diff --git a/internal/assembly/boot/agent_preset.go b/internal/assembly/boot/agent_preset.go
index 4ec14267ff..a9874e58a5 100644
--- a/internal/assembly/boot/agent_preset.go
+++ b/internal/assembly/boot/agent_preset.go
@@ -54,9 +54,7 @@ func CoreProviderToolNames() []string {
 		"read_file",
 		"edit_file",
 		"write_file",
-		"compress",
 		"recall",
-		"context_budget",
 		"use_capability",
 	}
 }
@@ -72,7 +70,6 @@ func HostControlToolNames() []string {
 		"update_goal",
 		"todo_write",
 		"complete_step",
-		"setting",
 	}
 }
 
@@ -158,7 +155,7 @@ func applyUnifiedProviderToolSurface(reg *tool.Registry, goalTurnsUnreachable bo
 	}
 	// advise, best_of_n, run_script, the peer tools and read_mcp_resource are
 	// registered only when config asks, at boot; present_files follows the same rule.
-	for _, name := range []string{advisor.Name, script.Name, "list_peers", "send_peer_message", "dispatch_to_project", "open_projects", "view_image", "present_files", plugin.ResourceReadToolName} {
+	for _, name := range []string{advisor.Name, script.Name, "list_peers", "send_peer_message", "dispatch_to_project", "open_projects", "view_image", plugin.ResourceReadToolName} {
 		if _, ok := reg.Get(name); ok {
 			allow = append(allow, name)
 		}
diff --git a/internal/assembly/boot/boot_test.go b/internal/assembly/boot/boot_test.go
index ccfa9ef408..73b20d6045 100644
--- a/internal/assembly/boot/boot_test.go
+++ b/internal/assembly/boot/boot_test.go
@@ -2500,12 +2500,13 @@ command = "reasonix-missing-mockmcp"
 	}
 	req := reqs[0]
 	assertSegmentedSurface(t, "light first request", toolSchemaNames(req.Tools))
-	for _, want := range []string{"compress", "use_capability", "read_file", "edit_file", "write_file", "bash", "ask"} {
+	for _, want := range []string{"use_capability", "read_file", "edit_file", "write_file", "bash", "ask"} {
 		if !requestHasTool(req, want) {
 			t.Fatalf("light first request missing tool %q; tools=%v", want, toolSchemaNames(req.Tools))
 		}
 	}
 	for _, forbidden := range []string{
+		"compress", "context_budget", "setting", "present_files", "browser_open",
 		"connect_tool_source", "web_fetch", "task", "read_only_task", "read_only_skill", "run_skill", "read_skill", "install_skill", "install_source",
 		"explore", "research", "review", "security_review",
 		"lsp_definition", "lsp_references", "lsp_hover", "lsp_diagnostics",
diff --git a/internal/assembly/boot/context_budget_effect_test.go b/internal/assembly/boot/context_budget_effect_test.go
index 27b6ac68c6..758f17202b 100644
--- a/internal/assembly/boot/context_budget_effect_test.go
+++ b/internal/assembly/boot/context_budget_effect_test.go
@@ -146,9 +146,12 @@ func TestEffectContextBudgetToolAnswersUnderPressure(t *testing.T) {
 	}
 }
 
-func TestEffectContextBudgetToolIsOffered(t *testing.T) {
+func TestEffectContextBudgetToolIsIndexedNotOffered(t *testing.T) {
 	reqs := effectRun(t, "boot-budget-tool-surface", "", ablation.Set{})
-	if !toolNames(reqs[0])["context_budget"] {
-		t.Fatalf("context_budget absent from the provider tool surface: %v", toolNames(reqs[0]))
+	if toolNames(reqs[0])["context_budget"] {
+		t.Fatalf("context_budget is in the first-round schema: %v", toolNames(reqs[0]))
+	}
+	if !strings.Contains(capabilityDescription(reqs[0]), "tool:context_budget") {
+		t.Fatalf("use_capability does not index context_budget: %s", capabilityDescription(reqs[0]))
 	}
 }
diff --git a/internal/assembly/boot/effect_present_files_test.go b/internal/assembly/boot/effect_present_files_test.go
index a1c408b730..b1457afd77 100644
--- a/internal/assembly/boot/effect_present_files_test.go
+++ b/internal/assembly/boot/effect_present_files_test.go
@@ -6,6 +6,7 @@ import (
 	"encoding/json"
 	"path/filepath"
 	"reflect"
+	"strings"
 	"testing"
 	"time"
 
@@ -135,24 +136,14 @@ func presentFilesEffect(t *testing.T, proxy bool, mode string) {
 			t.Fatal("disabled presenter leaked into the first provider request")
 		}
 	}
-	count = 0
+	// The presenter is deferred: never a schema of its own, indexed by
+	// use_capability's description exactly when it is enabled.
 	for _, schema := range requests[0].Tools {
-		if schema.Name != want.Name() {
-			continue
-		}
-		count++
-		var gotSchema, wantSchema any
-		if err := json.Unmarshal(schema.Parameters, &gotSchema); err != nil {
-			t.Fatal(err)
-		}
-		if err := json.Unmarshal(provider.CanonicalizeSchema(want.Schema()), &wantSchema); err != nil {
-			t.Fatal(err)
-		}
-		if schema.Description != want.Description() || !reflect.DeepEqual(gotSchema, wantSchema) {
-			t.Fatalf("unexpected provider contribution: %+v", schema)
+		if schema.Name == want.Name() {
+			t.Fatalf("present_files is in the first-round schema: %+v", schema)
 		}
 	}
-	if (count == 1) != enabled || count > 1 {
-		t.Fatalf("provider contribution = %d, enabled = %v", count, enabled)
+	if indexed := strings.Contains(capabilityDescription(requests[0]), "tool:"+want.Name()); indexed != enabled {
+		t.Fatalf("use_capability indexes present_files = %v, enabled = %v", indexed, enabled)
 	}
 }
diff --git a/internal/assembly/boot/effect_setting_tool_test.go b/internal/assembly/boot/effect_setting_tool_test.go
index 4597b214f0..6fdaa29c66 100644
--- a/internal/assembly/boot/effect_setting_tool_test.go
+++ b/internal/assembly/boot/effect_setting_tool_test.go
@@ -36,8 +36,8 @@ func (p *settingProvider) Stream(_ context.Context, req provider.Request) (<-cha
 	p.mu.Unlock()
 	ch := make(chan provider.Chunk, 3)
 	if first {
-		args, _ := json.Marshal(map[string]any{"op": "propose", "id": p.id, "value": json.RawMessage(p.val), "reason": "the task needs it"})
-		ch <- provider.Chunk{Type: provider.ChunkToolCall, ToolCall: &provider.ToolCall{ID: "call-set", Name: "setting", Arguments: string(args)}}
+		args, _ := json.Marshal(map[string]any{"action": "call", "capability_id": "tool:setting", "arguments": map[string]any{"op": "propose", "id": p.id, "value": json.RawMessage(p.val), "reason": "the task needs it"}})
+		ch <- provider.Chunk{Type: provider.ChunkToolCall, ToolCall: &provider.ToolCall{ID: "call-set", Name: "use_capability", Arguments: string(args)}}
 	} else {
 		ch <- provider.Chunk{Type: provider.ChunkText, Text: "done"}
 	}
@@ -53,7 +53,7 @@ func (p *settingProvider) toolResult() string {
 		return ""
 	}
 	for _, m := range p.reqs[1].Messages {
-		if m.Role == provider.RoleTool && m.Name == "setting" {
+		if m.Role == provider.RoleTool && m.Name == "use_capability" {
 			return m.Content
 		}
 	}
@@ -152,8 +152,8 @@ func TestEffectTheModelProposesASettingAndTheUserClickDecidesIt(t *testing.T) {
 			if !reflect.DeepEqual(rec.toolNames(0), rec.toolNames(1)) {
 				t.Fatalf("the tool block moved between requests:\n%v\n%v", rec.toolNames(0), rec.toolNames(1))
 			}
-			if !slices.Contains(rec.toolNames(0), "setting") {
-				t.Fatalf("setting is not offered: %v", rec.toolNames(0))
+			if slices.Contains(rec.toolNames(0), "setting") {
+				t.Fatalf("setting is in the first-round schema: %v", rec.toolNames(0))
 			}
 		})
 	}
diff --git a/internal/assembly/boot/testdata/golden/prefix_shape.json b/internal/assembly/boot/testdata/golden/prefix_shape.json
index 2d306983de..90650ce7ff 100644
--- a/internal/assembly/boot/testdata/golden/prefix_shape.json
+++ b/internal/assembly/boot/testdata/golden/prefix_shape.json
@@ -1,7 +1,7 @@
 {
   "SystemHash": "bc08dbc97c837bba",
-  "ToolsHash": "32d919766a6d1e78",
-  "PrefixHash": "dfa7eda18c0ad7c0",
+  "ToolsHash": "b779ff9e4407548b",
+  "PrefixHash": "fc00c9d2f68d114f",
   "LogRewriteVersion": 0,
-  "ToolSchemaTokens": 5032
+  "ToolSchemaTokens": 4305
 }
diff --git a/internal/assembly/boot/testdata/golden/provider_request.json b/internal/assembly/boot/testdata/golden/provider_request.json
index 6753ee0099..23e210e1f3 100644
--- a/internal/assembly/boot/testdata/golden/provider_request.json
+++ b/internal/assembly/boot/testdata/golden/provider_request.json
@@ -207,49 +207,6 @@
         "type": "object"
       }
     },
-    {
-      "name": "compress",
-      "description": "Compress a selected part of the current model-visible conversation without deleting visible history. Use only when the user explicitly asks for context compression. Choose `before` to summarize everything before the uniquely matched user turn while keeping that turn and later context, or `after` to summarize from that turn through the last completed turn while keeping the active turn. The anchor must be an exact, unique excerpt from a real user message; use a longer excerpt if the tool reports multiple matches.",
-      "parameters": {
-        "additionalProperties": false,
-        "properties": {
-          "anchor": {
-            "description": "An exact, unique excerpt from one real user message in the current model-visible conversation.",
-            "maxLength": 512,
-            "minLength": 1,
-            "type": "string"
-          },
-          "direction": {
-            "description": "Which side of the anchor user turn to compress.",
-            "enum": [
-              "before",
-              "after"
-            ],
-            "type": "string"
-          },
-          "focus": {
-            "description": "Optional guidance about facts or decisions the summary must preserve.",
-            "maxLength": 2000,
-            "type": "string"
-          }
-        },
-        "required": [
-          "anchor",
-          "direction"
-        ],
-        "type": "object"
-      }
-    },
-    {
-      "name": "context_budget",
-      "description": "Report how much context room is left before this conversation is automatically compacted. Nothing reports it unprompted: check it at the start of a long task, before a large change, and before output you may not finish reading (a broad search, a large file, a long build log), so you can narrow the command. `tokens_remaining` counts down to the compaction trigger, not the physical window.",
-      "parameters": {
-        "additionalProperties": false,
-        "properties": {},
-        "required": [],
-        "type": "object"
-      }
-    },
     {
       "name": "edit_file",
       "description": "Replace an exact string in a file with another. old_string must occur exactly once; add surrounding context to disambiguate. Use for targeted edits instead of rewriting the whole file.",
@@ -292,41 +249,6 @@
         "type": "object"
       }
     },
-    {
-      "name": "present_files",
-      "description": "Present existing final files as clickable cards, including shell-created documents. Paths must be in the workspace or authorized folders. Opens current files, not snapshots; does not certify correctness.",
-      "parameters": {
-        "additionalProperties": false,
-        "properties": {
-          "files": {
-            "items": {
-              "additionalProperties": false,
-              "properties": {
-                "description": {
-                  "maxLength": 256,
-                  "type": "string"
-                },
-                "path": {
-                  "maxLength": 4096,
-                  "type": "string"
-                }
-              },
-              "required": [
-                "path"
-              ],
-              "type": "object"
-            },
-            "maxItems": 4,
-            "minItems": 1,
-            "type": "array"
-          }
-        },
-        "required": [
-          "files"
-        ],
-        "type": "object"
-      }
-    },
     {
       "name": "read_file",
       "description": "Read a text file with optional line offset/limit. Output prefixes each line with its 1-based number (e.g. `   42→...`) so subsequent edit_file calls can target exact lines. Use `offset` and `limit` to page through large files; the tool reports total length and pagination hints in a trailer.",
@@ -384,42 +306,6 @@
         "type": "object"
       }
     },
-    {
-      "name": "setting",
-      "description": "Find and change the user's own settings. op=list finds settings by words of their id (for example \"peers dispatch\"). op=propose asks the user to confirm one change: give the id, the new value and a short reason; nothing changes until the user answers, and the result says what changed or that they declined. op=open points the user at a setting that only they may change. Use it when a refusal ends with (fix: setting ...), or when a tool you need is not offered because a switch is off. Do not change a setting the task did not call for.",
-      "parameters": {
-        "additionalProperties": false,
-        "properties": {
-          "id": {
-            "description": "The setting's id, for propose and open",
-            "type": "string"
-          },
-          "op": {
-            "enum": [
-              "list",
-              "propose",
-              "open"
-            ],
-            "type": "string"
-          },
-          "query": {
-            "description": "Words of the setting's id, for op=list",
-            "type": "string"
-          },
-          "reason": {
-            "description": "Why the change helps the user, shown on the card, for op=propose",
-            "type": "string"
-          },
-          "value": {
-            "description": "The new value as JSON, for op=propose"
-          }
-        },
-        "required": [
-          "op"
-        ],
-        "type": "object"
-      }
-    },
     {
       "name": "todo_write",
       "description": "Record and update a structured task list for the current work. Send the COMPLETE list every call — it replaces the previous one. Use it to plan multi-step work and show progress: keep exactly one item in_progress at a time, and flip an item to completed the moment it's done (don't batch completions). Skip it for trivial single-step tasks. The list is two-level: a `level` 0 item is a PHASE (a milestone) and the `level` 1 items after it are its concrete sub-steps. COPY `step_id` VERBATIM for every item that already has one: it is how a completion stays attached to its step when you retitle it, insert a step above it, or reorder the list. Give a new item a fresh unique id (e.g. \"plan_step_07\"); never reuse or renumber an existing one.",
@@ -530,7 +416,7 @@
     },
     {
       "name": "use_capability",
-      "description": "Search and call deferred tools, skills, and MCP through stable IDs such as tool:grep, skill:review, task:subagent, and mcp-tool:server/tool. Before saying a tool, network, or integration is unavailable, search by natural-language goal; results include ranked IDs and available schemas. list enumerates the catalog, inspect loads one definition, call invokes it, and decline records why a preferred capability is skipped. task:subagent delegates 1-8 tasks that run in parallel. Calls retain all permission, sandbox, and execution checks. Deferred tools, called by id: tool:web_fetch (Read HTTP(S) text.); tool:browser_open (Load a page and return its snapshot, one element per line with [eN]…); tool:browser_read (Read a browser page: snapshot); tool:browser_act (Operate a page with real input.); tool:grep (Search for a regular expression in a file, or recursively under a…); tool:glob (Find files matching a glob pattern); tool:ls (List the entries of a directory.). Read JSON or static text with tool:web_fetch; open a browser only for pages that need rendering, clicks or a login.",
+      "description": "Search and call deferred tools, skills, and MCP through stable IDs such as tool:grep, skill:review, task:subagent, and mcp-tool:server/tool. Before saying a tool, network, or integration is unavailable, search by natural-language goal; results include ranked IDs and available schemas. list enumerates the catalog, inspect loads one definition, call invokes it, and decline records why a preferred capability is skipped. task:subagent delegates 1-8 tasks that run in parallel. Calls retain all permission, sandbox, and execution checks. Deferred tools, called by id: tool:web_fetch (Read HTTP(S) text.); tool:browser_open (Load a page and return its snapshot, one element per line with [eN]…); tool:browser_read (Read a browser page: snapshot); tool:browser_act (Operate a page with real input.); tool:grep (Search for a regular expression in a file, or recursively under a…); tool:glob (Find files matching a glob pattern); tool:ls (List the entries of a directory.); tool:present_files (Present existing final files as clickable cards, including…); tool:setting (Find and change the user's own settings.); tool:compress (Compress a selected part of the current model-visible conversation…); tool:context_budget (Report how much context room is left before this conversation is…). Read JSON or static text with tool:web_fetch; open a browser only for pages that need rendering, clicks or a login.",
       "parameters": {
         "properties": {
           "action": {
diff --git a/internal/assembly/boot/testdata/golden/tool_schemas.json b/internal/assembly/boot/testdata/golden/tool_schemas.json
index 000b00721c..9e1d5b636f 100644
--- a/internal/assembly/boot/testdata/golden/tool_schemas.json
+++ b/internal/assembly/boot/testdata/golden/tool_schemas.json
@@ -182,51 +182,6 @@
       "type": "object"
     }
   },
-  {
-    "Name": "compress",
-    "Description": "Compress a selected part of the current model-visible conversation without deleting visible history. Use only when the user explicitly asks for context compression. Choose `before` to summarize everything before the uniquely matched user turn while keeping that turn and later context, or `after` to summarize from that turn through the last completed turn while keeping the active turn. The anchor must be an exact, unique excerpt from a real user message; use a longer excerpt if the tool reports multiple matches.",
-    "ReadOnly": true,
-    "Schema": {
-      "additionalProperties": false,
-      "properties": {
-        "anchor": {
-          "description": "An exact, unique excerpt from one real user message in the current model-visible conversation.",
-          "maxLength": 512,
-          "minLength": 1,
-          "type": "string"
-        },
-        "direction": {
-          "description": "Which side of the anchor user turn to compress.",
-          "enum": [
-            "before",
-            "after"
-          ],
-          "type": "string"
-        },
-        "focus": {
-          "description": "Optional guidance about facts or decisions the summary must preserve.",
-          "maxLength": 2000,
-          "type": "string"
-        }
-      },
-      "required": [
-        "anchor",
-        "direction"
-      ],
-      "type": "object"
-    }
-  },
-  {
-    "Name": "context_budget",
-    "Description": "Report how much context room is left before this conversation is automatically compacted. Nothing reports it unprompted: check it at the start of a long task, before a large change, and before output you may not finish reading (a broad search, a large file, a long build log), so you can narrow the command. `tokens_remaining` counts down to the compaction trigger, not the physical window.",
-    "ReadOnly": true,
-    "Schema": {
-      "additionalProperties": false,
-      "properties": {},
-      "required": [],
-      "type": "object"
-    }
-  },
   {
     "Name": "edit_file",
     "Description": "Replace an exact string in a file with another. old_string must occur exactly once; add surrounding context to disambiguate. Use for targeted edits instead of rewriting the whole file.",
@@ -271,42 +226,6 @@
       "type": "object"
     }
   },
-  {
-    "Name": "present_files",
-    "Description": "Present existing final files as clickable cards, including shell-created documents. Paths must be in the workspace or authorized folders. Opens current files, not snapshots; does not certify correctness.",
-    "ReadOnly": true,
-    "Schema": {
-      "additionalProperties": false,
-      "properties": {
-        "files": {
-          "items": {
-            "additionalProperties": false,
-            "properties": {
-              "description": {
-                "maxLength": 256,
-                "type": "string"
-              },
-              "path": {
-                "maxLength": 4096,
-                "type": "string"
-              }
-            },
-            "required": [
-              "path"
-            ],
-            "type": "object"
-          },
-          "maxItems": 4,
-          "minItems": 1,
-          "type": "array"
-        }
-      },
-      "required": [
-        "files"
-      ],
-      "type": "object"
-    }
-  },
   {
     "Name": "read_file",
     "Description": "Read a text file with optional line offset/limit. Output prefixes each line with its 1-based number (e.g. `   42→...`) so subsequent edit_file calls can target exact lines. Use `offset` and `limit` to page through large files; the tool reports total length and pagination hints in a trailer.",
@@ -366,43 +285,6 @@
       "type": "object"
     }
   },
-  {
-    "Name": "setting",
-    "Description": "Find and change the user's own settings. op=list finds settings by words of their id (for example \"peers dispatch\"). op=propose asks the user to confirm one change: give the id, the new value and a short reason; nothing changes until the user answers, and the result says what changed or that they declined. op=open points the user at a setting that only they may change. Use it when a refusal ends with (fix: setting ...), or when a tool you need is not offered because a switch is off. Do not change a setting the task did not call for.",
-    "ReadOnly": true,
-    "Schema": {
-      "additionalProperties": false,
-      "properties": {
-        "id": {
-          "description": "The setting's id, for propose and open",
-          "type": "string"
-        },
-        "op": {
-          "enum": [
-            "list",
-            "propose",
-            "open"
-          ],
-          "type": "string"
-        },
-        "query": {
-          "description": "Words of the setting's id, for op=list",
-          "type": "string"
-        },
-        "reason": {
-          "description": "Why the change helps the user, shown on the card, for op=propose",
-          "type": "string"
-        },
-        "value": {
-          "description": "The new value as JSON, for op=propose"
-        }
-      },
-      "required": [
-        "op"
-      ],
-      "type": "object"
-    }
-  },
   {
     "Name": "todo_write",
     "Description": "Record and update a structured task list for the current work. Send the COMPLETE list every call — it replaces the previous one. Use it to plan multi-step work and show progress: keep exactly one item in_progress at a time, and flip an item to completed the moment it's done (don't batch completions). Skip it for trivial single-step tasks. The list is two-level: a `level` 0 item is a PHASE (a milestone) and the `level` 1 items after it are its concrete sub-steps. COPY `step_id` VERBATIM for every item that already has one: it is how a completion stays attached to its step when you retitle it, insert a step above it, or reorder the list. Give a new item a fresh unique id (e.g. \"plan_step_07\"); never reuse or renumber an existing one.",
@@ -515,7 +397,7 @@
   },
   {
     "Name": "use_capability",
-    "Description": "Search and call deferred tools, skills, and MCP through stable IDs such as tool:grep, skill:review, task:subagent, and mcp-tool:server/tool. Before saying a tool, network, or integration is unavailable, search by natural-language goal; results include ranked IDs and available schemas. list enumerates the catalog, inspect loads one definition, call invokes it, and decline records why a preferred capability is skipped. task:subagent delegates 1-8 tasks that run in parallel. Calls retain all permission, sandbox, and execution checks. Deferred tools, called by id: tool:web_fetch (Read HTTP(S) text.); tool:browser_open (Load a page and return its snapshot, one element per line with [eN]…); tool:browser_read (Read a browser page: snapshot); tool:browser_act (Operate a page with real input.); tool:grep (Search for a regular expression in a file, or recursively under a…); tool:glob (Find files matching a glob pattern); tool:ls (List the entries of a directory.). Read JSON or static text with tool:web_fetch; open a browser only for pages that need rendering, clicks or a login.",
+    "Description": "Search and call deferred tools, skills, and MCP through stable IDs such as tool:grep, skill:review, task:subagent, and mcp-tool:server/tool. Before saying a tool, network, or integration is unavailable, search by natural-language goal; results include ranked IDs and available schemas. list enumerates the catalog, inspect loads one definition, call invokes it, and decline records why a preferred capability is skipped. task:subagent delegates 1-8 tasks that run in parallel. Calls retain all permission, sandbox, and execution checks. Deferred tools, called by id: tool:web_fetch (Read HTTP(S) text.); tool:browser_open (Load a page and return its snapshot, one element per line with [eN]…); tool:browser_read (Read a browser page: snapshot); tool:browser_act (Operate a page with real input.); tool:grep (Search for a regular expression in a file, or recursively under a…); tool:glob (Find files matching a glob pattern); tool:ls (List the entries of a directory.); tool:present_files (Present existing final files as clickable cards, including…); tool:setting (Find and change the user's own settings.); tool:compress (Compress a selected part of the current model-visible conversation…); tool:context_budget (Report how much context room is left before this conversation is…). Read JSON or static text with tool:web_fetch; open a browser only for pages that need rendering, clicks or a login.",
     "ReadOnly": true,
     "Schema": {
       "properties": {
diff --git a/internal/assembly/boot/tool_contract_surface_test.go b/internal/assembly/boot/tool_contract_surface_test.go
index 8f907662c2..79c6c580b8 100644
--- a/internal/assembly/boot/tool_contract_surface_test.go
+++ b/internal/assembly/boot/tool_contract_surface_test.go
@@ -18,13 +18,9 @@ func stableBootToolNames() []string {
 		"ask",
 		"await_user",
 		"bash",
-		"compress",
-		"context_budget",
 		"edit_file",
-		"present_files",
 		"read_file",
 		"recall",
-		"setting",
 		"todo_write",
 		"use_capability",
 		"write_file",
diff --git a/internal/runtime/usecap/deferred_index.go b/internal/runtime/usecap/deferred_index.go
index 9824350a98..31cc2adbb3 100644
--- a/internal/runtime/usecap/deferred_index.go
+++ b/internal/runtime/usecap/deferred_index.go
@@ -9,7 +9,7 @@ import (
 // deferredIndexTools are the built-ins a model reaches for by name before it
 // searches: hidden from the provider schema, so without a line here a
 // directly visible browser is the only way it can see to read a URL.
-var deferredIndexTools = []string{"web_fetch", "web_search", "browser_open", "browser_read", "browser_act", "grep", "glob", "ls"}
+var deferredIndexTools = []string{"web_fetch", "web_search", "browser_open", "browser_read", "browser_act", "grep", "glob", "ls", "present_files", "setting", "compress", "context_budget"}
 
 const deferredLeadBytes = 72