr/length-undef.js",
-    code: `return [
-      "abc".substr(0), "abc".substr(1), "abc".substr(2), "abc".substr(3),
-      "abc".substr(0, undefined), "abc".substr(1, undefined), "abc".substr(2, undefined), "abc".substr(3, undefined),
-    ]`,
-    expected: ["abc", "bc", "c", "", "abc", "bc", "c", ""],
-    labels: [
-      "start: 0, length: unspecified", "start: 1, length: unspecified", "start: 2, length: unspecified", "start: 3, length: unspecified",
-      "start: 0, length: undefined", "start: 1, length: undefined", "start: 2, length: undefined", "start: 3, length: undefined",
-    ],
-  },
-  {
-    path: "test/annexB/built-ins/String/prototype/substr/surrogate-pairs.js",
-    code: `return [
-      "\uD834\uDF06".substr(0), "\uD834\uDF06".substr(1), "\uD834\uDF06".substr(2),
-      "\uD834\uDF06".substr(0, 0), "\uD834\uDF06".substr(0, 1), "\uD834\uDF06".substr(0, 2),
-    ]`,
-    expected: ["\uD834\uDF06", "\uDF06", "", "", "\uD834", "\uD834\uDF06"],
-    labels: ["start: 0", "start: 1", "start: 2", "end: 0", "end: 1", "end: 2"],
-  },
   {
     path: "test/built-ins/String/prototype/includes/String.prototype.includes_FailMissingLetter.js",
     code: `return ["word".includes("a", 0)]`, expected: [false], labels: ['"word".includes("a", 0)'],

packages/codemode/tsconfig.json

@@ -2,6 +2,7 @@
   "$schema": "https://json.schemastore.org/tsconfig",
   "extends": "@tsconfig/bun/tsconfig.json",
   "compilerOptions": {
-    "noUncheckedIndexedAccess": false
+    "noUncheckedIndexedAccess": false,
+    "noUnusedLocals": true
   }
 }

packages/core/src/aisdk.ts

@@ -15,20 +15,43 @@ import type {
   SharedV3ProviderOptions,
 } from "@ai-sdk/provider"
 import {
+  APIError,
+  Authentication,
+  BadRequest,
+  ConnectionError,
   FinishReason,
-  InvalidProviderOutputReason,
+  HttpContext,
+  HttpRequestDetails,
+  HttpResponseDetails,
   LLMEvent,
-  LLMError,
+  MalformedResponse,
   Model,
+  NotFound,
   ProviderID,
   ProviderMetadata,
   ToolResultValue,
-  UnknownProviderReason,
+  classifyApiFailure,
+  isLLMError,
+  type LLMError,
   type ContentPart,
   type LLMRequest,
   type ToolDefinition,
   type UsageInput,
 } from "@opencode-ai/llm"
+import {
+  APICallError,
+  EmptyResponseBodyError,
+  InvalidArgumentError,
+  InvalidPromptError,
+  InvalidResponseDataError,
+  JSONParseError,
+  LoadAPIKeyError,
+  LoadSettingError,
+  NoContentGeneratedError,
+  NoSuchModelError,
+  TypeValidationError,
+  UnsupportedFunctionalityError,
+} from "@ai-sdk/provider"
 import { Auth, Endpoint, type AnyRoute } from "@opencode-ai/llm/route"
 import { Cause, Context, Effect, Layer, Option, Schema, Scope, Stream } from "effect"
 import { ModelV2 } from "./model"
@@ -490,12 +513,12 @@ function streamLanguage(language: LanguageModelV3, options: LanguageModelV3CallO
     Stream.unwrap(
       Effect.tryPromise({
         try: () => language.doStream(options),
-        catch: (error) => llmError("doStream", error),
+        catch: (error) => llmError(error),
       }).pipe(
         Effect.map((result) =>
           Stream.fromReadableStream({
             evaluate: () => result.stream,
-            onError: (error) => llmError("readStream", error),
+            onError: (error) => llmError(error),
           }).pipe(
             Stream.mapEffect((event) => streamPartEvents(state, event)),
             Stream.flatMap((events) => Stream.fromIterable(events)),
@@ -608,7 +631,7 @@ function streamPartEvents(
         }),
       ])
     case "error":
-      return Effect.fail(llmError("stream", event.error))
+      return Effect.fail(llmError(event.error))
   }
 }
 
@@ -666,16 +689,65 @@ function messageValue(input: unknown) {
   }
 }
 
-function llmError(method: string, error: unknown) {
-  const reason =
-    error instanceof LLMError
-      ? new InvalidProviderOutputReason({ message: error.message })
-      : new UnknownProviderReason({ message: error instanceof Error ? error.message : String(error) })
-  return new LLMError({
-    module: "AISDK",
-    method,
-    reason,
-  })
+const BODY_LIMIT = 16_384
+
+const headerRetryAfterMs = (headers: Record<string, string> | undefined) => {
+  if (!headers) return undefined
+  const millis = Number(headers["retry-after-ms"])
+  if (Number.isFinite(millis)) return Math.max(0, millis)
+  const value = headers["retry-after"]
+  if (!value) return undefined
+  const seconds = Number(value)
+  if (Number.isFinite(seconds)) return Math.max(0, seconds * 1000)
+  const date = Date.parse(value)
+  if (!Number.isNaN(date)) return Math.max(0, date - Date.now())
+  return undefined
+}
+
+// Classify AI SDK failures into the shared `LLMError` union so the synthetic
+// AI SDK route reports failures identically to native protocol routes. An
+// `APICallError` without a status code is the AI SDK's representation of a
+// network-level failure (connect refused, reset, DNS), not an API rejection.
+function llmError(error: unknown): LLMError {
+  if (isLLMError(error)) return error
+  if (APICallError.isInstance(error)) {
+    if (error.statusCode === undefined) {
+      return new ConnectionError({ message: error.message, url: error.url, cause: error })
+    }
+    return classifyApiFailure({
+      message: error.message,
+      status: error.statusCode,
+      retryAfterMs: headerRetryAfterMs(error.responseHeaders),
+      requestID: error.responseHeaders?.["x-request-id"] ?? error.responseHeaders?.["request-id"],
+      http: new HttpContext({
+        request: new HttpRequestDetails({ method: "POST", url: error.url, headers: {} }),
+        response: new HttpResponseDetails({ status: error.statusCode, headers: error.responseHeaders ?? {} }),
+        body: error.responseBody === undefined ? undefined : error.responseBody.slice(0, BODY_LIMIT),
+        bodyTruncated: error.responseBody !== undefined && error.responseBody.length > BODY_LIMIT ? true : undefined,
+      }),
+    })
+  }
+  if (LoadAPIKeyError.isInstance(error) || LoadSettingError.isInstance(error)) {
+    return new Authentication({ message: error.message })
+  }
+  if (NoSuchModelError.isInstance(error)) return new NotFound({ message: error.message })
+  if (
+    InvalidPromptError.isInstance(error) ||
+    InvalidArgumentError.isInstance(error) ||
+    UnsupportedFunctionalityError.isInstance(error)
+  ) {
+    return new BadRequest({ message: error.message })
+  }
+  if (
+    InvalidResponseDataError.isInstance(error) ||
+    JSONParseError.isInstance(error) ||
+    TypeValidationError.isInstance(error) ||
+    EmptyResponseBodyError.isInstance(error) ||
+    NoContentGeneratedError.isInstance(error)
+  ) {
+    return new MalformedResponse({ message: error.message })
+  }
+  return new APIError({ message: error instanceof Error ? error.message : String(error) })
 }
 
 export const node = makeLocationNode({ service: Service, layer: locationLayer, deps: [] })

packages/core/src/config.ts

@@ -200,7 +200,6 @@ const layer = Layer.effect(
             .up({
               targets: [".opencode", ".claude", ".agents", ...names.toReversed()],
               start: location.directory,
-              stop: location.project.directory,
             })
           .pipe(Effect.orDie)
 

packages/core/src/flag/flag.ts

@@ -52,9 +52,6 @@ export const Flag = {
   get OPENCODE_EXPERIMENTAL_REFERENCES() {
     return enabledByExperimental("OPENCODE_EXPERIMENTAL_REFERENCES")
   },
-  get CODEMODE_ENABLED() {
-    return process.env["CODEMODE_ENABLED"] === undefined || truthy("CODEMODE_ENABLED")
-  },
   get OPENCODE_TUI_CONFIG() {
     return process.env["OPENCODE_TUI_CONFIG"]
   },

packages/core/src/generate.ts

@@ -1,6 +1,6 @@
 export * as Generate from "./generate"
 
-import { LLM, LLMClient, LLMError } from "@opencode-ai/llm"
+import { LLM, LLMClient, type LLMError } from "@opencode-ai/llm"
 import { Context, Effect, Layer, Schema } from "effect"
 import { Catalog } from "./catalog"
 import { makeLocationNode } from "./effect/app-node"

packages/core/src/mcp/guidance.ts

@@ -1,7 +1,6 @@
 export * as McpGuidance from "./guidance"
 
 import { makeLocationNode } from "../effect/app-node"
-import { Flag } from "../flag/flag"
 import { Context, Effect, Layer, Schema } from "effect"
 import { AgentV2 } from "../agent"
 import { PermissionV2 } from "../permission"
@@ -18,11 +17,7 @@ type Summary = typeof Summary.Type
 const entries = (servers: ReadonlyArray<Summary>) =>
   servers.flatMap((server) => [
     `  <server name="${server.server}">`,
-    ...(Flag.CODEMODE_ENABLED
-      ? [
-          `    Use tools from this server through \`execute\` under \`tools[${JSON.stringify(McpTool.group(server.server))}]\`.`,
-        ]
-      : []),
+    `    Use tools from this server through \`execute\` under \`tools[${JSON.stringify(McpTool.group(server.server))}]\`.`,
     ...server.instructions.split("\n").map((line) => `    ${line}`),
     "  </server>",
   ])
@@ -81,7 +76,7 @@ export const layer = Layer.effect(
               removed: () => "MCP server instructions are no longer available.",
             },
           })
-        if (Flag.CODEMODE_ENABLED && PermissionV2.evaluate("execute", "*", agent.permissions).effect === "deny")
+        if (PermissionV2.evaluate("execute", "*", agent.permissions).effect === "deny")
           return source(Instructions.removed)
         const [instructions, tools] = yield* Effect.all([mcp.instructions(), mcp.tools()], {
           concurrency: "unbounded",
@@ -90,12 +85,9 @@ export const layer = Layer.effect(
         const visible = instructions
           .filter((item) => {
             const owned = tools.filter((tool) => tool.server === item.server)
-            return (
-              (!Flag.CODEMODE_ENABLED && owned.length === 0) ||
-              owned.some(
-                (tool) =>
-                  PermissionV2.evaluate(McpTool.name(tool.server, tool.name), "*", agent.permissions).effect !== "deny",
-              )
+            return owned.some(
+              (tool) =>
+                PermissionV2.evaluate(McpTool.name(tool.server, tool.name), "*", agent.permissions).effect !== "deny",
             )
           })
           .map((item) => ({ server: item.server, instructions: item.instructions }))

packages/core/src/session/compaction.ts

@@ -1,6 +1,6 @@
 export * as SessionCompaction from "./compaction"
 
-import { LLM, LLMClient, LLMError, LLMEvent, Message, type LLMRequest, type Model } from "@opencode-ai/llm"
+import { LLM, LLMClient, LLMEvent, Message, isLLMError, type LLMError, type LLMRequest, type Model } from "@opencode-ai/llm"
 import { SessionError } from "@opencode-ai/schema/session-error"
 import { Context, Effect, Layer, Stream } from "effect"
 import { Config } from "../config"
@@ -247,11 +247,6 @@ const make = (dependencies: Dependencies) => {
       )
       .pipe(
         Stream.runForEach((event) => {
-          if (LLMEvent.is.providerError(event))
-            failure = {
-              type: event.classification === "context-overflow" ? "provider.invalid-request" : "provider.error",
-              message: event.message,
-            }
           if (LLMEvent.is.textDelta(event)) {
             chunks.push(event.text)
             return dependencies.events.publish(SessionEvent.Compaction.Delta, {
@@ -261,7 +256,7 @@ const make = (dependencies: Dependencies) => {
           }
           return Effect.void
         }),
-        Effect.catchTag("LLM.Error", (error) =>
+        Effect.catchIf(isLLMError, (error) =>
           Effect.sync(() => {
             failure = toSessionError(error)
           }),

packages/core/src/session/runner/llm.ts

@@ -1,15 +1,6 @@
 export * as SessionRunnerLLM from "./llm"
 
-import {
-  LLM,
-  LLMClient,
-  LLMError,
-  LLMEvent,
-  Message,
-  SystemPart,
-  isContextOverflowFailure,
-  type ProviderErrorEvent,
-} from "@opencode-ai/llm"
+import { LLM, LLMClient, LLMEvent, Message, SystemPart, isLLMError, type LLMError } from "@opencode-ai/llm"
 import { SessionError } from "@opencode-ai/schema/session-error"
 import { Money } from "@opencode-ai/schema/money"
 import { Cause, Effect, Exit, Fiber, FiberSet, Layer, Option, Semaphore, Stream } from "effect"
@@ -227,17 +218,10 @@ const layer = Layer.effect(
       // mid-event.
       const serialized = <A, E, R>(effect: Effect.Effect<A, E, R>) => publication.withPermit(effect)
       const publish = (event: LLMEvent, error?: SessionError.Error) => serialized(publisher.publish(event, error))
-      let overflowFailure: ProviderErrorEvent | undefined
       const providerStream = llm.stream(request).pipe(
         Stream.runForEach((event) =>
           Effect.gen(function* () {
-            if (overflowFailure || publisher.hasProviderError()) return
-            if (LLMEvent.is.providerError(event)) {
-              if (isContextOverflowFailure(event) && !publisher.hasRetryEvidence()) {
-                overflowFailure = event
-                return
-              }
-            }
+            if (publisher.hasProviderError()) return
             yield* publish(event)
             if (event.type !== "tool-call" || event.providerExecuted) return
             if (!toolMaterialization) {
@@ -317,22 +301,21 @@ const layer = Layer.effect(
           // away non-interrupt failures, so both interrupt checks stay Cause-based.
           const streamInterrupted = stream._tag === "Failure" && Cause.hasInterrupts(stream.cause)
 
+          const llmFailure = streamFailure !== undefined && isLLMError(streamFailure) ? streamFailure : undefined
+
           // A context overflow before any assistant output is recoverable: compact and
           // restart the step instead of surfacing the provider error.
           if (
             recoverOverflow &&
             !publisher.hasRetryEvidence() &&
-            isContextOverflowFailure(overflowFailure ?? streamFailure) &&
+            llmFailure?._tag === "LLM.ContextOverflow" &&
             (yield* restore(recoverOverflow({ sessionID: session.id, messages: context, model }))).status ===
               "completed"
           )
             return { _tag: "RestartAfterOverflowCompaction", step: currentStep } as const
 
-          // An unrecovered held-back overflow becomes the step's durable provider error. A
-          // thrown LLM failure records the assistant failure unless a provider error was
-          // already recorded from the stream. Terminal publication waits for owned tools.
-          if (overflowFailure) yield* publish(overflowFailure)
-          const llmFailure = streamFailure instanceof LLMError ? streamFailure : undefined
+          // A thrown LLM failure records the assistant failure unless a provider failure
+          // was already recorded from the stream. Terminal publication waits for owned tools.
           if (llmFailure && !publisher.hasProviderError()) {
             const error = toSessionError(llmFailure)
             if (
@@ -349,7 +332,8 @@ const layer = Layer.effect(
             }
             yield* serialized(publisher.failAssistant(error))
           }
-          // Provider error events only arrive from the stream, so the flag is final here.
+          // The provider-failed flag is only set while consuming the stream (content-filter
+          // step finish), so it is final here.
           const providerFailed = publisher.hasProviderError()
 
           // Settle every owned tool fiber. FiberSet.join returns on the first failure, so retain
@@ -543,7 +527,8 @@ const layer = Layer.effect(
           needsContinuation = result.needsContinuation
           step = result.step + 1
           if (needsContinuation) {
-            promotion = (yield* SessionPending.compaction(db, input.sessionID)) ? undefined : "steer"
+            yield* runPendingCompaction(input.sessionID)
+            promotion = "steer"
             continue
           }
           yield* runPendingCompaction(input.sessionID)

packages/core/src/session/runner/publish-llm-event.ts

@@ -438,10 +438,6 @@ export const createLLMEventPublisher = (events: Pick<EventV2.Interface, "publish
         return
       case "finish":
         return
-      case "provider-error":
-        providerFailed = true
-        yield* failAssistant({ type: "provider.unknown", message: event.message }, true)
-        return
     }
   })
 

packages/core/src/session/runner/retry.ts

@@ -1,6 +1,6 @@
 export * as SessionRunnerRetry from "./retry"
 
-import { LLMError } from "@opencode-ai/llm"
+import type { LLMError } from "@opencode-ai/llm"
 import { SessionError } from "@opencode-ai/schema/session-error"
 import { Data, Duration, Effect, Schedule } from "effect"
 import { EventV2 } from "../../event"
@@ -17,29 +17,33 @@ export class RetryableFailure extends Data.TaggedError("SessionRunner.RetryableF
 }> {}
 
 export function isRetryable(error: LLMError) {
-  switch (error.reason._tag) {
-    case "RateLimit":
-    case "ProviderInternal":
-    case "Transport":
+  switch (error._tag) {
+    case "LLM.RateLimit":
+    case "LLM.ServerError":
+    case "LLM.ConnectionError":
+    case "LLM.TimeoutError":
       return true
-    case "Authentication":
-    case "QuotaExceeded":
-    case "ContentPolicy":
-    case "InvalidProviderOutput":
-    case "InvalidRequest":
-    case "NoRoute":
-    case "UnknownProvider":
+    case "LLM.Authentication":
+    case "LLM.PermissionDenied":
+    case "LLM.NotFound":
+    case "LLM.QuotaExceeded":
+    case "LLM.ContentPolicy":
+    case "LLM.ContextOverflow":
+    case "LLM.MalformedResponse":
+    case "LLM.BadRequest":
+    case "LLM.NoRoute":
+    case "LLM.APIError":
       return false
     default: {
-      const exhaustive: never = error.reason
+      const exhaustive: never = error
       return exhaustive
     }
   }
 }
 
 const retryAfter = (failure: RetryableFailure) => {
-  if (failure.cause.reason._tag === "RateLimit" || failure.cause.reason._tag === "ProviderInternal")
-    return failure.cause.reason.retryAfterMs
+  if (failure.cause._tag === "LLM.RateLimit" || failure.cause._tag === "LLM.ServerError")
+    return failure.cause.retryAfterMs
   return undefined
 }
 

packages/core/src/session/title.ts

@@ -1,6 +1,6 @@
 export * as SessionTitle from "./title"
 
-import { LLM, LLMClient, LLMError, LLMEvent, Message, type LLMRequest } from "@opencode-ai/llm"
+import { LLM, LLMClient, LLMEvent, Message, isLLMError, type LLMError, type LLMRequest } from "@opencode-ai/llm"
 import { Context, DateTime, Effect, Layer, Stream } from "effect"
 import { AgentV2 } from "../agent"
 import { Database } from "../database/database"
@@ -49,7 +49,6 @@ const make = (dependencies: Dependencies) => {
     ).pipe(Effect.catch(() => Effect.succeed(undefined)))
     if (!resolved) return
     const chunks: string[] = []
-    let failed = false
     const streamed = yield* dependencies.llm
       .stream(
         LLM.request({
@@ -61,14 +60,13 @@ const make = (dependencies: Dependencies) => {
       )
       .pipe(
         Stream.runForEach((event) => {
-          if (LLMEvent.is.providerError(event)) failed = true
           if (LLMEvent.is.textDelta(event)) chunks.push(event.text)
           return Effect.void
         }),
         Effect.as(true),
-        Effect.catchTag("LLM.Error", () => Effect.succeed(false)),
+        Effect.catchIf(isLLMError, () => Effect.succeed(false)),
       )
-    if (!streamed || failed) return
+    if (!streamed) return
     const title = chunks
       .join("")
       .split("\n")

packages/core/src/session/to-session-error.ts

@@ -1,4 +1,4 @@
-import { LLMError, ToolFailure } from "@opencode-ai/llm"
+import { isLLMError, ToolFailure } from "@opencode-ai/llm"
 import { Tool } from "@opencode-ai/plugin/v2/effect/tool"
 import { SessionError } from "@opencode-ai/schema/session-error"
 import { PermissionV2 } from "../permission"
@@ -9,30 +9,38 @@ import { AgentNotFoundError, StepFailedError, UserInterruptedError } from "./err
 import { SessionRunnerModel } from "./runner/model"
 
 export function toSessionError(cause: unknown): SessionError.Error {
-  if (cause instanceof LLMError) {
-    switch (cause.reason._tag) {
-      case "RateLimit":
-        return { type: "provider.rate-limit", message: cause.reason.message }
-      case "Authentication":
-        return { type: "provider.auth", message: cause.reason.message }
-      case "QuotaExceeded":
-        return { type: "provider.quota", message: cause.reason.message }
-      case "ContentPolicy":
-        return { type: "provider.content-filter", message: cause.reason.message }
-      case "Transport":
-        return { type: "provider.transport", message: cause.reason.message }
-      case "ProviderInternal":
-        return { type: "provider.internal", message: cause.reason.message }
-      case "InvalidProviderOutput":
-        return { type: "provider.invalid-output", message: cause.reason.message }
-      case "InvalidRequest":
-        return { type: "provider.invalid-request", message: cause.reason.message }
-      case "NoRoute":
-        return { type: "provider.no-route", message: cause.reason.message }
-      case "UnknownProvider":
-        return { type: "provider.unknown", message: cause.reason.message }
+  if (isLLMError(cause)) {
+    switch (cause._tag) {
+      case "LLM.RateLimit":
+        return { type: "provider.rate-limit", message: cause.message }
+      case "LLM.Authentication":
+        return { type: "provider.auth", message: cause.message }
+      case "LLM.PermissionDenied":
+        return { type: "provider.auth", message: cause.message }
+      case "LLM.NotFound":
+        return { type: "provider.not-found", message: cause.message }
+      case "LLM.QuotaExceeded":
+        return { type: "provider.quota", message: cause.message }
+      case "LLM.ContentPolicy":
+        return { type: "provider.content-filter", message: cause.message }
+      case "LLM.ContextOverflow":
+        return { type: "provider.context-overflow", message: cause.message }
+      case "LLM.ConnectionError":
+        return { type: "provider.transport", message: cause.message }
+      case "LLM.TimeoutError":
+        return { type: "provider.timeout", message: cause.message }
+      case "LLM.ServerError":
+        return { type: "provider.internal", message: cause.message }
+      case "LLM.MalformedResponse":
+        return { type: "provider.invalid-output", message: cause.message }
+      case "LLM.BadRequest":
+        return { type: "provider.invalid-request", message: cause.message }
+      case "LLM.NoRoute":
+        return { type: "provider.no-route", message: cause.message }
+      case "LLM.APIError":
+        return { type: "provider.unknown", message: cause.message }
       default: {
-        const exhaustive: never = cause.reason
+        const exhaustive: never = cause
         return exhaustive
       }
     }

packages/core/src/tool/AGENTS.md

@@ -29,7 +29,8 @@ Leaves own resolution, permission, and side-effect ordering. Translate only expe
 ## Registration
 
 Built-ins and plugin tools register through `Tools.Service.register({ [name]: tool })`. Registrations may provide a
-group, which flattens direct model names to `<group>_<tool>`, and may be deferred from direct model exposure.
+group, which flattens direct model names to `<group>_<tool>`, and default into CodeMode (`codemode` defaults true;
+`codemode: false` keeps the tool on the provider's native tool list).
 
 Registrations are scoped:
 

packages/core/src/tool/edit.ts

@@ -212,6 +212,7 @@ export const Plugin = {
             }),
             "edit",
           ),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/execute.ts

@@ -58,16 +58,16 @@ export const create = (registrations: ReadonlyMap<string, Registration>) => {
       })
       if (registration.group === undefined) {
         const path = registration.name
-        if (Object.hasOwn(tools, path)) throw new TypeError(`Deferred tool namespace conflict: ${path}`)
+        if (Object.hasOwn(tools, path)) throw new TypeError(`CodeMode tool namespace conflict: ${path}`)
         tools[path] = value
         continue
       }
       const path = registration.name
       const namespace = registration.group
       const group = tools[namespace]
-      if (group && Tool.isDefinition(group)) throw new TypeError(`Deferred tool namespace conflict: ${namespace}`)
+      if (group && Tool.isDefinition(group)) throw new TypeError(`CodeMode tool namespace conflict: ${namespace}`)
       if (group) {
-        if (Object.hasOwn(group, path)) throw new TypeError(`Deferred tool namespace conflict: ${namespace}.${path}`)
+        if (Object.hasOwn(group, path)) throw new TypeError(`CodeMode tool namespace conflict: ${namespace}.${path}`)
         group[path] = value
         continue
       }

packages/core/src/tool/glob.ts

@@ -106,6 +106,7 @@ export const Plugin = {
                 ),
               ),
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/grep.ts

@@ -137,6 +137,7 @@ export const Plugin = {
                 ),
               ),
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/mcp.ts

@@ -5,7 +5,7 @@ import { McpEvent } from "@opencode-ai/schema/mcp-event"
 import { Effect, Exit, type JsonSchema, Layer, Scope, Semaphore, Stream } from "effect"
 import { makeLocationNode } from "../effect/app-node"
 import { EventV2 } from "../event"
-import { Flag } from "../flag/flag"
+
 import { MCP } from "../mcp"
 import { PermissionV2 } from "../permission"
 import { Tool } from "./tool"
@@ -107,7 +107,7 @@ export const layer = Layer.effectDiscard(
         const next = yield* Scope.fork(scope)
         yield* Effect.forEach(
           groups,
-          ([group, record]) => tools.register(record, { group, deferred: Flag.CODEMODE_ENABLED }),
+          ([group, record]) => tools.register(record, { group }),
           {
             discard: true,
           },

packages/core/src/tool/patch.ts

@@ -191,6 +191,7 @@ export const Plugin = {
             }),
             "edit",
           ),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/question.ts

@@ -106,6 +106,7 @@ export const Plugin = {
                   }),
                 ),
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/read.ts

@@ -139,6 +139,7 @@ export const Plugin = {
               )
             },
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/registry.ts

@@ -3,7 +3,6 @@ export * as ToolRegistry from "./registry"
 import { ToolOutput, type ToolCall, type ToolDefinition, type ToolResultValue } from "@opencode-ai/llm"
 import { Context, Effect, Layer, Scope } from "effect"
 import type { AgentV2 } from "../agent"
-import { Flag } from "../flag/flag"
 import { PermissionV2 } from "../permission"
 import { SessionMessage } from "../session/message"
 import { SessionSchema } from "../session/schema"
@@ -56,7 +55,7 @@ const registryLayer = Layer.effect(
       readonly tool: AnyTool
       readonly name: string
       readonly group?: string
-      readonly deferred: boolean
+      readonly codemode: boolean
     }
     const local = new Map<string, Array<{ readonly token: object; readonly registration: Registration }>>()
 
@@ -132,7 +131,8 @@ const registryLayer = Layer.effect(
       register: Effect.fn("ToolRegistry.register")(function* (tools, options) {
         const entries = registrationEntries(tools, options?.group)
         if (entries.length === 0) return
-        const reserved = options?.deferred ? undefined : entries.find((entry) => entry.key === "execute")
+        const codemode = options?.codemode ?? true
+        const reserved = codemode ? undefined : entries.find((entry) => entry.key === "execute")
         if (reserved)
           return yield* Effect.fail(
             new RegistrationError({ name: reserved.key, message: 'Tool name "execute" is reserved for CodeMode' }),
@@ -149,7 +149,7 @@ const registryLayer = Layer.effect(
                     tool: entry.tool,
                     name: entry.name,
                     group: entry.group,
-                    deferred: options?.deferred ?? false,
+                    codemode,
                   },
                 },
               ])
@@ -168,18 +168,17 @@ const registryLayer = Layer.effect(
       }),
       materialize: Effect.fn("ToolRegistry.materialize")(function* (permissions) {
         const direct = new Map<string, Registration>()
-        const deferred = new Map<string, Registration>()
+        const codemode = new Map<string, Registration>()
         const rules = permissions ?? []
         for (const [name, entries] of local) {
           const registration = entries.at(-1)?.registration
           if (!registration) continue
-          if (registration.deferred && !Flag.CODEMODE_ENABLED) continue
           if (whollyDisabled(permission(registration.tool, name), rules)) continue
-          if (registration.deferred) deferred.set(name, registration)
+          if (registration.codemode) codemode.set(name, registration)
           else direct.set(name, registration)
         }
         const execute =
-          deferred.size > 0 && !whollyDisabled("execute", rules) ? ExecuteTool.create(deferred) : undefined
+          codemode.size > 0 && !whollyDisabled("execute", rules) ? ExecuteTool.create(codemode) : undefined
         return {
           definitions: [
             ...Array.from(direct, ([name, registration]) => definition(name, registration.tool)),

packages/core/src/tool/shell.ts

@@ -132,6 +132,8 @@ export const Plugin = {
           return runtime.session.synthetic({
             sessionID,
             text: `<shell id="${callID}" state="${state}" command="${command}">\n${text}\n</shell>`,
+            description: command,
+            metadata: { source: "shell", state },
           })
         }),
         Effect.forkIn(scope, { startImmediately: true }),
@@ -276,6 +278,7 @@ export const Plugin = {
                 ),
               ),
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/skill.ts

@@ -97,6 +97,7 @@ export const Plugin = {
                 }).pipe(Effect.mapError((error) => unableToLoad(input.id, error)))
               }),
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/subagent.ts

@@ -208,6 +208,7 @@ export const Plugin = {
                 return { sessionID: child.id, status: "completed" as const, output: result?.info.output ?? NO_TEXT }
               }),
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/webfetch.ts

@@ -174,6 +174,7 @@ export const Plugin = {
                 }
               }).pipe(Effect.mapError((error) => new ToolFailure({ message: `Unable to fetch ${input.url}`, error }))),
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/websearch.ts

@@ -252,6 +252,7 @@ export const Plugin = {
               )
             },
           }),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/src/tool/write.ts

@@ -91,6 +91,7 @@ export const Plugin = {
             }),
             "edit",
           ),
+          { codemode: false },
         ),
       )
       .pipe(Effect.orDie)

packages/core/test/config/config.test.ts

@@ -876,7 +876,7 @@ describe("Config", () => {
     ),
   )
 
-  it.live("loads global, ancestor, and .opencode configuration up to the project boundary", () =>
+  it.live("loads global and ancestor configuration across the project boundary", () =>
     Effect.acquireRelease(
       Effect.promise(() => tmpdir()),
       (tmp) => Effect.promise(() => tmp[Symbol.asyncDispose]()),
@@ -936,6 +936,7 @@ describe("Config", () => {
             ])
             expect(documents.map((document) => document.info.$schema)).toEqual([
               "global",
+              "outside",
               "root",
               "parent",
               "directory",
@@ -951,6 +952,8 @@ describe("Config", () => {
               AbsolutePath.make(path.join(root, ".agents")),
               "global",
               AbsolutePath.make(global),
+              "outside",
+              AbsolutePath.make(path.join(tmp.path, "opencode.json")),
               "root",
               AbsolutePath.make(path.join(root, "opencode.json")),
               "parent",

packages/core/test/plugin.test.ts

@@ -260,6 +260,7 @@ describe("PluginV2", () => {
                   output: Schema.Struct({ ok: Schema.Boolean }),
                   execute: () => Effect.succeed({ ok: true }),
                 }),
+                { codemode: false },
               ),
             )
             .pipe(Effect.orDie),
@@ -273,7 +274,7 @@ describe("PluginV2", () => {
     }),
   )
 
-  it.effect("groups tool names and defers registrations from direct exposure", () =>
+  it.effect("groups tool names and routes codemode registrations through execute", () =>
     Effect.gen(function* () {
       const plugins = yield* PluginV2.Service
       const registry = yield* ToolRegistry.Service
@@ -289,9 +290,9 @@ describe("PluginV2", () => {
         effect: (ctx) =>
           ctx.tool
             .transform((draft) => {
-              draft.add("plain", tool("Plain"))
-              draft.add("look/up", tool("Lookup"), { group: "context 7" })
-              draft.add("search", tool("Search"), { group: "context 7", deferred: true })
+              draft.add("plain", tool("Plain"), { codemode: false })
+              draft.add("look/up", tool("Lookup"), { group: "context 7", codemode: false })
+              draft.add("search", tool("Search"), { group: "context 7" })
             })
             .pipe(Effect.orDie),
       })
@@ -330,6 +331,7 @@ describe("PluginV2", () => {
                     output: Schema.Struct({ text: Schema.String }),
                     execute: ({ text }) => Effect.sync(() => executed.push({ text })).pipe(Effect.as({ text })),
                   }),
+                  { codemode: false },
                 ),
               )
               .pipe(Effect.orDie)

packages/core/test/plugin/promise.test.ts

@@ -135,6 +135,7 @@ describe("fromPromise", () => {
           await ctx.tool.transform((tools) => {
             tools.add({
               name: "hello",
+              options: { codemode: false },
               description: "Hello",
               input: Schema.Struct({ name: Schema.String }),
               output: Schema.String,

packages/core/test/session-error.test.ts

@@ -1,18 +1,22 @@
 import { describe, expect, test } from "bun:test"
 import {
-  AuthenticationReason,
-  ContentPolicyReason,
-  InvalidProviderOutputReason,
-  InvalidRequestReason,
-  LLMError,
-  NoRouteReason,
+  APIError,
+  Authentication,
+  BadRequest,
+  ConnectionError,
+  ContentPolicy,
+  ContextOverflow,
+  MalformedResponse,
   ModelID,
+  NoRoute,
+  NotFound,
+  PermissionDenied,
   ProviderID,
-  ProviderInternalReason,
-  QuotaExceededReason,
-  RateLimitReason,
-  TransportReason,
-  UnknownProviderReason,
+  QuotaExceeded,
+  RateLimit,
+  RouteID,
+  ServerError,
+  TimeoutError,
   ToolFailure,
 } from "@opencode-ai/llm"
 import { PermissionV2 } from "@opencode-ai/core/permission"
@@ -20,39 +24,33 @@ import { Tool } from "@opencode-ai/plugin/v2/effect/tool"
 import { toSessionError } from "@opencode-ai/core/session/to-session-error"
 import { SessionRunnerRetry } from "@opencode-ai/core/session/runner/retry"
 
-const llm = (reason: LLMError["reason"]) => new LLMError({ module: "test", method: "stream", reason })
-
 describe("toSessionError", () => {
-  test("maps every LLM reason to the open wire type", () => {
-    expect(toSessionError(llm(new RateLimitReason({ message: "rate", retryAfterMs: 123 })))).toEqual({
+  test("maps every LLM error tag to the open wire type", () => {
+    expect(toSessionError(new RateLimit({ message: "rate", retryAfterMs: 123 }))).toEqual({
       type: "provider.rate-limit",
       message: "rate",
     })
-    expect(toSessionError(llm(new AuthenticationReason({ message: "auth", kind: "invalid" }))).type).toBe(
-      "provider.auth",
-    )
-    expect(toSessionError(llm(new QuotaExceededReason({ message: "quota" }))).type).toBe("provider.quota")
-    expect(toSessionError(llm(new ContentPolicyReason({ message: "blocked" }))).type).toBe("provider.content-filter")
-    expect(toSessionError(llm(new TransportReason({ message: "transport" }))).type).toBe("provider.transport")
-    expect(toSessionError(llm(new ProviderInternalReason({ message: "internal", status: 500 }))).type).toBe(
-      "provider.internal",
-    )
-    expect(toSessionError(llm(new InvalidProviderOutputReason({ message: "output" }))).type).toBe(
-      "provider.invalid-output",
-    )
-    expect(toSessionError(llm(new InvalidRequestReason({ message: "request" }))).type).toBe("provider.invalid-request")
+    expect(toSessionError(new Authentication({ message: "auth" })).type).toBe("provider.auth")
+    expect(toSessionError(new PermissionDenied({ message: "forbidden" })).type).toBe("provider.auth")
+    expect(toSessionError(new NotFound({ message: "missing" })).type).toBe("provider.not-found")
+    expect(toSessionError(new QuotaExceeded({ message: "quota" })).type).toBe("provider.quota")
+    expect(toSessionError(new ContentPolicy({ message: "blocked" })).type).toBe("provider.content-filter")
+    expect(toSessionError(new ContextOverflow({ message: "too long" })).type).toBe("provider.context-overflow")
+    expect(toSessionError(new ConnectionError({ message: "reset" })).type).toBe("provider.transport")
+    expect(toSessionError(new TimeoutError({ message: "timed out" })).type).toBe("provider.timeout")
+    expect(toSessionError(new ServerError({ message: "internal", status: 500 })).type).toBe("provider.internal")
+    expect(toSessionError(new MalformedResponse({ message: "output" })).type).toBe("provider.invalid-output")
+    expect(toSessionError(new BadRequest({ message: "request" })).type).toBe("provider.invalid-request")
     expect(
       toSessionError(
-        llm(
-          new NoRouteReason({
-            route: "route",
-            provider: ProviderID.make("provider"),
-            model: ModelID.make("model"),
-          }),
-        ),
+        new NoRoute({
+          route: RouteID.make("route"),
+          provider: ProviderID.make("provider"),
+          model: ModelID.make("model"),
+        }),
       ).type,
     ).toBe("provider.no-route")
-    expect(toSessionError(llm(new UnknownProviderReason({ message: "unknown" }))).type).toBe("provider.unknown")
+    expect(toSessionError(new APIError({ message: "unknown", status: 418 })).type).toBe("provider.unknown")
   })
 
   test("preserves the permission rejection type without exposing internal fields", () => {
@@ -71,23 +69,31 @@ describe("toSessionError", () => {
     })
   })
 
-  test("retries only rate limits, provider-internal failures, and transport failures", () => {
+  test("retries only rate limits, server errors, connection failures, and timeouts", () => {
     const eligible = [
-      llm(new RateLimitReason({ message: "rate" })),
-      llm(new ProviderInternalReason({ message: "internal", status: 500 })),
-      llm(new TransportReason({ message: "transport" })),
+      new RateLimit({ message: "rate" }),
+      new ServerError({ message: "internal", status: 500 }),
+      new ConnectionError({ message: "reset" }),
+      new TimeoutError({ message: "timed out" }),
     ]
     const ineligible = [
-      llm(new AuthenticationReason({ message: "auth", kind: "invalid" })),
-      llm(new QuotaExceededReason({ message: "quota" })),
-      llm(new ContentPolicyReason({ message: "blocked" })),
-      llm(new InvalidProviderOutputReason({ message: "output" })),
-      llm(new InvalidRequestReason({ message: "request" })),
-      llm(new NoRouteReason({ route: "route", provider: ProviderID.make("provider"), model: ModelID.make("model") })),
-      llm(new UnknownProviderReason({ message: "unknown" })),
+      new Authentication({ message: "auth" }),
+      new PermissionDenied({ message: "forbidden" }),
+      new NotFound({ message: "missing" }),
+      new QuotaExceeded({ message: "quota" }),
+      new ContentPolicy({ message: "blocked" }),
+      new ContextOverflow({ message: "too long" }),
+      new MalformedResponse({ message: "output" }),
+      new BadRequest({ message: "request" }),
+      new NoRoute({
+        route: RouteID.make("route"),
+        provider: ProviderID.make("provider"),
+        model: ModelID.make("model"),
+      }),
+      new APIError({ message: "unknown" }),
     ]
 
-    expect(eligible.map(SessionRunnerRetry.isRetryable)).toEqual([true, true, true])
-    expect(ineligible.map(SessionRunnerRetry.isRetryable)).toEqual([false, false, false, false, false, false, false])
+    expect(eligible.map(SessionRunnerRetry.isRetryable)).toEqual([true, true, true, true])
+    expect(ineligible.map(SessionRunnerRetry.isRetryable)).toEqual(ineligible.map(() => false))
   })
 })

packages/core/test/session-execution.test.ts

@@ -1,5 +1,5 @@
 import { describe, expect, test } from "bun:test"
-import { LLMError, TransportReason } from "@opencode-ai/llm"
+import { ConnectionError } from "@opencode-ai/llm"
 import { Database } from "@opencode-ai/core/database/database"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
@@ -25,17 +25,10 @@ const it = testEffect(AppNodeBuilder.build(LayerNode.group([Database.node, Event
 describe("SessionExecution lifecycle", () => {
   test("classifies success and typed failure terminals", () => {
     expect(SessionExecution.terminal(Exit.succeed(undefined))).toEqual({ type: "succeeded" })
-    expect(
-      SessionExecution.terminal(
-        Exit.fail(
-          new LLMError({
-            module: "test",
-            method: "stream",
-            reason: new TransportReason({ message: "Disconnected" }),
-          }),
-        ),
-      ),
-    ).toEqual({ type: "failed", error: { type: "provider.transport", message: "Disconnected" } })
+    expect(SessionExecution.terminal(Exit.fail(new ConnectionError({ message: "Disconnected" })))).toEqual({
+      type: "failed",
+      error: { type: "provider.transport", message: "Disconnected" },
+    })
     const storage = new ToolOutputStore.StorageError({ operation: "encode", cause: new Error("invalid output") })
     expect(SessionExecution.terminal(Exit.fail(storage))).toEqual({
       type: "failed",

packages/core/test/session-runner-tool-registry.test.ts

@@ -70,7 +70,7 @@ describe("ToolRegistry", () => {
         bash: make(),
         edit: make("edit"),
         write: make("edit"),
-      })
+      }, { codemode: false })
       const names = (permissions: PermissionV2.Ruleset) =>
         toolDefinitions(service, permissions).pipe(Effect.map((definitions) => definitions.map((tool) => tool.name)))
 
@@ -95,8 +95,8 @@ describe("ToolRegistry", () => {
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service
       const shared = make()
-      yield* service.register({ first: shared })
-      yield* service.register({ second: Tool.withPermission(shared, "edit") })
+      yield* service.register({ first: shared }, { codemode: false })
+      yield* service.register({ second: Tool.withPermission(shared, "edit") }, { codemode: false })
       Tool.withPermission(shared, "question")
 
       expect(
@@ -110,7 +110,7 @@ describe("ToolRegistry", () => {
   it.effect("reuses model definitions across requests", () =>
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service
-      yield* service.register({ echo: make() })
+      yield* service.register({ echo: make() }, { codemode: false })
       const first = yield* toolDefinitions(service)
       const second = yield* toolDefinitions(service)
 
@@ -122,7 +122,7 @@ describe("ToolRegistry", () => {
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service
       const scope = yield* Scope.make()
-      yield* service.register({ echo: make() }).pipe(Scope.provide(scope))
+      yield* service.register({ echo: make() }, { codemode: false }).pipe(Scope.provide(scope))
       expect((yield* toolDefinitions(service)).map((tool) => tool.name)).toEqual(["echo"])
       yield* Scope.close(scope, Exit.void)
       expect(yield* toolDefinitions(service)).toEqual([])
@@ -135,7 +135,7 @@ describe("ToolRegistry", () => {
       const scope = yield* Scope.make()
       const registered = yield* Deferred.make<void>()
       const fiber = yield* service
-        .register({ echo: make() })
+        .register({ echo: make() }, { codemode: false })
         .pipe(
           Effect.andThen(Deferred.succeed(registered, undefined)),
           Effect.andThen(Effect.never),
@@ -161,7 +161,7 @@ describe("ToolRegistry", () => {
           output: Schema.Struct({ ok: Schema.Boolean }),
           execute: () => Effect.fail(new Tool.Failure({ message: "Denied" })),
         }),
-      })
+      }, { codemode: false })
       expect(
         yield* executeTool(service, {
           sessionID,
@@ -184,7 +184,7 @@ describe("ToolRegistry", () => {
           output: Schema.Struct({}),
           execute: () => Effect.die("unexpected executor defect"),
         }),
-      })
+      }, { codemode: false })
       expect(
         yield* service.materialize().pipe(
           Effect.flatMap((materialized) =>
@@ -203,7 +203,7 @@ describe("ToolRegistry", () => {
   it.effect("propagates retention failures through settlement", () =>
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service
-      yield* service.register({ echo: make() })
+      yield* service.register({ echo: make() }, { codemode: false })
       const materialized = yield* service.materialize()
       const exit = yield* materialized.settle(call("echo", "call-retention-failure")).pipe(Effect.exit)
 
@@ -234,7 +234,7 @@ describe("ToolRegistry", () => {
           output: Schema.Struct({ ok: Schema.Boolean }),
           execute: (_, context) => Effect.sync(() => contexts.push(context)).pipe(Effect.as({ ok: true })),
         }),
-      })
+      }, { codemode: false })
       yield* executeTool(service, {
         sessionID,
         ...identity,
@@ -248,7 +248,7 @@ describe("ToolRegistry", () => {
     Effect.gen(function* () {
       bounds.length = 0
       const service = yield* ToolRegistry.Service
-      yield* service.register({ bounded: make() })
+      yield* service.register({ bounded: make() }, { codemode: false })
       expect(
         yield* settleTool(service, {
           sessionID,
@@ -282,7 +282,7 @@ describe("ToolRegistry", () => {
           execute: ({ value }) => Effect.sync(() => executed.push(value)).pipe(Effect.as({ value })),
           toModelOutput: ({ output }) => [{ type: "text", text: String(output.value) }],
         }),
-      })
+      }, { codemode: false })
 
       expect(
         yield* executeTool(service, {
@@ -319,7 +319,7 @@ describe("ToolRegistry", () => {
           }),
           execute: () => Effect.succeed({ value: "invalid" }),
         }),
-      })
+      }, { codemode: false })
       expect(
         yield* executeTool(service, {
           sessionID,
@@ -334,10 +334,10 @@ describe("ToolRegistry", () => {
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service
       const scope = yield* Scope.make()
-      yield* service.register({ echo: constant("advertised") }).pipe(Scope.provide(scope))
+      yield* service.register({ echo: constant("advertised") }, { codemode: false }).pipe(Scope.provide(scope))
       const request = yield* service.materialize()
       yield* Scope.close(scope, Exit.void)
-      yield* service.register({ echo: constant("replacement") })
+      yield* service.register({ echo: constant("replacement") }, { codemode: false })
 
       expect((yield* request.settle(call("echo"))).result).toEqual({ type: "text", value: "advertised" })
       expect(yield* executeTool(service, call("echo"))).toEqual({ type: "text", value: "replacement" })
@@ -347,9 +347,9 @@ describe("ToolRegistry", () => {
   it.effect("reveals the previous registration after an overlay closes", () =>
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service
-      yield* service.register({ echo: constant("base") })
+      yield* service.register({ echo: constant("base") }, { codemode: false })
       const overlay = yield* Scope.make()
-      yield* service.register({ echo: constant("overlay") }).pipe(Scope.provide(overlay))
+      yield* service.register({ echo: constant("overlay") }, { codemode: false }).pipe(Scope.provide(overlay))
 
       expect(yield* executeTool(service, call("echo"))).toEqual({ type: "text", value: "overlay" })
       yield* Scope.close(overlay, Exit.void)
@@ -357,37 +357,31 @@ describe("ToolRegistry", () => {
     }),
   )
 
-  it.effect("executes deferred tools advertised in a model request", () =>
+  it.effect("executes codemode tools advertised in a model request", () =>
     Effect.gen(function* () {
       const service = yield* ToolRegistry.Service
       const executed: string[] = []
       const scope = yield* Scope.make()
       yield* service
-        .register(
-          {
-            echo: Tool.make({
-              description: "Echo text",
-              input: Schema.Struct({ text: Schema.String }),
-              output: Schema.Struct({ text: Schema.String }),
-              execute: ({ text }) => Effect.sync(() => executed.push(`old:${text}`)).pipe(Effect.as({ text })),
-            }),
-          },
-          { deferred: true },
-        )
-        .pipe(Scope.provide(scope))
-      const materialized = yield* service.materialize()
-      yield* Scope.close(scope, Exit.void)
-      yield* service.register(
-        {
+        .register({
           echo: Tool.make({
             description: "Echo text",
             input: Schema.Struct({ text: Schema.String }),
             output: Schema.Struct({ text: Schema.String }),
-            execute: ({ text }) => Effect.sync(() => executed.push(`new:${text}`)).pipe(Effect.as({ text })),
+            execute: ({ text }) => Effect.sync(() => executed.push(`old:${text}`)).pipe(Effect.as({ text })),
           }),
-        },
-        { deferred: true },
-      )
+        })
+        .pipe(Scope.provide(scope))
+      const materialized = yield* service.materialize()
+      yield* Scope.close(scope, Exit.void)
+      yield* service.register({
+        echo: Tool.make({
+          description: "Echo text",
+          input: Schema.Struct({ text: Schema.String }),
+          output: Schema.Struct({ text: Schema.String }),
+          execute: ({ text }) => Effect.sync(() => executed.push(`new:${text}`)).pipe(Effect.as({ text })),
+        }),
+      })
 
       const settlement = yield* materialized.settle({
         ...call("execute"),

packages/core/test/session-runner.test.ts

@@ -1,15 +1,17 @@
 import { describe, expect, test } from "bun:test"
 import {
+  APIError,
+  BadRequest,
+  ConnectionError,
+  ContextOverflow,
   LLMClient,
-  LLMError,
   LLMEvent,
+  MalformedResponse,
   Model,
+  RateLimit,
   ToolFailure,
-  TransportReason,
-  InvalidProviderOutputReason,
-  InvalidRequestReason,
-  RateLimitReason,
   type LLMClientShape,
+  type LLMError,
   type LLMRequest,
 } from "@opencode-ai/llm"
 import * as OpenAIChat from "@opencode-ai/llm/protocols/openai-chat"
@@ -69,8 +71,9 @@ import { asc, eq } from "drizzle-orm"
 import { testEffect } from "./lib/effect"
 
 const requests: LLMRequest[] = []
+type ScriptedResponse = LLMEvent[] | Stream.Stream<LLMEvent, LLMError>
 let response: LLMEvent[] = []
-let responses: LLMEvent[][] | undefined
+let responses: ScriptedResponse[] | undefined
 let responseStream: Stream.Stream<LLMEvent, LLMError> | undefined
 let responseStreams: Stream.Stream<LLMEvent, LLMError>[] | undefined
 let streamGate: Deferred.Deferred<void> | undefined
@@ -93,9 +96,12 @@ const client = Layer.succeed(
         responseStream = undefined
         return stream
       }
+      const scripted = responses === undefined ? response : (responses.shift() ?? [])
       const events = streamFailure
         ? Stream.fail(streamFailure)
-        : Stream.fromIterable(responses === undefined ? response : (responses.shift() ?? []))
+        : Array.isArray(scripted)
+          ? Stream.fromIterable(scripted)
+          : scripted
       if (!streamGate) return events
       return Stream.unwrap(
         (streamStarted ? Deferred.succeed(streamStarted, undefined) : Effect.void).pipe(
@@ -261,7 +267,7 @@ const echo = Layer.effectDiscard(
         output: Schema.Any,
         execute: () => Effect.succeed({ big: 1n }),
       }),
-    }),
+    }, { codemode: false }),
   ),
 )
 const echoNode = makeLocationNode({ name: "test/session-runner-tools", layer: echo, deps: [ToolRegistry.node] })
@@ -483,26 +489,16 @@ const setup = Effect.gen(function* () {
   return yield* SessionV2.Service
 })
 
-const providerUnavailable = () =>
-  new LLMError({
-    module: "test",
-    method: "stream",
-    reason: new TransportReason({ message: "Provider unavailable" }),
-  })
+const providerUnavailable = () => new ConnectionError({ message: "Provider unavailable" })
 
-const invalidRequest = () =>
-  new LLMError({
-    module: "test",
-    method: "stream",
-    reason: new InvalidRequestReason({ message: "Invalid request" }),
-  })
+const contextOverflow = () => new ContextOverflow({ message: "prompt too long" })
 
-const rateLimited = (retryAfterMs?: number) =>
-  new LLMError({
-    module: "test",
-    method: "stream",
-    reason: new RateLimitReason({ message: "Rate limited", retryAfterMs }),
-  })
+const failingResponse = (events: LLMEvent[], failure: LLMError): Stream.Stream<LLMEvent, LLMError> =>
+  Stream.fromIterable(events).pipe(Stream.concat(Stream.fail(failure)))
+
+const invalidRequest = () => new BadRequest({ message: "Invalid request" })
+
+const rateLimited = (retryAfterMs?: number) => new RateLimit({ message: "Rate limited", retryAfterMs })
 
 const setupOverflowRecovery = Effect.gen(function* () {
   const session = yield* setup
@@ -782,7 +778,7 @@ describe("SessionRunnerLLM", () => {
               return { answer: query.toUpperCase() }
             }),
         }),
-      })
+      }, { codemode: false })
       yield* admit(session, "Use application context")
       responses = [reply.tool("call-location", "location_context", { query: "hello" }), []]
 
@@ -827,7 +823,7 @@ describe("SessionRunnerLLM", () => {
             output: Schema.Struct({ value: Schema.String }),
             execute: () => Effect.sync(() => executions.push("advertised")).pipe(Effect.as({ value: "advertised" })),
           }),
-        })
+        }, { codemode: false })
         .pipe(Scope.provide(scope))
       yield* admit(session, "Use the reloaded tool")
       responses = [
@@ -852,7 +848,7 @@ describe("SessionRunnerLLM", () => {
           output: Schema.Struct({ value: Schema.String }),
           execute: () => Effect.sync(() => executions.push("replacement")).pipe(Effect.as({ value: "replacement" })),
         }),
-      })
+      }, { codemode: false })
       yield* Deferred.succeed(streamGate, undefined)
       yield* Fiber.join(run)
 
@@ -1614,14 +1610,14 @@ describe("SessionRunnerLLM", () => {
     }),
   )
 
-  it.effect("runs one durable compaction barrier before later steer and queued prompts", () =>
+  it.effect("runs one durable compaction barrier after tool settlement and before later inputs", () =>
     Effect.gen(function* () {
       const session = yield* setup
       currentModel = recoveryModel
       streamGate = yield* Deferred.make<void>()
       streamStarted = yield* Deferred.make<void>()
       responses = [
-        reply.text("Active complete", "text-active"),
+        reply.tool("call-active", "echo", { text: "active" }),
         [LLMEvent.textDelta({ id: "summary", text: "durable summary" })],
         reply.text("Steer complete", "text-steer"),
         reply.text("Queue complete", "text-queue"),
@@ -1758,14 +1754,14 @@ describe("SessionRunnerLLM", () => {
       yield* admit(session, "Earlier question")
       yield* session.resume(sessionID)
 
-      response = [LLMEvent.providerError({ message: "summary unavailable" })]
+      responseStream = Stream.fail(new APIError({ message: "summary unavailable" }))
       const compaction = yield* session.compact({ sessionID })
       yield* session.resume(sessionID)
 
       expect((yield* session.messages({ sessionID })).find((message) => message.id === compaction.id)).toMatchObject({
         type: "compaction",
         status: "failed",
-        error: { type: "provider.error", message: "summary unavailable" },
+        error: { type: "provider.unknown", message: "summary unavailable" },
       })
     }),
   )
@@ -1875,7 +1871,7 @@ describe("SessionRunnerLLM", () => {
       currentModel = compactModel
       requests.length = 0
       responses = [
-        [LLMEvent.providerError({ message: "Unsupported parameter: max_output_tokens" })],
+        Stream.fail(new BadRequest({ message: "Unsupported parameter: max_output_tokens" })),
         reply.text("Must not run", "text-after-failed-compaction"),
       ]
       yield* admit(session, "Recent exact request ".repeat(180))
@@ -1898,10 +1894,7 @@ describe("SessionRunnerLLM", () => {
     Effect.gen(function* () {
       const session = yield* setupOverflowRecovery
       responses = [
-        [
-          LLMEvent.stepStart({ index: 0 }),
-          LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" }),
-        ],
+        failingResponse([LLMEvent.stepStart({ index: 0 })], contextOverflow()),
         reply.text("## Objective\n- Recover overflow", "text-summary"),
         reply.text("Recovered", "text-final"),
       ]
@@ -1928,7 +1921,7 @@ describe("SessionRunnerLLM", () => {
       const session = yield* setupOverflowRecovery
       currentModel = model
       responses = [
-        [LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" })],
+        Stream.fail(contextOverflow()),
         reply.text("## Objective\n- Recover unknown limit", "text-summary-unknown-limit"),
         reply.text("Recovered", "text-final-unknown-limit"),
       ]
@@ -1948,7 +1941,7 @@ describe("SessionRunnerLLM", () => {
       const session = yield* setupOverflowRecovery
       currentModel = undersizedContextModel
       responses = [
-        [LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" })],
+        Stream.fail(contextOverflow()),
         reply.text("## Objective\n- Recover undersized limit", "text-summary-undersized-limit"),
         reply.text("Recovered", "text-final-undersized-limit"),
       ]
@@ -1966,10 +1959,7 @@ describe("SessionRunnerLLM", () => {
   it.effect("persists a second context overflow after one recovery", () =>
     Effect.gen(function* () {
       const session = yield* setupOverflowRecovery
-      const overflow = () => [
-        LLMEvent.stepStart({ index: 0 }),
-        LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" }),
-      ]
+      const overflow = () => failingResponse([LLMEvent.stepStart({ index: 0 })], contextOverflow())
       responses = [overflow(), reply.text("## Objective\n- Recover once", "text-summary"), overflow()]
       yield* admit(session, "Continue")
       expect((yield* session.resume(sessionID).pipe(Effect.flip)).message).toBe("prompt too long")
@@ -1985,16 +1975,7 @@ describe("SessionRunnerLLM", () => {
   it.effect("recovers once from a raw context overflow failure", () =>
     Effect.gen(function* () {
       const session = yield* setupOverflowRecovery
-      responseStream = Stream.fail(
-        new LLMError({
-          module: "test",
-          method: "stream",
-          reason: new InvalidRequestReason({
-            message: "prompt too long",
-            classification: "context-overflow",
-          }),
-        }),
-      )
+      responseStream = Stream.fail(contextOverflow())
       responses = [
         reply.text("## Objective\n- Recover raw overflow", "text-summary"),
         reply.text("Recovered", "text-final"),
@@ -2013,10 +1994,7 @@ describe("SessionRunnerLLM", () => {
   it.effect("publishes the original overflow when recovery summarization fails", () =>
     Effect.gen(function* () {
       const session = yield* setupOverflowRecovery
-      responses = [
-        [LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" })],
-        [LLMEvent.providerError({ message: "summary unavailable" })],
-      ]
+      responses = [Stream.fail(contextOverflow()), Stream.fail(new APIError({ message: "summary unavailable" }))]
       yield* admit(session, "Continue")
       expect((yield* session.resume(sessionID).pipe(Effect.flip)).message).toBe("prompt too long")
 
@@ -2027,7 +2005,7 @@ describe("SessionRunnerLLM", () => {
           type: "compaction",
           status: "failed",
           reason: "auto",
-          error: { type: "provider.error", message: "summary unavailable" },
+          error: { type: "provider.unknown", message: "summary unavailable" },
         }),
       )
       expect(context.slice(-3)).toMatchObject([
@@ -2041,10 +2019,7 @@ describe("SessionRunnerLLM", () => {
   it.effect("interrupts overflow recovery while the summary provider is running", () =>
     Effect.gen(function* () {
       const session = yield* setupOverflowRecovery
-      responses = [
-        [LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" })],
-        reply.text("## Objective\n- Interrupted", "text-summary"),
-      ]
+      responses = [Stream.fail(contextOverflow()), reply.text("## Objective\n- Interrupted", "text-summary")]
       const firstGate = yield* Deferred.make<void>()
       const summaryGate = yield* Deferred.make<void>()
       streamGate = firstGate
@@ -3189,7 +3164,7 @@ describe("SessionRunnerLLM", () => {
               Effect.mapError(() => new Tool.Failure({ message: "Permission blocked" })),
             ),
         }),
-      })
+      }, { codemode: false })
       yield* admit(session, "Call blocked")
 
       responses = [reply.tool("call-blocked", "blocked", {}), reply.stop()]
@@ -3221,7 +3196,7 @@ describe("SessionRunnerLLM", () => {
           output: Schema.Struct({}),
           execute: () => Effect.die(new PermissionV2.DeclinedError()),
         }),
-      })
+      }, { codemode: false })
       yield* admit(session, "Call declined")
 
       response = reply.tool("call-declined", "declined", {})
@@ -3261,7 +3236,7 @@ describe("SessionRunnerLLM", () => {
               Effect.mapError(() => new Tool.Failure({ message: "Use another tool" })),
             ),
         }),
-      })
+      }, { codemode: false })
       yield* admit(session, "Call corrected")
 
       responses = [reply.tool("call-corrected", "corrected", {}), reply.stop()]
@@ -3321,7 +3296,7 @@ describe("SessionRunnerLLM", () => {
     Effect.gen(function* () {
       const session = yield* setup
       const registry = yield* ToolRegistry.Service
-      yield* registry.register({ permissionfail: permissionFail })
+      yield* registry.register({ permissionfail: permissionFail }, { codemode: false })
       yield* admit(session, "Reject permission")
       responses = [
         reply.tool("call-permission", "permissionfail", {}),
@@ -3366,7 +3341,7 @@ describe("SessionRunnerLLM", () => {
           output: Schema.Struct({}),
           execute: () => Effect.die(new QuestionTool.CancelledError()),
         }),
-      })
+      }, { codemode: false })
       yield* admit(session, "Ask then stop")
 
       responses = [reply.tool("call-question", "question", {}), []]
@@ -3627,7 +3602,10 @@ describe("SessionRunnerLLM", () => {
       const session = yield* setup
       yield* admit(session, "Fail durably")
 
-      response = [LLMEvent.stepStart({ index: 0 }), LLMEvent.providerError({ message: "Provider unavailable" })]
+      responseStream = failingResponse(
+        [LLMEvent.stepStart({ index: 0 })],
+        new APIError({ message: "Provider unavailable" }),
+      )
 
       expect((yield* session.resume(sessionID).pipe(Effect.flip)).message).toBe("Provider unavailable")
 
@@ -3644,7 +3622,7 @@ describe("SessionRunnerLLM", () => {
       const session = yield* setup
       yield* admit(session, "Fail before step")
 
-      response = [LLMEvent.providerError({ message: "Provider unavailable" })]
+      responseStream = Stream.fail(new APIError({ message: "Provider unavailable" }))
 
       expect((yield* session.resume(sessionID).pipe(Effect.flip)).message).toBe("Provider unavailable")
 
@@ -3732,13 +3710,15 @@ describe("SessionRunnerLLM", () => {
       const session = yield* setup
       yield* admit(session, "Fail after output")
 
-      response = [
-        LLMEvent.stepStart({ index: 0 }),
-        LLMEvent.textStart({ id: "text-partial" }),
-        LLMEvent.textDelta({ id: "text-partial", text: "Partial" }),
-        LLMEvent.textEnd({ id: "text-partial" }),
-        LLMEvent.providerError({ message: "prompt too long", classification: "context-overflow" }),
-      ]
+      responseStream = failingResponse(
+        [
+          LLMEvent.stepStart({ index: 0 }),
+          LLMEvent.textStart({ id: "text-partial" }),
+          LLMEvent.textDelta({ id: "text-partial", text: "Partial" }),
+          LLMEvent.textEnd({ id: "text-partial" }),
+        ],
+        contextOverflow(),
+      )
       expect((yield* session.resume(sessionID).pipe(Effect.flip)).message).toBe("prompt too long")
 
       expect(requests).toHaveLength(1)
@@ -3892,11 +3872,7 @@ describe("SessionRunnerLLM", () => {
     Effect.gen(function* () {
       const session = yield* setup
       yield* admit(session, "Call a malformed tool")
-      const failure = new LLMError({
-        module: "test",
-        method: "stream",
-        reason: new InvalidProviderOutputReason({ message: "Invalid JSON input for tool call echo" }),
-      })
+      const failure = new MalformedResponse({ message: "Invalid JSON input for tool call echo" })
       responseStream = Stream.fromIterable([
         LLMEvent.stepStart({ index: 0 }),
         LLMEvent.toolInputStart({ id: "call-malformed", name: "echo" }),
@@ -3935,11 +3911,13 @@ describe("SessionRunnerLLM", () => {
       toolExecutionGate = yield* Deferred.make<void>()
       toolExecutionsStarted = yield* Deferred.make<void>()
       toolExecutionsReady = 1
-      response = [
-        LLMEvent.stepStart({ index: 0 }),
-        LLMEvent.toolCall({ id: "call-before-provider-error", name: "echo", input: { text: "settled" } }),
-        LLMEvent.providerError({ message: "Provider unavailable" }),
-      ]
+      responseStream = failingResponse(
+        [
+          LLMEvent.stepStart({ index: 0 }),
+          LLMEvent.toolCall({ id: "call-before-provider-error", name: "echo", input: { text: "settled" } }),
+        ],
+        new APIError({ message: "Provider unavailable" }),
+      )
 
       const run = yield* session.resume(sessionID).pipe(Effect.forkChild)
       yield* Deferred.await(toolExecutionsStarted)
@@ -3966,11 +3944,10 @@ describe("SessionRunnerLLM", () => {
       const session = yield* setup
       yield* admit(session, "Fail hosted tool durably")
 
-      response = [
-        LLMEvent.stepStart({ index: 0 }),
-        hostedCall("call-hosted-provider-error", "effect"),
-        LLMEvent.providerError({ message: "Provider unavailable" }),
-      ]
+      responseStream = failingResponse(
+        [LLMEvent.stepStart({ index: 0 }), hostedCall("call-hosted-provider-error", "effect")],
+        new APIError({ message: "Provider unavailable" }),
+      )
 
       expect((yield* session.resume(sessionID).pipe(Effect.flip)).message).toBe("Provider unavailable")
 
@@ -3997,11 +3974,13 @@ describe("SessionRunnerLLM", () => {
     Effect.gen(function* () {
       const session = yield* setup
       yield* admit(session, "Defect while provider fails")
-      response = [
-        LLMEvent.stepStart({ index: 0 }),
-        LLMEvent.toolCall({ id: "call-defect-provider-error", name: "defect", input: {} }),
-        LLMEvent.providerError({ message: "Provider unavailable" }),
-      ]
+      responseStream = failingResponse(
+        [
+          LLMEvent.stepStart({ index: 0 }),
+          LLMEvent.toolCall({ id: "call-defect-provider-error", name: "defect", input: {} }),
+        ],
+        new APIError({ message: "Provider unavailable" }),
+      )
 
       expect((yield* session.resume(sessionID).pipe(Effect.flip)).message).toBe("Provider unavailable")
 

packages/core/test/tool-shell.test.ts

@@ -2,7 +2,7 @@ import fs from "fs/promises"
 import { realpathSync } from "node:fs"
 import path from "path"
 import { describe, expect, test } from "bun:test"
-import { DateTime, Duration, Effect, Fiber, Layer, Scope } from "effect"
+import { DateTime, Duration, Effect, Fiber, Layer, Scope, Stream } from "effect"
 import { Money } from "@opencode-ai/schema/money"
 import { AppNodeBuilder } from "@opencode-ai/core/effect/app-node-builder"
 import { LayerNode } from "@opencode-ai/core/effect/layer-node"
@@ -443,6 +443,12 @@ describe("ShellTool", () => {
         reset()
         return withSession(tmp.path, (registry) =>
           Effect.gen(function* () {
+            const events = yield* EventV2.Service
+            const admitted = yield* events.subscribe(SessionEvent.InputAdmitted).pipe(
+              Stream.filter((event) => event.data.sessionID === sessionID && event.data.input.type === "synthetic"),
+              Stream.runHead,
+              Effect.forkScoped({ startImmediately: true }),
+            )
             const settled = yield* settleTool(registry, call({ command: idleCommand, timeout: 50, background: true }))
             const structured = settled.output?.structured as Record<string, unknown> | undefined
             const shellID = typeof structured?.shellID === "string" ? structured.shellID : undefined
@@ -454,6 +460,13 @@ describe("ShellTool", () => {
             const id = ShellSchema.ID.make(shellID)
             expect((yield* shell.list()).map((info) => info.id)).toContain(id)
             expect((yield* shell.wait(id)).status).toBe("timeout")
+            expect((yield* Fiber.join(admitted)).valueOrUndefined?.data.input.data).toMatchObject({
+              description: idleCommand,
+              metadata: {
+                source: "shell",
+                state: "completed",
+              },
+            })
           }),
         )
       },

packages/docs/build/plugins.mdx

@@ -315,11 +315,12 @@ export default Plugin.define({
 Unsupported characters in tool and group names are normalized to underscores.
 The resulting exposed key must begin with a letter and contain at most 64
 letters, digits, underscores, or hyphens. Set `options` on the declaration to
-configure registration with `{ group, deferred }`:
+configure registration with `{ group, codemode }`:
 
 - `group` prefixes and groups the exposed tool name.
-- `deferred: true` makes the tool available through the deferred `execute`
-  tool instead of exposing it directly.
+- `codemode` defaults to `true` and makes the tool available through the
+  `execute` CodeMode tool. Set `codemode: false` to expose it directly to the
+  provider.
 
 The executor receives a second context argument containing `sessionID`,
 `agent`, `assistantMessageID`, and `toolCallID`.

packages/docs/build/sdk.mdx

@@ -1,12 +1,21 @@
 ---
 title: "SDK"
-description: "Embed an OpenCode host in an Effect application."
+description: "Embed OpenCode directly in your application."
 ---
 
-`@opencode-ai/sdk-next` is the Effect-native SDK for applications that need to
-host OpenCode in-process. Unlike the [network client](/build/client), it assembles the
-OpenCode server and routes API calls through its HTTP router in memory. It opens
-no HTTP listener and adds no network hop between the client and server.
+We're working on a general-purpose SDK for embedding OpenCode directly inside
+your application. The regular SDK is coming soon.
+
+An Effect-native version is available now for applications built with Effect.
+Its current documentation is below. For other applications, run OpenCode as a
+server and use the [TypeScript client](/build/client) in the meantime.
+
+## Effect
+
+`@opencode-ai/sdk-next` hosts OpenCode in-process. Unlike the
+[network client](/build/client), it assembles the OpenCode server and routes API
+calls through its HTTP router in memory. It opens no HTTP listener and adds no
+network hop between the client and server.
 
 <Warning>
   The V2 SDK is beta and currently private to the OpenCode workspace. It is not

packages/docs/migrate-v1.mdx

@@ -3,14 +3,19 @@ title: "Migrate from V1"
 description: "Move from OpenCode V1 to the OpenCode 2.0 beta."
 ---
 
-<Note>
-  The only intentional breaking changes in V2 are the server API and the plugin API. All other functionality is intended
-  to remain compatible with V1.
-</Note>
+## Breaking changes
 
-Existing config files, agent definitions, command definitions, skills, and other files in `.opencode/` should continue to
-work without changes. If one of these stops working in V2, treat it as a beta compatibility bug rather than an expected
-migration requirement.
+V2 has three intentional breaking changes:
+
+- [Plugins](#plugins) use a new plugin API.
+- The [server API and clients](#server-api-and-clients) have new contracts.
+- [TUI configuration](#tui-configuration) moves from layered `tui.json(c)` files to one global `cli.json` file (auto migrated).
+
+All other functionality is intended to remain compatible with V1.
+
+Existing server config files, agent definitions, command definitions, skills, and other files in `.opencode/` should
+continue to work without changes. If one of these stops working in V2, treat it as a beta compatibility bug rather than
+an expected migration requirement.
 
 <Tip>
   Run `/report` if existing V1 functionality does not work in V2. The report skill collects diagnostics and helps you file
@@ -466,6 +471,58 @@ If a V1 setup relied on a `CLAUDE.md` fallback, move that guidance into the appl
 discovers `AGENTS.md`; because non-API V1 behavior is intended to remain compatible, also run `/report` with the affected
 project details. See [Instructions](/instructions).
 
+## TUI configuration
+
+V1 loaded `tui.json(c)` from the global config directory and from project directories discovered while walking up from
+the current directory. V2 instead stores CLI and TUI settings in one global file:
+
+```text
+~/.config/opencode/cli.json
+```
+
+The CLI owns this file. The background service does not load it, and V2 does not discover or merge project-local
+`tui.json(c)` or `cli.json` files.
+
+The native V2 format groups related settings. For example:
+
+```jsonc
+// V1: ~/.config/opencode/tui.json
+{
+  "theme": "tokyonight",
+  "scroll_speed": 2,
+  "scroll_acceleration": {
+    "enabled": true
+  }
+}
+
+// V2: ~/.config/opencode/cli.json
+{
+  "theme": {
+    "name": "tokyonight"
+  },
+  "scroll": {
+    "speed": 2,
+    "acceleration": true
+  }
+}
+```
+
+V2 migrates the global TUI configuration automatically. On the first CLI or TUI startup, when `cli.json` does not already
+exist, it:
+
+- Reads `~/.config/opencode/tui.json`.
+- Reads persisted TUI preferences from the legacy `kv.json` state file.
+- Converts supported settings to the native grouped format and writes `~/.config/opencode/cli.json`.
+- Leaves the V1 files unchanged so V1 can continue using them.
+
+Migration runs only while `cli.json` is absent. Once that file exists, V2 treats it as the source of truth and does not
+continually synchronize later changes from `tui.json` or `kv.json`. If you created `cli.json` before starting V2, merge any
+V1 settings you still need into it manually.
+
+Project-local V1 TUI configuration is not migrated because V2 has no project-local CLI configuration. Move settings you
+still want into the global `cli.json`; when multiple projects used different values for the same setting, choose the
+global behavior you want V2 to use.
+
 ## Plugins
 
 Rename `plugin` to `plugins`. Replace a package-and-options tuple with an object:

packages/llm/AGENTS.md

@@ -182,8 +182,8 @@ The dependency arrow points down: `providers/*.ts` files import protocol routes
 - `joinText(parts)` — joins an array of `TextPart` (or anything with a `.text`) with newlines. Use this anywhere a protocol flattens text content into a single string for a provider field.
 - `parseToolInput(route, name, raw)` — Schema-decodes a tool-call argument string with the canonical "Invalid JSON input for `<route>` tool call `<name>`" error message. Treats empty input as `{}`.
 - `parseJson(route, raw, message)` — generic JSON-via-Schema decode for non-tool bodies.
-- `eventError(route, message, ...)` — typed `InvalidProviderOutput` constructor for stream-time decode failures.
-- `validateWith(decoder)` — maps Schema decode errors to `InvalidRequest`. `Route.make(...)` uses this for body validation; lower-level routes can reuse it.
+- `eventError(route, message, ...)` — typed `MalformedResponse` constructor for stream-time decode failures.
+- `validateWith(decoder)` — maps Schema decode errors to `BadRequest`. `Route.make(...)` uses this for body validation; lower-level routes can reuse it.
 - `matchToolChoice(provider, choice, branches)` — branches over `LLMRequest["toolChoice"]` for provider-specific lowering.
 
 If you find yourself copying a 3-to-5-line snippet between two protocols, lift it into `ProviderShared` next to these helpers rather than duplicating.
@@ -291,7 +291,7 @@ Use this order for every protocol module:
 
 - Keep protocol files focused on the protocol. Move provider-specific projection, signing, media normalization, or other bulky transformations into `src/protocols/utils/*`.
 - Use `Effect.fn("Provider.fromRequest")` for request body construction entrypoints. Use `Effect.fn(...)` for event handlers that yield effects; keep purely synchronous handlers as plain functions returning a `StepResult` that the dispatcher lifts via `Effect.succeed(...)`.
-- Parser state owns terminal information. The state machine records finish reason, usage, and pending tool calls; emit one terminal `finish` event (or `provider-error`) for each completed response. If a provider splits reason and usage across events, merge them in parser state before flushing.
+- Parser state owns terminal information. The state machine records finish reason, usage, and pending tool calls; emit one terminal `finish` event for each completed response. Provider-reported failures (SSE error events, exception frames) fail the stream with a typed `LLMError` via `classifyApiFailure` — never an ordinary event. If a provider splits reason and usage across events, merge them in parser state before flushing.
 - Emit exactly one terminal `finish` event for a completed response, normally after a matching `step-finish`. Use `stream.terminal` to stop reading when the provider has a completion sentinel; use `stream.onHalt` when the final event must be flushed after the framed stream ends.
 - Use shared helpers for repeated protocol policy such as text joining, usage totals, JSON parsing, and tool-call accumulation. `ToolStream` (`protocols/utils/tool-stream.ts`) accumulates streamed tool-call arguments uniformly.
 - Make intentional provider differences explicit in helper names or comments. If two protocol files differ visually, the reason should be obvious from the names.

packages/llm/src/index.ts

@@ -2,7 +2,7 @@ export { LLMClient } from "./route/client"
 export { Auth } from "./route/auth"
 export { Provider } from "./provider"
 export { ProviderPackage } from "./provider-package"
-export { isContextOverflow, isContextOverflowFailure } from "./provider-error"
+export { classifyApiFailure, isContextOverflow, type ApiFailure } from "./provider-error"
 export type {
   RouteModelInput,
   RouteRoutedModelInput,

packages/llm/src/llm.ts

@@ -3,8 +3,8 @@ import { LLMClient } from "./route/client"
 import {
   GenerationOptions,
   HttpOptions,
-  InvalidProviderOutputReason,
-  LLMError,
+  MalformedResponse,
+  type LLMError,
   LLMEvent,
   LLMRequest,
   LLMResponse,
@@ -121,22 +121,14 @@ const runGenerateObject = Effect.fn("LLM.generateObject")(function* (
     (event) => LLMEvent.is.toolCall(event) && event.name === GENERATE_OBJECT_TOOL_NAME,
   )
   if (!call || !LLMEvent.is.toolCall(call))
-    return yield* new LLMError({
-      module: "LLM",
-      method: "generateObject",
-      reason: new InvalidProviderOutputReason({
-        message: `generateObject: model did not call the forced \`${GENERATE_OBJECT_TOOL_NAME}\` tool`,
-      }),
+    return yield* new MalformedResponse({
+      message: `generateObject: model did not call the forced \`${GENERATE_OBJECT_TOOL_NAME}\` tool`,
     })
   const object = yield* tool._decode(call.input).pipe(
     Effect.mapError(
       (error) =>
-        new LLMError({
-          module: "LLM",
-          method: "generateObject",
-          reason: new InvalidProviderOutputReason({
-            message: `generateObject: tool input failed schema decode: ${error.message}`,
-          }),
+        new MalformedResponse({
+          message: `generateObject: tool input failed schema decode: ${error.message}`,
         }),
     ),
   )

packages/llm/src/protocols/anthropic-messages.ts

@@ -19,7 +19,7 @@ import {
   type ToolResultPart,
 } from "../schema"
 import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
-import { isContextOverflow } from "../provider-error"
+import { classifyApiFailure } from "../provider-error"
 import * as Cache from "./utils/cache"
 import { Lifecycle } from "./utils/lifecycle"
 import { ToolSchemaProjection } from "./utils/tool-schema"
@@ -832,15 +832,11 @@ const providerErrorMessage = (event: AnthropicEvent): string => {
   return message || type || "Anthropic Messages stream error"
 }
 
-const onError = (state: ParserState, event: AnthropicEvent): StepResult => [
-  state,
-  [
-    LLMEvent.providerError({
-      message: providerErrorMessage(event),
-      classification: isContextOverflow(event.error?.message ?? "") ? "context-overflow" : undefined,
-    }),
-  ],
-]
+const onError = (event: AnthropicEvent) =>
+  classifyApiFailure({
+    message: providerErrorMessage(event),
+    code: event.error?.type,
+  })
 
 const step = (state: ParserState, event: AnthropicEvent) => {
   if (event.type === "message_start") return Effect.succeed(onMessageStart(state, event))
@@ -848,7 +844,7 @@ const step = (state: ParserState, event: AnthropicEvent) => {
   if (event.type === "content_block_delta") return onContentBlockDelta(state, event)
   if (event.type === "content_block_stop") return onContentBlockStop(state, event)
   if (event.type === "message_delta") return Effect.succeed(onMessageDelta(state, event))
-  if (event.type === "error") return Effect.succeed(onError(state, event))
+  if (event.type === "error") return Effect.fail(onError(event))
   return Effect.succeed<StepResult>([state, NO_EVENTS])
 }
 

packages/llm/src/protocols/bedrock-converse.ts

@@ -17,7 +17,7 @@ import {
   type ToolResultPart,
 } from "../schema"
 import { BedrockEventStream } from "./bedrock-event-stream"
-import { isContextOverflow } from "../provider-error"
+import { classifyApiFailure } from "../provider-error"
 import { JsonObject, optionalArray, ProviderShared } from "./shared"
 import { BedrockAuth } from "./utils/bedrock-auth"
 import { BedrockCache } from "./utils/bedrock-cache"
@@ -586,27 +586,20 @@ const step = (state: ParserState, event: BedrockEvent) =>
       return [{ ...state, pendingFinish: { reason: state.pendingFinish?.reason ?? "stop", usage } }, []] as const
     }
 
-    if (event.internalServerException || event.modelStreamErrorException || event.serviceUnavailableException) {
-      const message =
-        event.internalServerException?.message ??
-        event.modelStreamErrorException?.message ??
-        event.serviceUnavailableException?.message ??
-        "Bedrock Converse stream error"
-      return [state, [LLMEvent.providerError({ message })]] as const
-    }
-
-    if (event.validationException || event.throttlingException) {
-      const message =
-        event.validationException?.message ?? event.throttlingException?.message ?? "Bedrock Converse error"
-      return [
-        state,
-        [
-          LLMEvent.providerError({
-            message,
-            classification: event.validationException && isContextOverflow(message) ? "context-overflow" : undefined,
-          }),
-        ],
+    const exception = (
+      [
+        ["internalServerException", event.internalServerException],
+        ["modelStreamErrorException", event.modelStreamErrorException],
+        ["serviceUnavailableException", event.serviceUnavailableException],
+        ["throttlingException", event.throttlingException],
+        ["validationException", event.validationException],
       ] as const
+    ).find((entry) => entry[1] !== undefined)
+    if (exception) {
+      return yield* classifyApiFailure({
+        message: exception[1]?.message ?? "Bedrock Converse stream error",
+        code: exception[0],
+      })
     }
 
     return [state, []] as const

packages/llm/src/protocols/openai-responses.ts

@@ -19,7 +19,7 @@ import {
   type ToolResultPart,
 } from "../schema"
 import { JsonObject, optionalArray, optionalNull, ProviderShared } from "./shared"
-import { isContextOverflow } from "../provider-error"
+import { classifyApiFailure } from "../provider-error"
 import { OpenAIOptions } from "./utils/openai-options"
 import { Lifecycle } from "./utils/lifecycle"
 import { ToolSchemaProjection } from "./utils/tool-schema"
@@ -606,9 +606,9 @@ type StepResult = readonly [ParserState, ReadonlyArray<LLMEvent>]
 const NO_EVENTS: StepResult["1"] = []
 
 // `response.completed` / `response.incomplete` are clean finishes that emit a
-// `finish` event; `response.failed` is a hard failure that emits a
-// `provider-error`. All three end the stream — kept in one set so `step` and
-// the protocol's `terminal` predicate stay in sync.
+// `finish` event; `response.failed` is a hard failure that fails the stream
+// with a classified `LLMError`. All three end the stream — kept in one set so
+// `step` and the protocol's `terminal` predicate stay in sync.
 const TERMINAL_TYPES = new Set(["response.completed", "response.incomplete", "response.failed"])
 
 const onOutputTextDelta = (state: ParserState, event: OpenAIResponsesEvent): StepResult => {
@@ -907,24 +907,11 @@ const providerErrorMessage = (event: OpenAIResponsesEvent, fallback: string): st
   return message || code || fallback
 }
 
-const providerError = (event: OpenAIResponsesEvent, fallback: string) => {
-  const code = event.code || event.error?.code || event.response?.error?.code || undefined
-  const message = providerErrorMessage(event, fallback)
-  return LLMEvent.providerError({
-    message,
-    classification: code === "context_length_exceeded" || isContextOverflow(message) ? "context-overflow" : undefined,
+const providerError = (event: OpenAIResponsesEvent, fallback: string) =>
+  classifyApiFailure({
+    message: providerErrorMessage(event, fallback),
+    code: event.code || event.error?.code || event.response?.error?.code || undefined,
   })
-}
-
-const onResponseFailed = (state: ParserState, event: OpenAIResponsesEvent): StepResult => [
-  state,
-  [providerError(event, "OpenAI Responses response failed")],
-]
-
-const onError = (state: ParserState, event: OpenAIResponsesEvent): StepResult => [
-  state,
-  [providerError(event, "OpenAI Responses stream error")],
-]
 
 const step = (state: ParserState, event: OpenAIResponsesEvent) => {
   if (event.type === "response.output_text.delta") return Effect.succeed(onOutputTextDelta(state, event))
@@ -950,8 +937,8 @@ const step = (state: ParserState, event: OpenAIResponsesEvent) => {
   if (event.type === "response.output_item.done") return onOutputItemDone(state, event)
   if (event.type === "response.completed" || event.type === "response.incomplete")
     return Effect.succeed(onResponseFinish(state, event))
-  if (event.type === "response.failed") return Effect.succeed(onResponseFailed(state, event))
-  if (event.type === "error") return Effect.succeed(onError(state, event))
+  if (event.type === "response.failed") return Effect.fail(providerError(event, "OpenAI Responses response failed"))
+  if (event.type === "error") return Effect.fail(providerError(event, "OpenAI Responses stream error"))
   return Effect.succeed<StepResult>([state, NO_EVENTS])
 }
 

packages/llm/src/protocols/shared.ts

@@ -3,9 +3,9 @@ import { Effect, Schema, Stream } from "effect"
 import * as Sse from "effect/unstable/encoding/Sse"
 import { Headers, HttpClientRequest } from "effect/unstable/http"
 import {
-  InvalidProviderOutputReason,
-  InvalidRequestReason,
-  LLMError,
+  BadRequest,
+  MalformedResponse,
+  type LLMError,
   type ContentPart,
   type LLMRequest,
   type MediaPart,
@@ -88,11 +88,7 @@ export const sumTokens = (...values: ReadonlyArray<number | undefined>): number
 }
 
 export const eventError = (route: string, message: string, raw?: string) =>
-  new LLMError({
-    module: "ProviderShared",
-    method: "stream",
-    reason: new InvalidProviderOutputReason({ route, message, raw }),
-  })
+  new MalformedResponse({ route, message, raw })
 
 export const parseJson = (route: string, input: string, message: string) =>
   Effect.try({
@@ -252,15 +248,9 @@ export const sseFraming = (bytes: Stream.Stream<Uint8Array, LLMError>): Stream.S
  * Canonical invalid-request constructor. Lift one-line `const invalid =
  * (message) => invalidRequest(message)` aliases out of every
  * route so the error constructor lives in one place. If we ever extend
- * `InvalidRequestReason` with route context or trace metadata, the change
- * lands here.
+ * `BadRequest` with route context or trace metadata, the change lands here.
  */
-export const invalidRequest = (message: string) =>
-  new LLMError({
-    module: "ProviderShared",
-    method: "request",
-    reason: new InvalidRequestReason({ message }),
-  })
+export const invalidRequest = (message: string) => new BadRequest({ message })
 
 export const matchToolChoice = <Auto, None, Required, Tool>(
   route: string,

packages/llm/src/protocols/utils/tool-stream.ts

@@ -1,5 +1,5 @@
 import { Effect } from "effect"
-import { LLMError, LLMEvent, type ProviderMetadata, type ToolCall } from "../../schema"
+import { isLLMError, LLMEvent, type LLMError, type ProviderMetadata, type ToolCall } from "../../schema"
 import { eventError, parseToolInput, type ToolAccumulator } from "../shared"
 
 type StreamKey = string | number
@@ -95,7 +95,7 @@ const appendTool = <K extends StreamKey>(
 }
 
 export const isError = <K extends StreamKey>(result: AppendOutcome<K> | LLMError): result is LLMError =>
-  result instanceof LLMError
+  isLLMError(result)
 
 /**
  * Register a tool call whose start event arrived before any argument deltas.

packages/llm/src/provider-error.ts

@@ -1,5 +1,19 @@
-import { Schema } from "effect"
-import { LLMError, ProviderErrorEvent } from "./schema"
+import {
+  APIError,
+  Authentication,
+  BadRequest,
+  ContentPolicy,
+  ContextOverflow,
+  HttpContext,
+  HttpRateLimitDetails,
+  NotFound,
+  PermissionDenied,
+  ProviderMetadata,
+  QuotaExceeded,
+  RateLimit,
+  ServerError,
+  type LLMError,
+} from "./schema"
 
 const patterns = [
   /prompt is too long/i,
@@ -27,7 +41,102 @@ const patterns = [
 export const isContextOverflow = (message: string) =>
   patterns.some((pattern) => pattern.test(message)) || /^4(00|13)\s*(status code)?\s*\(no body\)/i.test(message)
 
-export const isContextOverflowFailure = (failure: unknown) =>
-  failure instanceof LLMError
-    ? failure.reason._tag === "InvalidRequest" && failure.reason.classification === "context-overflow"
-    : Schema.is(ProviderErrorEvent)(failure) && failure.classification === "context-overflow"
+const OVERFLOW_CODES = new Set(["context_length_exceeded", "model_context_window_exceeded"])
+const QUOTA_CODES = new Set(["insufficient_quota", "usage_not_included", "billing_error"])
+const QUOTA_TEXT = /insufficient[-_\s]?quota|quota[-_\s]?exceeded/i
+const CONTENT_POLICY_TEXT = /content[-_\s]?policy|content_filter|safety/i
+const SERVER_ERROR_STATUS = (status: number) => status >= 500 || status === 529
+
+const CODE_CLASSIFICATION: Record<string, (input: ApiFailure, common: CommonFields) => LLMError> = {
+  overloaded_error: serverError,
+  api_error: serverError,
+  server_error: serverError,
+  internal_error: serverError,
+  server_is_overloaded: serverError,
+  internalServerException: serverError,
+  serviceUnavailableException: serverError,
+  modelStreamErrorException: serverError,
+  rate_limit_error: rateLimit,
+  rate_limit_exceeded: rateLimit,
+  too_many_requests: rateLimit,
+  throttlingException: rateLimit,
+  authentication_error: (_input, common) => new Authentication(common),
+  permission_error: (_input, common) => new PermissionDenied(common),
+  not_found_error: (_input, common) => new NotFound(common),
+  invalid_request_error: (_input, common) => new BadRequest(common),
+  invalid_prompt: (_input, common) => new BadRequest(common),
+  validationException: (_input, common) => new BadRequest(common),
+}
+
+export interface ApiFailure {
+  readonly message: string
+  readonly status?: number | undefined
+  /** Provider machine-readable error code or type string (e.g. `context_length_exceeded`, `overloaded_error`). */
+  readonly code?: string | undefined
+  readonly retryAfterMs?: number | undefined
+  readonly rateLimit?: HttpRateLimitDetails | undefined
+  readonly requestID?: string | undefined
+  readonly http?: HttpContext | undefined
+  readonly providerMetadata?: ProviderMetadata | undefined
+}
+
+type CommonFields = {
+  readonly message: string
+  readonly status: number | undefined
+  readonly code: string | undefined
+  readonly requestID: string | undefined
+  readonly http: HttpContext | undefined
+  readonly providerMetadata: ProviderMetadata | undefined
+}
+
+function serverError(input: ApiFailure, common: CommonFields) {
+  return new ServerError({ ...common, retryAfterMs: input.retryAfterMs })
+}
+
+function rateLimit(input: ApiFailure, common: CommonFields) {
+  return new RateLimit({ ...common, retryAfterMs: input.retryAfterMs, rateLimit: input.rateLimit })
+}
+
+/**
+ * One classifier for every failure a remote API deliberately reports.
+ * Protocols call it with in-stream error payloads, the request executor with
+ * non-2xx responses, and the AI SDK adapter with `APICallError`s, so all
+ * three surfaces produce identical `LLMError` tags.
+ *
+ * Precedence: context overflow (most specific, 4xx-scoped), content policy,
+ * HTTP status, provider code, then the generic `APIError` fallback.
+ */
+export const classifyApiFailure = (input: ApiFailure): LLMError => {
+  const common: CommonFields = {
+    message: input.message,
+    status: input.status,
+    code: input.code,
+    requestID: input.requestID,
+    http: input.http,
+    providerMetadata: input.providerMetadata,
+  }
+  const body = input.http?.body ?? ""
+  const clientScoped = input.status === undefined || (input.status >= 400 && input.status < 500)
+  if (
+    clientScoped &&
+    ((input.code !== undefined && OVERFLOW_CODES.has(input.code)) ||
+      isContextOverflow(input.message) ||
+      (body.length > 0 && isContextOverflow(body)))
+  )
+    return new ContextOverflow(common)
+  if (CONTENT_POLICY_TEXT.test(body.length > 0 ? body : input.message)) return new ContentPolicy(common)
+  if (input.code !== undefined && QUOTA_CODES.has(input.code)) return new QuotaExceeded(common)
+  if (input.status === 401) return new Authentication(common)
+  if (input.status === 403) return new PermissionDenied(common)
+  if (input.status === 404) return new NotFound(common)
+  if (input.status === 429) {
+    if (QUOTA_TEXT.test(body.length > 0 ? body : input.message)) return new QuotaExceeded(common)
+    return rateLimit(input, common)
+  }
+  if (input.status !== undefined && SERVER_ERROR_STATUS(input.status)) return serverError(input, common)
+  if (input.status === 400 || input.status === 409 || input.status === 413 || input.status === 422)
+    return new BadRequest(common)
+  const byCode = input.code === undefined ? undefined : CODE_CLASSIFICATION[input.code]
+  if (byCode) return byCode(input, common)
+  return new APIError(common)
+}