dsh-plugin-rea

English · 简体中文

Which version is this? 2.x — and main is it. 2.x provisions the runtime itself (Node, the REA CLI, a JDK, Ghidra) and then wires the row: install the plugin, run scripts/setup.mjs once, restart dsh. 1.x is the older configuration-only adapter — it expects rea on PATH and Ghidra installed by you — kept on the v1 branch and on npm as [email protected]. Full comparison: Two lines: 1.x and 2.x · upgrading: Upgrading from 1.x.

REA 6.8.0 — this package's pin — still accepts Ghidra 12.1.x only (12.2 is unsupported, and the whole Ghidra decompile/analysis group disappears) and the jadx bridge 0.7.1 only (0.8.0 fails on its first call with Expected engine 0.7.1, before any APK is loaded). Both gates were re-checked against 6.8.0 on 2026-10-26 and neither one moved. JDK 21.0.13 is out upstream, but Adoptium publishes no Windows build of it, so there is nothing to point at here — and it drops only three of the four eMudhra anchors. The trade-off is therefore "all tools green, CVE still open": Ghidra 12.1.4's CVE-2026-100504 / 100505 / 100503 / 113604 stay in the chain — do not analyse untrusted samples with this environment — and the JDK still trusts the eMudhra anchors. Evidence, mitigations and revisit triggers: docs/update-review-2026-10-24.md and docs/update-review-2026-10-26.md.

Ghidra 12.2 itself works — measured 2026-10-26: same 9 segments, same 341 procedures, byte-identical entry pseudocode as 12.1.4, through REA's own bridge. What blocks it is REA's version-equality check, not the bridge. If you want the CVE-fixed engine anyway, node scripts/ghidra-gate-override.mjs --enable --version 12.2 rewrites that one constant in the vendored copy, records what it changed in <bundle>/ghidra-gate-override.json, and --disable restores it; re-running setup reverts it too. It needs JDK 25+ (application.java.min=25).

You must run node scripts/setup.mjs yourself — once after installing, and once after every plugin upgrade. Nothing runs it for you: a market install or update replaces the package and stops there. Since 2.0.8 a market upgrade keeps the row working (the mirrored launcher re-execs the new one), but it does not move the bundle to the new pins — REA and the engines stay at the versions recorded by the last setup run. Skip it on a first install and the row exits MODULE_NOT_FOUND with no tools and no message; skip it after an upgrade and you keep running the old REA silently (check-updates.ps1 reports the drift). If you had enabled the Ghidra gate override (see §9 of the 2026-10-26 review), that setup run re-provisions the bundled REA and the patch goes with it — re-enable it afterwards, or accept the pinned Ghidra again. The full rule set is under Updating.

A first install and every plugin upgrade also show a one-time guide in the Web GUI. It says whether the runtime is missing or behind this package's pin, shows the exact node "<install path>/scripts/setup.mjs" command for that install, and remembers the dismissal per plugin version. It is a notice only: it never runs setup for you and never edits the bundle. Run it once, restart dsh, and the guide stays away until the next plugin version.

Brings the upstream REA — Reverse Engineer Anything MCP server into DeepSeek Harness (dsh), so its tools show up in a session as mcp__rea__*.

What this is

A Cordis bundle with a provisioning layer. package.json declares dsh.bundle.patch, and that patch inserts one row of the official MCP client plugin (@deepseek-ai/dsh-mcp-client). The row launches bin/rea-shim.mjs with the running Node executable; the shim reads the machine-local state that scripts/setup.mjs wrote and starts the REA CLI it provisioned. There is no analysis code here, and the published patch contains no machine path.

Third-party adapter. This repository is not part of REA, ships none of REA's code, and is not maintained by REA's authors. It composes one configuration row and fetches upstream's own releases. Upstream: github.com/morluto/rea (MIT) · docs · rea.tools · npm package rea-agents. REA's own problems go to its tracker; this plugin's problems go to this repository's issues.

It adds exactly three things: the tool registration under the mcp__rea__* names, the per-machine engine environment (GHIDRA_INSTALL_DIR, JAVA_HOME, REA_ADB_PATH, REA_APKTOOL_COMMAND, REA_JADX_MCP_JAR, REA_BROWSER_EXECUTABLE, …) scoped to that MCP child instead of the whole harness, and the provisioning that puts those components on disk.

What it can and cannot do

The analysis is upstream's; this section only restates upstream's own contract so expectations are right.

• It cannot recover complete source code. For native binaries upstream returns pseudocode and assembly, for Android APKs decompiled methods, for JEB units pseudocode, and for JavaScript/Electron modules and their relations. Upstream's tool contract states the pseudocode "is decompiler output, not original source and not text-equivalent" to another provider's output.
• Not recoverable in practice: original identifier names, types, comments, macros and template expansion, build scripts and assets. Optimisation (inlining, reordering), symbol stripping, obfuscation, packing and VM protection only make the output worse.
• Every result carries confidence and limitations (observed / derived / inferred, plus an explicit limitations list). Read them as evidence, never as recovered source.
• Good for: understanding behaviour, locating an implementation, interoperability and compatible re-implementation, security review, migration reference. To obtain actual source code, use the legitimate route — the upstream repository, vendor authorisation, or the vendor's own SDK.

What provisioning cannot buy you. These are upstream host gates, not settings this plugin forgot:

Capability	Gate
Hopper provider	macOS only
IDA provider	needs your own IDA and REA_IDA_MCP_CONFIG; REA never launches IDA
firmware binwalk / unblob	Linux only
Wakaru JS recovery, pwntools ELF/recorded-crash paths	Linux x64 only
pwndbg / gdb core context	Linux, macOS
native mitmproxy captures	Linux (REA_MITMDUMP_COMMAND); HAR needs no engine
Apple toolchain (Mach-O, plist, asset catalog, LLDB)	macOS only
capture_process_scenario	Linux/macOS PTY
apktool (Android resource decoding)	available only when REA_APKTOOL_COMMAND points at something Node can execute directly: upstream execFiles that command without a shell, and Windows cannot execute a .bat/.cmd that way (EINVAL). This bundle wires the official apktool.bat as upstream documents, so on Windows the launcher reports no version and the capability stays unavailable; JADX (a jar) is unaffected

Two lines: 1.x and 2.x

	2.x — this line, main	1.x — v1
What it is	the adapter plus provisioning: it prepares Node, the REA CLI, a JDK and Ghidra, then wires the row	configuration-only adapter: one MCP row pointing at a rea you installed yourself
What you need first	nothing for the runtime — provisioning fetches what is missing (with --reuse it adopts an existing Ghidra/JDK instead)	Node.js on REA's engines, npm i -g rea-agents, and Ghidra 12.1.x for native analysis
First run	run scripts/setup.mjs (or the bootstrap) once, then restart dsh	nothing to run
The row launches	the running Node executable plus <DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs, which reads that machine's state	cmd.exe /c rea mcp on Windows, rea mcp on macOS/Linux
Machine state	<DSH_HOME>/plugins/rea-bundle/ — bundle.json, vendor/, cache/	your own REA and Ghidra installs
Install	dsh plugin --profile <profile> add dsh-plugin-rea (npm latest ≥ 2.0.0)	… add [email protected], or github:czcdr/dsh-plugin-rea#v1.0.6

Both lines ship the same patch shape — one insert row of @deepseek-ai/dsh-mcp-client — and offer the same mcp__rea__* surface; they differ in who prepares the runtime. Moving from 1.x to 2.x takes two commands: see Upgrading from 1.x.

Upstream also registers itself with dsh now (rea setup --client dsh); that route is essentially the 1.x model maintained upstream — configuration only, your own Node/REA/Ghidra. This bundle is the zero-dependency route. Pick one of the three, and never keep two serverName: rea rows at once.

Install

Nothing to install first. On a bare Windows x64 machine:

powershell -ExecutionPolicy Bypass -File scripts\rea-bundle-bootstrap.ps1 -Reuse

-Reuse adopts a Ghidra/JDK you already have; omit it to download both (about 775 MB from the official releases, checksum-verified). Then install the plugin for your profile:

dsh plugin --profile <profile> add dsh-plugin-rea
# or pin this release instead of tracking npm latest:
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea#v2.0.16

On an Electron host (the DeepSeek Harness desktop app) the profile is owned by the application. dsh plugin --profile desktop … is refused by the CLI (profile "desktop" is managed exclusively by the Electron application) — install or update through the in-app plugin manager instead. The CLI form above is for a profile you created yourself with the CLI under another name.

If Node is already installed, you can skip the bootstrap and run provisioning directly:

node scripts/setup.mjs --reuse      # or --dry-run to see the plan first

Node, when you provide it instead of letting setup fetch one, must satisfy REA's engines — ^22.23.4 || ^24.22.1 || >=26.12.1 (check with node -v). REA raised those floors in 6.8.0: the pinned 24.22.1 is now the oldest acceptable 24.x, so a 24.11–24.21 runtime that 6.7.0 accepted is rejected here. A lower Node does not fail at install time; the MCP row fails later, which reads as silence.

Installing the plugin first also works, but it takes two steps. The row resolves to <DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs, and that file is created by setup — so until the first provisioning run the row exits with MODULE_NOT_FOUND, and nothing is there to start a background provisioning (the harness itself is unaffected: failOnStartupError: false is deliberate). Run node scripts/setup.mjs once; from then on the shim is in place, and every later missing state is picked up by the shim itself, which provisions in the background.

Which DSH versions

package.json declares engines.dsh: ">=0.1.7-rc.2 <0.3.0-0", and that declaration is the answer the plugin market reads: with it the card says compatible; without it the market can only say unknown. The range is not a guess — it is the dsh release list this plugin is checked against, and node scripts/host-compat.mjs fails the build when the two disagree:

Host	Covered
0.1.7-rc.2 … 0.2.2-rc.1 — every release in between, alpha and rc (16 tags)	yes
0.3.0 and later	no — a new minor line is a host nobody updated this row for yet

2.0.14 checked the six tags published after 2026-10-23 individually (0.2.1-rc.3, 0.2.1-rc.4, 0.2.2-alpha.2, 0.2.1-rc.5, 0.2.2-alpha.3, 0.2.2-rc.1): none of their release notes touch the five surfaces this plugin depends on (patch dialect, MCP row schema, host plugin API, client injection contract, market engines verdict). The one adjacent item is 0.2.2-rc.1's Cordis Core 4.1.0 / Loader 1.1.0, and it was measured on this machine: GET /dsh-plugin-rea/guide-state answered 200 and the mcp__rea__* tools were available.

What was actually compared across that range is the row provider: packages/mcp/mcp-client at dsh-v0.1.7-rc.2, dsh-v0.2.1-rc.2 and dsh-v0.2.2-rc.1 exposes the same config surface the shipped row uses (serverName, transport, command, args, env, failOnStartupError), and dsh.bundle.patch exists at all three. A host outside the range can still work; it is simply not claimed.

The market reads engines.dsh only, and not engines.node (measured in dshmarket/lib/verify.js and discovery-compatibility.js). A host whose bundled Node sits below this package's engines.node floor is therefore not blocked by the market; only someone running npm with engine-strict=true is stopped at install time. Run time is unaffected either way: REA is started with the paths.node recorded in bundle.json, not the host's Node.

A REA you already have is used as-is

2.x exists to supply what a machine is missing. If REA is already installed — rea on PATH, the 1.x setup — the row runs that, exactly as the 1.x line did, and nothing is provisioned:

What the machine has	What the row runs
rea on PATH, at least as new as the provisioned build	your REA (rea mcp), with your Node and your PATH
rea on PATH but older than the provisioned build	the provisioned bundle — a stale global CLI would take the tool count backwards (6.4.0 ships 170 tools, 6.7.0 ships 207, 6.8.0 ships 229)
no rea on PATH	the provisioned bundle, provisioning itself on first use

• REA_BUNDLE_LAUNCH=auto|system|bundle overrides the decision (auto is the default); REA_BUNDLE_SYSTEM_REA=<path> names a binary that is not on PATH.
• The probe is the CLI's own --version, bounded at 8 s: a .cmd shim on Windows takes ~2.4 s to answer, so the check runs once per start and only when a rea actually exists.
• --shim-status prints the decision — launch.mode, launch.reason, launch.systemRea — and its launch.rejected list says why a candidate was skipped ("exit 7", "inside the bundle", "timeout").
• A delegated child carries REA_BUNDLE_DELEGATED=1, so a wrapper that calls this shim cannot make it delegate to itself forever.
• On Windows a forced kill of the shim does not reap the delegated REA. There is no job object here: Task Manager or a supervisor terminating the shim leaves your rea running with the row's pipes (a console Ctrl+C reaches the whole process group and does stop it). On POSIX the child is in the same group and dies with the terminal.

Configure — nothing

The shipped row carries no machine path:

- insert:
    - id: mcp-rea
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: rea
        transport: stdio
        command: !!js process.execPath
        args:
          - !!js "(process.env.DSH_HOME && process.env.DSH_HOME.trim() !== '' ? process.env.DSH_HOME : (process.env.USERPROFILE || ((process.env.HOMEDRIVE || '') + (process.env.HOMEPATH || '')) || process.env.HOME || '') + '/.dsh') + '/plugins/rea-bundle/bin/rea-shim.mjs'"
          - mcp
        env:
          ELECTRON_RUN_AS_NODE: '1'
        failOnStartupError: false

Three details in that row are load-bearing, and all three were bugs first:

• The !!js scalar is quoted. A bare scalar containing ? and : is not one scalar: YAML reads it as an explicit key/value pair, so the loader receives a mapping where an expression belongs. Measured against the harness's own dialect (JSON_SCHEMA plus the !!js type from app.asar), the unquoted form made args[0] into {"[object Object]": "..."} — an object the row cannot launch with. Quoting removes the ambiguity, and the JS string literals inside then use single quotes.
• Forward slashes, never backslashes. A single-quoted YAML scalar keeps \ literal, and JavaScript then consumes it: '\plugins' is not a path but the BEL control character, '\rea-bundle' becomes a carriage return and '\bin' a backspace — the path is destroyed silently. Forward slashes work on Windows and on POSIX.
• The DSH_HOME fallback is <home>/.dsh, not <home>. The shim, setup.mjs and bootstrap.ps1 all resolve an unset DSH_HOME to <home>/.dsh/...; a fallback that stopped one level higher made the row point at a file provisioning never writes — invisible on any machine where DSH_HOME happens to be set.
• The home fallback mirrors os.homedir(): USERPROFILE, then HOMEDRIVE+HOMEPATH, then HOME. The middle pair is what a Windows service account gets instead of USERPROFILE; an expression that skipped it produced /.dsh/... — the root of the current drive — for exactly the accounts that cannot read that path. Only when all four are empty (a container account with no home at all) does the concatenation land on /.dsh/..., and failOnStartupError: false keeps that quiet, so the only symptom is "no mcp__rea__* tools". Set DSH_HOME explicitly on such an account.
• No path/require in a !!js expression: ctx is the loader scope, so the DSH_HOME fallback is written as plain concatenation.

npm run bundle:audit enforces the first three: it evaluates this exact expression with DSH_HOME unset, empty and set, and rejects a bare scalar that YAML would read as a mapping.

ELECTRON_RUN_AS_NODE only matters if the harness itself runs on Electron; it is harmless for a plain Node host, and stating it explicitly beats relying on it surviving the child-env scrub.

command is the running Node executable and args[0] is computed from DSH_HOME, so the same patch works on Windows, macOS and Linux without a PATH entry. Everything machine-specific lives in <DSH_HOME>/plugins/rea-bundle/bundle.json, written by setup and readable with:

node bin/rea-shim.mjs --shim-status      # resolved paths, versions, verification state
node bin/rea-shim.mjs --shim-print-env   # the environment the MCP child will see

Both commands answer with the same bundle whether you run them from the plugin directory (the checkout copy) or from <DSH_HOME>/plugins/rea-bundle (the mirrored copy): the checkout resolves the location by the same rule setup uses, while the mirror uses its own position — it cannot rely on DSH_HOME, which the harness strips from an MCP child's environment.

Setting REA_BUNDLE_AUTO_PROVISION=0 disables the first-run background provisioning.

To pin a different launch command on this machine (for example your own globally installed rea), override the row from the profile's own cordis.patch.yml — that file is machine-local and never published. Only one row may use serverName: rea in a profile.

A patch replaces the row's config wholesale — it does not deep-merge — so a partial override such as config: {env: {...}} also deletes command, args and serverName, and the row stops starting. Supply the complete config, or use REA_BUNDLE_LAUNCH=system (above), which needs no patch at all.

Ghidra and the JDK are not read from your environment at launch: setup.mjs records their absolute paths in bundle.json and the shim exports GHIDRA_INSTALL_DIR / JAVA_HOME from that record — which is why a non-standard install location works without touching the row. To adopt an existing install, point GHIDRA_INSTALL_DIR / JAVA_HOME at it and run node scripts/setup.mjs --reuse: the manifest's probe checks the candidate and falls back to a download when it does not match, so a wrong path is reported during setup instead of failing later inside a tool call.

One REA registration only

Upstream REA 6.5.0 can register itself with DSH:

rea setup --client dsh

That writes a row with id: rea-mcp and serverName: rea into <DSH_HOME>/cordis.patch.yml. This bundle's row is id: mcp-rea with the same serverName, and DSH rejects the later of two entries that share a server name — so which REA row survives depends on load order. Use one, not both.

Which one? Upstream's is the configuration-only route: it expects Node on REA's engines, rea-agents reachable, and a Ghidra/JDK you provisioned yourself. This bundle is the zero-dependency route: it fetches Node, REA, JDK and Ghidra on first run. The patch shape and the mcp__rea__* surface are the same either way.

The shim reports the conflict on startup — setup.mjs records the DSH home it ran under in mirror.json, and the mirrored copy reads that when the harness has stripped DSH_* from its environment; node bin/rea-shim.mjs --shim-status lists it under duplicateRegistration.

Verify

Ask for a mcp__rea__* call — binary_session reports the engine checks (configuration / installation / headless / native_decompiler / java), all of which should be ok: true. Then open_binary a PE, binary_overview, and procedure_pseudo_code.

Provisioning itself ends with three checks: rea doctor --json (reported, with unmet checks named), a Ghidra smoke import that runs rea analyze on the JDK's own bin/java and fails the run when it cannot import (--skip-smoke omits it), and the exact shim launch the MCP row will use. The smoke step exists because doctor validates paths, versions and which Java would be chosen — it never imports a binary, so a broken bridge or a mismatched native addon used to surface only as an MCP timeout.

The first analysis of each new target is slow. Measured here: 36 s for a small PE, 64 s for a 6 MB DLL — every import starts a JVM and runs Ghidra's default auto-analysis. If the first open_binary → list_segments on a large file times out, that is this cost and not a broken engine: call it again and the open target answers immediately, because it keeps its ephemeral project.

Verified against

Author's machine, 2026-10-13 / 14 / 16 / 19 / 23 / 26:

Component	Version
REA (rea-agents)	6.8.0 (229 tools)
DeepSeek Harness	desktop runtime 0.2.2-rc.1 (the host 2.0.14 was measured on; full range in the table above)
@deepseek-ai/dsh-mcp-client (row provider)	bundled by the harness
Ghidra	12.1.4 (this machine also runs 12.2 through the gate override — see below)
JDK	Temurin 21 (provisioned) / Amazon Corretto 25.0.5 (adopted; its major satisfies the pin 21)
Node.js	24.22.1 adopted (REA engines: ^22.23.4 || ^24.22.1 || >=26.12.1; ABI 137)
Host	Windows 11 x64 (build 26200)

Exercised live: a clean setup.mjs --reuse run reached setup complete; an MCP initialize handshake through the shim with no rea on PATH answered serverInfo {name: rea, version: 6.8.0}; and the whole chain ran under the real desktop host — its own Electron binary launched <DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs mcp, the shim started the vendored REA, and a live tool call reported path_source: environment for a path only that machine's bundle.json knows.

Not exercised end to end here, and therefore not claimed: the ~775 MB JDK + Ghidra download branch (planned by --dry-run, but this machine adopted its existing installs with --reuse), and first runs on macOS/Linux.

On 2026-10-23 the same chain was re-verified after the host Node moved to 24.22.1: the vendored rea-windows-x64.node and both conpty addons loaded under ABI 137; binary_overview on conpty.node reported 10 segments / 1178 procedures / 618 strings; procedure_pseudo_code returned real decompiled output; and java -version from the vendored JDK answered 21.0.12.1. The one thing that did not work on the first try was the first list_segments — it timed out at 64 s of cold Ghidra analysis, which is what the smoke step and the note above are for.

On 2026-10-26 the pin moved to REA 6.8.0 and the chain was re-verified on the same machine: setup.mjs --reuse reported probe cli=6.8.0 tools=229 and its Ghidra smoke imported java.exe in 26 s; an independent MCP handshake over stdio answered serverInfo {name: rea, version: 6.8.0} with 229 tools / 6 prompts on protocol 2024-11-05; apktool still answers 3.0.3, adb still reports 37.0.1 (1.0.41), and the jadx jar on disk still hashes to 6e5eacf5… — the value REA 6.8.0 requires. Executing 6.8.0's own release policy gave 12.1.4 → verified, 12.1.5 → compatible, 12.2 → unsupported, so Ghidra stays where it was.

On 2026-11-06 (2.0.14 — an 18-item external audit plus host adaptation) no upstream component moved (REA 6.8.0 / Ghidra 12.1.4 / JDK 21 / apktool 3.0.3 / jadx 0.7.1 / Node 24.22.1). The same chain was re-verified: --dry-run skipped all 7 components on their pins (the Ghidra 12.2 gate override and JDK 25 were not mistaken for drift) → a forced apktool re-provision went through the full staging path (no leftover .staging/.backup, and the directory held exactly the two files the manifest declares) → a full --reuse run covered 8 components plus fridaBindings FRIDA_OK 3 → probe cli=6.8.0 tools=229, the Ghidra smoke imported java.exe in 28 s, and the shim resolved 6.8.0, exit 0. Gates: 154 tests (151 pass / 0 fail / 3 skipped), audit 24/24 hash-pinned clean, the dependency gate at zod 4.6.7 · minimatch 11.0.5 · glob 14.0.4, host-compat 16 releases, and verify-branch ALL CHECKS PASSED. check-updates.ps1 went from a false-green exit 0 to reporting honestly, and exits 0 on this machine once repaired.

Scope and caveats

• REA exposes 229 tools at 6.8.0, so every session's tool list grows.
• Provisioning is honest about checksums: where a vendor publishes one it is pinned and verified; where none exists (jadx-headless-mcp, the apktool wrapper, Android platform-tools, the rolling "latest" URLs) the hash is measured once and recorded, and later drift is reported rather than silently accepted. bundle.json lists which components are in that state under unverifiedChecksums.
• Downloads are slow on some networks (measured 0.35–0.54 MiB/s from the GitHub release CDN here, so Ghidra alone takes ~25–45 min). Fetching is resumable on both routes: a partial .part is continued with a Range request, and a .part that already holds the whole file is verified and kept instead of re-fetched.
• Some REA tools launch processes, write files and access the network — DSH's permission policy applies. Analyse only artifacts you are authorized to inspect.
• The MCP child and the engines it launches are ordinary processes running as you. DSH's tool approvals govern what the model asks for, not what that child may read, write or send on its own; treat the plugin like any third-party code you install.
• On Linux, upstream's process-ownership checks need a procps-compatible ps on PATH; the Ghidra/JADX and firmware paths use it. A minimal image without it fails those tools, not the install.
• scripts/setup.mjs resolves the REA CLI's closure with npm ci against the shipped scripts/vendor-lock/ (package.json + package-lock.json). REA pins zod exactly ("4.4.3"), and 4.4.3 is inside GHSA-mrhx-c7xc-m3w5, so the closure is raised through overrides to zod 4.6.7 / minimatch 11.0.5 / glob 14.0.4. npm ci verifies every tarball's sha512; if a repackaging mirror makes that fail with EINTEGRITY, it falls back to "same versions via overrides + npm install" (same versions, not byte-reproducible, and the log says so). After landing, the version in node_modules/zod/package.json is read once more and anything below 4.6.6 is refused. node scripts/check-dependency-advisories.mjs is the offline gate, wired into CI. Package lifecycle scripts are still skipped on purpose — with exactly one named exception: frida, whose native binding (build/frida_binding.node) only exists after its install script runs, so setup runs npm rebuild frida for that one package and then proves it with import('frida') plus device enumeration (recorded under components.reaAgents.frida in bundle.json). A missing binding makes the Frida-backed tools unavailable and the rest of REA untouched, so it is reported as skip rather than failing the whole provision. This step is also the only one that cannot use this bundle's own route/proxy resolution — npm reads HTTPS_PROXY/HTTP_PROXY and NPM_CONFIG_REGISTRY from the environment, so on a network that needs a mirror, set one of those before provisioning (NPM_CONFIG_REGISTRY=https://registry.npmmirror.com). With --offline the step is refused outright instead of quietly going online.
• On POSIX, extraction falls back when the usual tools are missing: a zip is opened with unzip, then python3 -m zipfile, then the provisioned JDK's jar; a tarball with tar, then python3. A minimal image that has none of them fails with an explicit "no zip extractor available" line rather than a bare "extraction failed". This matters because Ghidra is a zip on every platform and unzip is not installed everywhere by default.

Updating — three independent things

Two standing rules — for this release and for every later 2.x release.

setup.mjs reconciles the bundle, never the plugin: it reads the manifest of the plugin that is installed right now. Installed 2.0.9? setup provisions 6.5.0 even after a newer release exists. The package changes only through the market (or npm i -g dsh-plugin-rea@<version> / a file: reinstall); running setup more often never substitutes for that.

And nothing runs it for you. A market install or update replaces the package and stops; the first setup.mjs — and one after every upgrade — is yours to run, from the plugin directory.

1. The row executes a copy of this plugin's launcher. <DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs is a mirror made by setup, and a market upgrade replaces the package without replacing it — so the mirror can keep running old code with no symptom at all. Since 2.0.8 the mirror compares itself with the launcher recorded in mirror.json and re-execs the plugin's copy when the two differ (one hop, guarded by REA_BUNDLE_SHIM_REEXEC). A mirror older than 2.0.8 cannot do this, so upgrading from 2.0.0–2.0.7 still needs one node scripts/setup.mjs. --shim-status reports mirror.stale; check-updates.ps1 says when to re-run.
2. The mirrored scripts/setup.mjs is a copy too, and it has no self-heal of its own. Running setup from the bundle root therefore executes the version from when the bundle was created, not the one you just installed. Run it once from the plugin directory to refresh that copy. Since 2.0.9 the copy strips the UTF-8 BOM from mirror.json / bundle.json exactly as the shim does, so a BOM-ed record no longer reads as "no record" on the setup side (which used to re-mirror the previous manifest).

Updating from the market

Update it where you installed it. Updating to 2.0.5 or later is routine; updating an installed 2.0.1–2.0.4 through the in-app market failed with this, three times on a healthy profile:

dsh-plugin-rea 更新后缺少入口文件（package.json 的 main/exports 指向的文件不存在），且未能验证恢复原版本文件

That message is not about a broken tarball. On every update the market asks whether the new build's package.json points at an entry file that exists (hasLoadableEntry in dshmarket), and rolls the build back when the answer is false. A bundle may legitimately own no entry — but only if a package it mounts has one, and the package this bundle mounts (@deepseek-ai/dsh-mcp-client) ships inside the harness build rather than in node_modules, so every location that test probes comes back empty. The 1.x and 2.0.x manifests also declared no main at all, so the answer was false even though the profile was working; the rollback was then reported as unverifiable for the same reason. Nothing in the profile was actually damaged — the update simply did not apply.

2.0.5 declares main: ./index.mjs. That file is a real, deliberately empty Cordis plugin: it registers nothing, spawns nothing, touches no state — the working row is still inserted by cordis.patch.yml — and it exists so the question "can this package load?" has an honest answer, and so import 'dsh-plugin-rea' resolves. On an older version, update to >=2.0.5; if the market refuses before that version exists, install [email protected] (Desktop: through the plugin manager) instead of updating in place.

Upgrading from 1.x

Install 2.x, then run provisioning once before restarting dsh:

node scripts/setup.mjs          # or: powershell -File scripts\rea-bundle-bootstrap.ps1 -Reuse

Why it matters: the 1.x row launched rea mcp from PATH, while the 2.x row launches <DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs — a file only setup.mjs creates. Upgrade and restart without provisioning and that row exits with MODULE_NOT_FOUND: the harness stays up (failOnStartupError: false is deliberate), but the mcp__rea__* tools are gone and you get no remediation message, because the shim that prints one does not exist yet. After the setup run they are back on the next start.

• This plugin never updates itself. Installed from a market? Update it there. Installed from a file: checkout? Overwrite the directory and reinstall.
• The row runs a COPY of this plugin's launcher. <DSH_HOME>/plugins/rea-bundle/bin/rea-shim.mjs is a mirror made by setup, so a market upgrade replaces the package but not the file the row executes — until setup runs again. From 2.0.8 the mirror notices that the installed plugin ships a different launcher and re-execs it (one hop, guarded by REA_BUNDLE_SHIM_REEXEC), so upgrades carry. On an older version the mirror stays behind silently; --shim-status now reports mirror.stale, and check-updates.ps1 says to re-run setup. If you are upgrading from 2.0.0–2.0.7, run node scripts/setup.mjs once after the upgrade.
• The mirrored scripts/setup.mjs is a copy too. A market upgrade replaces the package but not that copy, so a remediation command that points at <DSH_HOME>/plugins/rea-bundle/scripts/setup.mjs runs the setup from when the bundle was created — not the one you just installed. From 2.0.9 both the shim and setup strip a UTF-8 BOM from mirror.json / bundle.json (a BOM-ed record used to read as "no record" on the setup side, which re-mirrored the previous manifest). Running node scripts/setup.mjs once from the plugin directory refreshes that copy too.
• Upgrading from 2.0.0-bundle.1: that release mirrored the shim and setup.mjs but not the manifest, and recorded nothing about the plugin checkout, so a bundle it provisioned cannot repair itself — its background provisioning dies on the missing manifest, invisibly, and its own lock suppresses retries for six hours. Re-run node scripts/setup.mjs from the plugin directory once; it re-mirrors all three files and writes the record that lets later runs find the checkout. check-updates.ps1 reports this state.
• Provisioned components: re-run node scripts/setup.mjs; it re-verifies what is on disk and only fetches what is missing or wrong. Pins live in manifest/bundle.manifest.json.
• Upstream REA / Ghidra: update the pins in the manifest, then re-run setup. REA version-gates Ghidra (12.1.x today), so update REA first.

check-updates.ps1 (Windows PowerShell) reports the state of all of these at once. npm run check-updates forwards to it on Windows (PowerShell 5.1, then pwsh); on macOS/Linux it prints why the script cannot answer there and names the two portable commands instead — node bin/rea-shim.mjs --shim-status and node scripts/setup.mjs --dry-run — exiting non-zero rather than reporting a green it did not earn.

Still prefer the configuration-only 1.x line?

1.x is the adapter without provisioning: one row that launches rea mcp from PATH and expects Ghidra to be installed by you. It is the right choice when you already manage REA and Ghidra yourself, when you do not want the plugin fetching anything, or when you want the row to keep pointing at your own rea binary:

dsh plugin --profile <profile> add [email protected]
# or pin the branch instead of the version:
dsh plugin --profile <profile> add github:czcdr/dsh-plugin-rea#v1.0.6

Both lines publish the same patch shape and the same mcp__rea__* surface. 1.x is maintained on the v1 branch, where its own install notes live (Node on REA's engines, npm i -g rea-agents, Ghidra 12.1.x for native analysis).

License

MIT (this adapter). REA is MIT and owned by its authors; Ghidra is Apache-2.0 and owned by the NSA; Temurin is GPLv2+CE; Node.js is MIT; apktool, jadx-headless-mcp and Android platform-tools keep their own licences. No upstream code or binary is redistributed in this package — provisioning fetches each component from its official source.