feat(setup): register Kimi Code client with compact schema profile

[yadnuses]

Summary

Kimi Code (kimi-cli, Moonshot's coding agent) is the one major local MCP client with no rea setup integration, so its users hand-write mcpServers.rea into ~/.kimi-code/mcp.json (the exact path #1484's reproduction uses). That leaves setup detection, doctor, and uninstall blind to the registration, rea update never re-pins it, and a naive entry reproduces the Moonshot per-schema 400s of #1484 on current releases. This PR registers Kimi Code as a supported client and seeds the compact schema presentation into its entry so the first prompt after setup works. Fixes #1791.

Problem and expected behavior

• Use case: A Kimi Code user runs npx rea-agents setup and gets a Kimi Code checkbox the way OMP (Support OMP (oh-my-pi) in rea setup client integrations #1257 → feat(setup): register OMP during setup #1301) and Qwen Code ([Feature] Support Qwen Code in rea setup client integrations #1409) did, then loads both the MCP server and the reverse-engineer-anything workflow with no manual file edits.
• Observable outcome: rea setup --client kimi_code writes a pinned mcpServers.rea entry to user-level ~/.kimi-code/mcp.json with "env": { "REA_MCP_INPUT_SCHEMA_PROFILE": "compact" }; doctor --client kimi_code reports the registration; uninstall --client kimi_code removes it. No [Bug] Moonshot clients reject every request while REA is connected: advertised tool schemas exceed the per-schema size cap (kimi-cli unusable) #1484-style 400 cascade on the first prompt.

Change and scope

• SupportedClients.ts — kimi_code definition (configPath mcp.json under the Kimi home, markerPath the same directory, format: "json"). The existing JSON dialect writes exactly the mcpServers.<name> → {command, args, env} shape Kimi Code loads; no new dialect.
• Home resolution honors KIMI_CODE_HOME with codex-style semantics (non-empty override used as-is, empty falls back to ~/.kimi-code, no tilde expansion) — matched to Kimi's own resolution and pinned by tests.
• ClientRegistrationEnvironment.ts — the existing environment-retention merge gains a lowest-precedence defaults layer ({...defaults, ...existing, ...overrides}), so seeded values never clobber explicitly configured ones (preserves the "explicitly configured server settings" invariant).
• SetupClientConfiguration.ts — clientRegistrationDefaults(client) returns the compact profile seed for kimi_code (empty map otherwise) at the single clientConfigurationDesired convergence point, so register/inspect/aligned/update all seed it and stay idempotent.
• SetupSkill.ts — unchanged: Kimi Code discovers the shared ~/.agents/skills, the existing fallback destination; no client skill family needed (simpler than [Feature] Support Qwen Code in rea setup client integrations #1409's Qwen case).
• Tests + docs/installation.md (table row, Kimi Code section, restart note).
• Intentionally not included: a verify:kimi-client script, project-level config writes, tool-contract changes, Windows Kimi verification.

Contract and boundary impact

• Semantic owner and earliest changed stage: setup client registration (application layer: SetupClientConfiguration / ClientRegistrationEnvironment).
• CLI and MCP/tool-catalog contract: none — REA_MCP_INPUT_SCHEMA_PROFILE is presentation-only advertising; canonical schemas and server-side validation are untouched (catalog still 170 tools).
• Provider, bridge, target-format, or platform compatibility: none — no provider changes. Windows Kimi Code path behavior untested (same code shape as codex/qwen).
• Evidence, artifact, provenance, or reconstruction contract: none.
• Process execution, authorization, cleanup, or containment impact: setup writes only user-level client config through the existing reviewed plan/backup/atomic-replacement transaction; no process execution added.
• Generated metadata, package, or installation impact: setup_clients client metadata derives from SUPPORTED_CLIENT_DEFINITIONS (CI-generated; nothing committed); npm pack --dry-run and verify:package pass; docs/installation.md updated.

Evidence and regression coverage

• Tests added or updated: tests/boundary/filesystem/kimiClientConfiguration.test.ts (9 tests: path resolution on linux/darwin, KIMI_CODE_HOME semantics incl. empty/~//relative cases, register → preserve-unrelated → aligned-read → idempotent-reconfigure → uninstall round trip, compact seed + explicit full precedence + provider-key stacking); tests/acceptance/setup/kimiClientSetup.test.ts (CLI dry-run plan → --yes apply → .rea.backup → compact env → doctor --skill → idempotent re-run → uninstall, under KIMI_CODE_HOME); clientLifecycle.test.ts detection-order list; clientEnvironment.ts fixture.
• Base reproduction or other evidence: on the base revision no kimi_code client exists in SUPPORTED_CLIENT_DEFINITIONS (setup multi-select and --client have no Kimi Code option); [Bug] Moonshot clients reject every request while REA is connected: advertised tool schemas exceed the per-schema size cap (kimi-cli unusable) #1484 documents the schema-cap failure mode. Schema-size table in [Feature] Support Kimi Code in rea setup client integrations #1791 (2026-10-11, rea-agents@6.4.0): full profile 5 inputSchemas over ~14 KB (largest 61,876 B); compact profile 0 over (largest 12,090 B).
• User-visible CLI/MCP output: dry-run plan lists the Kimi Code registration action; doctor --client kimi_code covers it like other clients; docs section shows the written entry shape.
• Remaining proof gaps: no live Kimi Code session round trip against the MCP server (no verify:kimi-client lane here); Moonshot cap behavior measured previously, not re-triggered from this host; Windows unverified; full test:acceptance/test:integration and docs:build not run; dev toolchain is Node 24.14.0 vs .nvmrc 24.18.0 (nvm unavailable on the host).

For evidence-bearing changes:

• Observed, derived, and inferred claims remain distinguishable.
• Artifact identity, source provenance, and failed attempts remain preserved.
• Unsupported, incomplete, unavailable, or uncertain outcomes remain visible.

Validation performed

• npm ci — pass
• npm run test:focused -- tests/boundary/filesystem/kimiClientConfiguration.test.ts tests/acceptance/setup/kimiClientSetup.test.ts (plus clientLifecycle, qwen, clientConfigurationEnvironment, productCatalog) — 6 files / 90 tests pass
• npm run check:fast — pass (typecheck + lint incl. module boundaries)
• npm run verify:test-discovery — pass (821 test files, each discovered once)
• npm run verify:package — pass (setup planned→idempotent, backupReadback, skill, mcpTools=170)
• npm pack --dry-run — pass (1162 files, no artifacts written)
• npm run docs:check — pass
• npm run test:fast — 4298 passed / 7 skipped
• npm run test:boundary — 27 failed / 3039 passed; the failing set is identical on the main baseline of this host (missing ibtool, hdiutil behavior, Ghidra MCP fixture, JS inspector fixture); one additional 30 s timeout flake passes standalone 8/8. Unrelated to this change; no assertion weakened.
• Real Hopper/Ghidra, browser, or OS coverage: not run — registration-only change; no provider lanes on this host.

Compatibility, safety, and release

• Breaking changes or migration steps: none. Fresh kimi_code registrations seed the compact profile; explicitly configured REA_MCP_INPUT_SCHEMA_PROFILE values keep precedence over the seed. Existing manual entries are only touched if the user runs setup for kimi_code (standard reviewed transaction).
• Real Hopper/Ghidra, browser, or OS coverage: unavailable on this host (see above).
• Package or release metadata impact: adds one client to derived setup metadata; release via the usual Release Please flow.
• Security, privacy, process, or containment review: writes one user-level config file via the existing backup/atomic transaction; no credentials, no new process execution; the env seed is a presentation hint, not a capability change.

Review checklist

• This PR addresses a concrete problem or an agreed enhancement.
• The PR has one focused outcome and the title follows type(scope): outcome.
• Related issue is linked (Fixes #1791).
• Tests cover changed observable behavior and meaningful failure paths.
• Owning docs, contracts, and generated metadata are updated where needed.
• User-visible CLI/MCP changes include representative output.
• I checked the final diff for secrets, unrelated cleanup, and unsupported claims.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-11T11:37:58.916470Z	ac955f5	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.