[PR #2148] feat(mcp): accept common argument-name aliases for procedure and search tools #5059

Open
opened 2026-10-31 01:01:35 +00:00 by bsakaguchi · 0 comments
Owner

📋 Pull Request Information

Original PR: https://github.com/morluto/rea/pull/2148
Author: @pablodiazjorge
Created: 10/12/2026
Status: 🔄 Open

Base: main ← Head: feat/mcp-input-key-aliases


📝 Commits (10+)

  • 06b761a feat(mcp): accept common argument-name aliases for procedure and search tools
  • f9620ac test(mcp): assert alias Evidence parameters
  • 00a0711 test(mcp): reject invalid alias values before dispatch
  • 4d06d3c fix(mcp): own compatible input spellings in named contracts
  • 25e93ae fix(mcp): retain selector constraints after schema presentation
  • 32a2809 fix(mcp): preserve alias schema input type
  • 52d2a2d fix(mcp): digest advertised tool schemas
  • 12a989a fix(mcp): describe optional argument aliases accurately
  • 13582fc Merge branch 'main' into feat/mcp-input-key-aliases
  • 324863c Merge branch 'main' into feat/mcp-input-key-aliases

📊 Changes

9 files changed (+562 additions, -10 deletions)

View changed files

📝 docs/mcp-contracts.md (+9 -0)
📝 src/contracts/enhancedToolContracts.ts (+8 -5)
📝 src/contracts/functionWorkflowToolContracts.ts (+4 -1)
➕ src/contracts/inputKeyAliases.test.ts (+141 -0)
➕ src/contracts/inputKeyAliases.ts (+165 -0)
📝 src/contracts/officialToolContracts.ts (+2 -1)
📝 src/contracts/toolContractHelpers.ts (+1 -1)
📝 src/contracts/toolSchemaMetadata.ts (+4 -2)
➕ tests/boundary/mcp/inputKeyAliasesMcp.test.ts (+228 -0)

📄 Description

Summary

Fixes #2146.

Agents carry argument names from one tool to the next. After resolve_containing_procedure, xrefs or read_bytes, which take address, they send address to procedure-scoped tools that take procedure; and they send query to the search tools, which take pattern. Strict input validation rejects these calls before they reach REA. In the study described in the issue (40 runs, 321 REA calls) this caused 18 input validation failures in 10 runs, and the agent recovered in 9 of them.

The two later studies in rea-lab, with the same pinned REA 6.4.0, saw the same mistake: in study 2, 7 of the 9 MCP input validation errors sent address or query, and 6 of those went to tools this PR covers (the seventh was list_strings, which takes no filter); in study 3, both of them were procedure_pseudo_code with address.

This PR makes the named tool contracts accept those two spellings, advertise them, and parse them into the canonical key before the handler runs. The current implementation is @morluto's rework of the first version (see Change and scope).

Problem and expected behavior

Before: search_strings {"query": "ACCESS"} returns Input validation error: Invalid arguments for tool search_strings: pattern: Invalid input: expected string, received undefined, Unrecognized key: "query".

After: the call is parsed as {"pattern": "ACCESS"}, and the handler and the recorded Evidence parameters see only pattern. The same applies to address on tools whose input declares procedure.

Unchanged:

  • An input that sends both spellings is rejected, now with Supply either pattern or query, not both (or procedure or address).
  • Unknown keys, including misspellings, are still rejected.
  • Canonical value checks, defaults and cross-field refinements apply to either spelling.

Change and scope

  • src/contracts/inputKeyAliases.ts holds a fixed list of two pairs, address for procedure and query for pattern, and compatibleToolInputSchema wraps a named contract's input: it accepts strict alternatives with either spelling, rejects both together, and parses into the canonical key. A pair applies only when the input declares the canonical key and not the alias. On the current head that is:
    • address (12 tools): analyze_function, annotate_native_function, inspect_native_api, procedure_address, procedure_assembly, procedure_callees, procedure_callers, procedure_info, procedure_pseudo_code, procedure_references, read_function_instructions, trace_native_values. In all of them procedure takes a procedure name or address.
    • query (5 tools): analyze_swift_types, get_objc_classes, search_procedures, search_strings, and search_bytes. In the first four pattern is a name or text filter. search_bytes reached main in #2341 after this PR was opened, and there pattern is a masked byte pattern such as 48 8B ?? ?? 89; query gets exactly the same validation, so text sent as query is rejected with the byte-pattern format error rather than reinterpreted. If you would rather keep query to text filters, excluding search_bytes is a one-line change.
  • The enhanced, function-workflow and official tool contracts build their input with compatibleToolInputSchema, so one contract owns both spellings and the canonical parsed output. The MCP server no longer rewrites raw caller arguments.
  • Advertised input schemas are one strict object that accepts exactly the alternatives: the properties of both spellings with additionalProperties: false, the keys every alternative requires, and a not rule that rejects both spellings (and requires one when the selector is required), in the full and compact profiles. An alias property is described as Same as <canonical>. (for example Same as procedure.) rather than with a copy of the canonical description.
  • docs/mcp-contracts.md documents this in a new "Compatible input spellings" subsection.

Catalog budget: the aliases as first advertised cost 7,104 bytes across the 17 tools, mostly text repeated in every tool: a selector sentence, the shared root-union presentation's input-group sentence and minProperties, and alias properties that copied their canonical description. That took the full input catalog with output schemas omitted 1,373 bytes over the 1 MiB frame that outputSchemaProfile.test.ts enforced then. 7b6b9034d, 918a144e7 and dd000cbca remove the repetition: the projection advertises the one strict object above instead of root alternatives, so the shared presentation adds nothing. #7379 has since removed that frame for full input. With main at 65f1c4473, the aliases cost 2,734 bytes in the full input catalog with output schemas omitted (908,170 bytes on Linux, 905,436 without them) and 2,136 bytes in the default compact catalog (568,933 bytes on Linux). I kept the flat object because it costs 2,734 bytes where the alternatives cost 5,082 for the same accepted set. It departs from the earlier comment that kept the exact alternatives in the projection for schema consumers; the unit and MCP tests check the accepted set with Ajv. If you would rather go back to the alternatives now that the frame is gone, that is your call.

The first version of this PR renamed aliases inside EvidenceMcpServer.registerTool and kept advertised schemas canonical-only. @morluto moved ownership into the named contracts and advertised the alternatives (4d06d3c0a, 25e93ae21, 12a989a6e, 52d2a2d37); 663ea6fb6 types the compatible schemas under exactOptionalPropertyTypes.

Not included:

  • No fuzzy matching of arbitrary keys.
  • No renaming of existing parameters.
  • No change to CLI argument parsing.

Contract and boundary impact

  • Semantic owner and earliest changed stage: the named tool input contracts (src/contracts/inputKeyAliases.ts and the contract builders that use it).
  • CLI and MCP/tool-catalog contract: MCP calls with only address (on the 12 tools above) or only query (on the 5 tools above) now validate, and tools/list advertises both spellings for those tools. The CLI is unchanged.
  • Provider, bridge, target-format, or platform compatibility: none.
  • Evidence, artifact, provenance, or reconstruction contract: none. Recorded parameters use the canonical key, so Evidence identity does not depend on which spelling the agent sent.
  • Process execution, authorization, cleanup, or containment impact: none. annotate_native_function is among the 12 tools; the alias only names its target procedure and does not change its mutation authority.
  • Generated metadata (docs/public/product-catalog.json), package, or installation impact: none; productCatalog.test.ts passes on this head.

Evidence and regression coverage

  • Tests added or updated:

    • src/contracts/inputKeyAliases.test.ts (8 tests): which inputs get aliases, caller values preserved, both spellings advertised as one strict object and parsed while ambiguity, a missing selector and unknown keys are rejected, alias properties described by their canonical spelling, canonical checks and refinements retained, optional selectors accepted with either spelling or neither but not both, transforms parsed once, and contracts without compatible keys left unwrapped.
    • tests/boundary/mcp/inputKeyAliasesMcp.test.ts (9 tests): the full tool catalog on EvidenceMcpServer over an in-memory MCP client, with and without the compact profile: aliases accepted, alias values validated before dispatch, both spellings rejected, accepted aliases advertised through SDK conversion (checked with Ajv), alias descriptions that name the canonical spelling instead of a key-based fallback in the full profile, and canonical parameters recorded in Evidence after an alias call.
  • Base reproduction or other evidence: on main, search_strings {"query"} and procedure_pseudo_code {"address"} fail with Unrecognized key. @CassieFox23 verified head 46ad54d2e on Linux over stdio MCP (comment).

  • User-visible CLI/MCP output, from the built server over stdio on Windows, head e1716846d, identical with the default compact profile and with REA_MCP_INPUT_SCHEMA_PROFILE=full (no binary open, so a call that passes validation returns target_unavailable; a rejected call returns invalid_request, and the table quotes its details.issues[].message):

    Call Result
    search_strings {"query": "ACCESS"} reaches the tool
    procedure_pseudo_code {"address": "0x401000"} reaches the tool
    search_bytes {"query": "48 8B ?? ?? 89"} reaches the tool
    search_bytes {"query": "ACCESS"} pattern: Use hexadecimal byte pairs or ?? wildcards, the same as {"pattern": "ACCESS"}
    search_strings {"query": "a", "pattern": "b"} Supply either pattern or query, not both
    procedure_pseudo_code {"adress": "0x401000"} procedure: Invalid input: expected string, received undefined, Unrecognized key: "adress"
  • Remaining proof gaps: no real-provider session was run with the change; validation happens before any provider call.

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

Windows 11 x64 (10.0.26300), Node 24.22.1, head e1716846d (this branch updated with main at the 6.8.0 release, 160d821ac), with the Windows native controls from the published rea-agents@6.8.0 package:

  • npx vitest run src/contracts tests/boundary/mcp/inputKeyAliasesMcp.test.ts: 215 passed.
  • npm run build, npm run test:prepare, npm run check and npm run docs:check: passed; then npx vitest run tests/boundary/filesystem/productCatalog.test.ts: 10 passed.
  • npx vitest run src/contracts src/server src/config tests/boundary/mcp: 1867 passed, 23 failed, 96 skipped. The same 23 tests, in 18 files outside this change, fail on main at 160d821ac on this host with the same native controls, so the change adds no failure.
  • The stdio calls in the table above, with REA_MCP_INPUT_SCHEMA_PROFILE unset (compact, the default since 6.8.0) and set to full.
  • After merging main at ff6d96bce (head 66dde5165; one import conflict in src/contracts/enhancedToolContracts.ts, both imports kept): tsc, src/contracts plus the alias and analysis-detail boundary tests (249 passed), the catalog tests (54 passed), npm run check and npm run docs:check passed.
  • On dd000cbca: npm run build, npm run test:prepare, npm run check and npm run docs:check passed. src/contracts, src/server and the alias, compact-schema, schema-validity, contract-presentation, doctor, output-schema-profile, v-mode and catalog tests: 462 passed and 1 failed, advertisedPatternsVMode.test.ts, which fails the same way on main on Windows (#7151). outputSchemaProfile.test.ts passes on Windows (1,046,979 bytes); with process.platform reported as linux while the modules load, it measures 1,045,579 bytes (1,049,949 on 66dde5165, matching CI). npx vitest run src/contracts src/server src/config tests/boundary/mcp: 2,007 tests, with the same 23 failures as the run on e1716846d above and no new ones.
  • After merging main at 235e8bc05 (4c4069178; one import conflict in src/contracts/functionWorkflowToolContracts.ts, both imports kept; npm ci for zod 4.6.6) and at 65f1c4473 (7dbfd6b54, no conflicts): npm run build, npm run test:prepare, npm run docs:check, typecheck, lint, format, knip, architecture guards, knip:production and lint:typed passed. The alias, output-schema-profile and v-mode tests, plus the four files that failed CI on 4c4069178 and were fixed on main by #7479 and #7349 (WindowsNativeLoader, safeOutputTreeCloseFailures, electronMcp, providerBinding): 58 passed. npx vitest run src/contracts src/server src/config tests/boundary/mcp: 2,015 passed and 22 failed, all among the 25 that fail on clean main at 235e8bc05 on this host (symlinks, Unix sockets and a file name with a tab), none new.

Compatibility, safety, and release

  • Breaking changes or migration steps: none. Every previously valid call keeps its result.
  • Real Hopper/Ghidra, browser, or OS coverage: none needed for validation, which runs before any provider call; checked on Windows here and on Linux by @CassieFox23.
  • Package or release metadata impact: none.
  • Security, privacy, process, or containment review: the alias list is fixed and small, and both spellings together are rejected, so validation never chooses between two values.

Author readiness

Human owner: @pablodiazjorge

  • I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work.
  • I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps.
  • I declare this PR ready for initial maintainer review.

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, or the reason for not linking one is stated above.
  • 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.

🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/morluto/rea/pull/2148 **Author:** [@pablodiazjorge](https://github.com/pablodiazjorge) **Created:** 10/12/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `feat/mcp-input-key-aliases` --- ### 📝 Commits (10+) - [`06b761a`](https://github.com/morluto/rea/commit/06b761a54d462094d5a8da57a83392c8fc5f942e) feat(mcp): accept common argument-name aliases for procedure and search tools - [`f9620ac`](https://github.com/morluto/rea/commit/f9620ac86a96fddcd314d3a7f4387abf8cf6f328) test(mcp): assert alias Evidence parameters - [`00a0711`](https://github.com/morluto/rea/commit/00a07114245d2d96f48ac941cc82db37f358fcd8) test(mcp): reject invalid alias values before dispatch - [`4d06d3c`](https://github.com/morluto/rea/commit/4d06d3c0a0a0fcd9346276fbb816e64c82fc3ac7) fix(mcp): own compatible input spellings in named contracts - [`25e93ae`](https://github.com/morluto/rea/commit/25e93ae2149d7f8d141cb04265eb84b5dcf2ebbc) fix(mcp): retain selector constraints after schema presentation - [`32a2809`](https://github.com/morluto/rea/commit/32a280904d4954831a54db21b6f27ab5c6ba21a8) fix(mcp): preserve alias schema input type - [`52d2a2d`](https://github.com/morluto/rea/commit/52d2a2d37b25fb1edd2de440f2f0263035009a6a) fix(mcp): digest advertised tool schemas - [`12a989a`](https://github.com/morluto/rea/commit/12a989a6e5821367de4ec05faf086da6102b920a) fix(mcp): describe optional argument aliases accurately - [`13582fc`](https://github.com/morluto/rea/commit/13582fcd0a9e447866d357fe5b535f3dfcb96949) Merge branch 'main' into feat/mcp-input-key-aliases - [`324863c`](https://github.com/morluto/rea/commit/324863ce9055f7a465d85c4534340568a473fad4) Merge branch 'main' into feat/mcp-input-key-aliases ### 📊 Changes **9 files changed** (+562 additions, -10 deletions) <details> <summary>View changed files</summary> 📝 `docs/mcp-contracts.md` (+9 -0) 📝 `src/contracts/enhancedToolContracts.ts` (+8 -5) 📝 `src/contracts/functionWorkflowToolContracts.ts` (+4 -1) ➕ `src/contracts/inputKeyAliases.test.ts` (+141 -0) ➕ `src/contracts/inputKeyAliases.ts` (+165 -0) 📝 `src/contracts/officialToolContracts.ts` (+2 -1) 📝 `src/contracts/toolContractHelpers.ts` (+1 -1) 📝 `src/contracts/toolSchemaMetadata.ts` (+4 -2) ➕ `tests/boundary/mcp/inputKeyAliasesMcp.test.ts` (+228 -0) </details> ### 📄 Description ## Summary Fixes #2146. Agents carry argument names from one tool to the next. After `resolve_containing_procedure`, `xrefs` or `read_bytes`, which take `address`, they send `address` to procedure-scoped tools that take `procedure`; and they send `query` to the search tools, which take `pattern`. Strict input validation rejects these calls before they reach REA. In the study described in the issue (40 runs, 321 REA calls) this caused 18 input validation failures in 10 runs, and the agent recovered in 9 of them. The two later studies in [rea-lab](https://github.com/pablodiazjorge/rea-lab), with the same pinned REA 6.4.0, saw the same mistake: in [study 2](https://github.com/pablodiazjorge/rea-lab/tree/main/writeups/2026-10-study-2), 7 of the 9 MCP input validation errors sent `address` or `query`, and 6 of those went to tools this PR covers (the seventh was `list_strings`, which takes no filter); in [study 3](https://github.com/pablodiazjorge/rea-lab/tree/main/writeups/2026-10-study-3), both of them were `procedure_pseudo_code` with `address`. This PR makes the named tool contracts accept those two spellings, advertise them, and parse them into the canonical key before the handler runs. The current implementation is @morluto's rework of the first version (see Change and scope). ## Problem and expected behavior Before: `search_strings {"query": "ACCESS"}` returns `Input validation error: Invalid arguments for tool search_strings: pattern: Invalid input: expected string, received undefined, Unrecognized key: "query"`. After: the call is parsed as `{"pattern": "ACCESS"}`, and the handler and the recorded Evidence parameters see only `pattern`. The same applies to `address` on tools whose input declares `procedure`. Unchanged: - An input that sends both spellings is rejected, now with `Supply either pattern or query, not both` (or `procedure or address`). - Unknown keys, including misspellings, are still rejected. - Canonical value checks, defaults and cross-field refinements apply to either spelling. ## Change and scope - `src/contracts/inputKeyAliases.ts` holds a fixed list of two pairs, `address` for `procedure` and `query` for `pattern`, and `compatibleToolInputSchema` wraps a named contract's input: it accepts strict alternatives with either spelling, rejects both together, and parses into the canonical key. A pair applies only when the input declares the canonical key and not the alias. On the current head that is: - `address` (12 tools): `analyze_function`, `annotate_native_function`, `inspect_native_api`, `procedure_address`, `procedure_assembly`, `procedure_callees`, `procedure_callers`, `procedure_info`, `procedure_pseudo_code`, `procedure_references`, `read_function_instructions`, `trace_native_values`. In all of them `procedure` takes a procedure name or address. - `query` (5 tools): `analyze_swift_types`, `get_objc_classes`, `search_procedures`, `search_strings`, and `search_bytes`. In the first four `pattern` is a name or text filter. `search_bytes` reached main in #2341 after this PR was opened, and there `pattern` is a masked byte pattern such as `48 8B ?? ?? 89`; `query` gets exactly the same validation, so text sent as `query` is rejected with the byte-pattern format error rather than reinterpreted. If you would rather keep `query` to text filters, excluding `search_bytes` is a one-line change. - The enhanced, function-workflow and official tool contracts build their input with `compatibleToolInputSchema`, so one contract owns both spellings and the canonical parsed output. The MCP server no longer rewrites raw caller arguments. - Advertised input schemas are one strict object that accepts exactly the alternatives: the properties of both spellings with `additionalProperties: false`, the keys every alternative requires, and a `not` rule that rejects both spellings (and requires one when the selector is required), in the full and compact profiles. An alias property is described as `Same as <canonical>.` (for example `Same as procedure.`) rather than with a copy of the canonical description. - `docs/mcp-contracts.md` documents this in a new "Compatible input spellings" subsection. Catalog budget: the aliases as first advertised cost 7,104 bytes across the 17 tools, mostly text repeated in every tool: a selector sentence, the shared root-union presentation's input-group sentence and `minProperties`, and alias properties that copied their canonical description. That took the full input catalog with output schemas omitted 1,373 bytes over the 1 MiB frame that `outputSchemaProfile.test.ts` enforced then. 7b6b9034d, 918a144e7 and dd000cbca remove the repetition: the projection advertises the one strict object above instead of root alternatives, so the shared presentation adds nothing. #7379 has since removed that frame for full input. With main at `65f1c4473`, the aliases cost 2,734 bytes in the full input catalog with output schemas omitted (908,170 bytes on Linux, 905,436 without them) and 2,136 bytes in the default compact catalog (568,933 bytes on Linux). I kept the flat object because it costs 2,734 bytes where the alternatives cost 5,082 for the same accepted set. It departs from the earlier comment that kept the exact alternatives in the projection for schema consumers; the unit and MCP tests check the accepted set with Ajv. If you would rather go back to the alternatives now that the frame is gone, that is your call. The first version of this PR renamed aliases inside `EvidenceMcpServer.registerTool` and kept advertised schemas canonical-only. @morluto moved ownership into the named contracts and advertised the alternatives (4d06d3c0a, 25e93ae21, 12a989a6e, 52d2a2d37); 663ea6fb6 types the compatible schemas under `exactOptionalPropertyTypes`. Not included: - No fuzzy matching of arbitrary keys. - No renaming of existing parameters. - No change to CLI argument parsing. ## Contract and boundary impact - Semantic owner and earliest changed stage: the named tool input contracts (`src/contracts/inputKeyAliases.ts` and the contract builders that use it). - CLI and MCP/tool-catalog contract: MCP calls with only `address` (on the 12 tools above) or only `query` (on the 5 tools above) now validate, and `tools/list` advertises both spellings for those tools. The CLI is unchanged. - Provider, bridge, target-format, or platform compatibility: none. - Evidence, artifact, provenance, or reconstruction contract: none. Recorded parameters use the canonical key, so Evidence identity does not depend on which spelling the agent sent. - Process execution, authorization, cleanup, or containment impact: none. `annotate_native_function` is among the 12 tools; the alias only names its target procedure and does not change its mutation authority. - Generated metadata (`docs/public/product-catalog.json`), package, or installation impact: none; `productCatalog.test.ts` passes on this head. ## Evidence and regression coverage - Tests added or updated: - `src/contracts/inputKeyAliases.test.ts` (8 tests): which inputs get aliases, caller values preserved, both spellings advertised as one strict object and parsed while ambiguity, a missing selector and unknown keys are rejected, alias properties described by their canonical spelling, canonical checks and refinements retained, optional selectors accepted with either spelling or neither but not both, transforms parsed once, and contracts without compatible keys left unwrapped. - `tests/boundary/mcp/inputKeyAliasesMcp.test.ts` (9 tests): the full tool catalog on `EvidenceMcpServer` over an in-memory MCP client, with and without the compact profile: aliases accepted, alias values validated before dispatch, both spellings rejected, accepted aliases advertised through SDK conversion (checked with Ajv), alias descriptions that name the canonical spelling instead of a key-based fallback in the full profile, and canonical parameters recorded in Evidence after an alias call. - Base reproduction or other evidence: on main, `search_strings {"query"}` and `procedure_pseudo_code {"address"}` fail with `Unrecognized key`. @CassieFox23 verified head 46ad54d2e on Linux over stdio MCP ([comment](https://github.com/morluto/rea/pull/2148#issuecomment-6183906469)). - User-visible CLI/MCP output, from the built server over stdio on Windows, head e1716846d, identical with the default compact profile and with `REA_MCP_INPUT_SCHEMA_PROFILE=full` (no binary open, so a call that passes validation returns `target_unavailable`; a rejected call returns `invalid_request`, and the table quotes its `details.issues[].message`): | Call | Result | |---|---| | `search_strings {"query": "ACCESS"}` | reaches the tool | | `procedure_pseudo_code {"address": "0x401000"}` | reaches the tool | | `search_bytes {"query": "48 8B ?? ?? 89"}` | reaches the tool | | `search_bytes {"query": "ACCESS"}` | `pattern: Use hexadecimal byte pairs or ?? wildcards`, the same as `{"pattern": "ACCESS"}` | | `search_strings {"query": "a", "pattern": "b"}` | `Supply either pattern or query, not both` | | `procedure_pseudo_code {"adress": "0x401000"}` | `procedure: Invalid input: expected string, received undefined, Unrecognized key: "adress"` | - Remaining proof gaps: no real-provider session was run with the change; validation happens before any provider call. For evidence-bearing changes: - [x] Observed, derived, and inferred claims remain distinguishable. - [x] Artifact identity, source provenance, and failed attempts remain preserved. - [x] Unsupported, incomplete, unavailable, or uncertain outcomes remain visible. ## Validation performed Windows 11 x64 (10.0.26300), Node 24.22.1, head e1716846d (this branch updated with main at the 6.8.0 release, 160d821ac), with the Windows native controls from the published `rea-agents@6.8.0` package: - `npx vitest run src/contracts tests/boundary/mcp/inputKeyAliasesMcp.test.ts`: 215 passed. - `npm run build`, `npm run test:prepare`, `npm run check` and `npm run docs:check`: passed; then `npx vitest run tests/boundary/filesystem/productCatalog.test.ts`: 10 passed. - `npx vitest run src/contracts src/server src/config tests/boundary/mcp`: 1867 passed, 23 failed, 96 skipped. The same 23 tests, in 18 files outside this change, fail on main at 160d821ac on this host with the same native controls, so the change adds no failure. - The stdio calls in the table above, with `REA_MCP_INPUT_SCHEMA_PROFILE` unset (compact, the default since 6.8.0) and set to `full`. - After merging main at `ff6d96bce` (head `66dde5165`; one import conflict in `src/contracts/enhancedToolContracts.ts`, both imports kept): `tsc`, `src/contracts` plus the alias and analysis-detail boundary tests (249 passed), the catalog tests (54 passed), `npm run check` and `npm run docs:check` passed. - On `dd000cbca`: `npm run build`, `npm run test:prepare`, `npm run check` and `npm run docs:check` passed. `src/contracts`, `src/server` and the alias, compact-schema, schema-validity, contract-presentation, doctor, output-schema-profile, v-mode and catalog tests: 462 passed and 1 failed, `advertisedPatternsVMode.test.ts`, which fails the same way on main on Windows (#7151). `outputSchemaProfile.test.ts` passes on Windows (1,046,979 bytes); with `process.platform` reported as `linux` while the modules load, it measures 1,045,579 bytes (1,049,949 on `66dde5165`, matching CI). `npx vitest run src/contracts src/server src/config tests/boundary/mcp`: 2,007 tests, with the same 23 failures as the run on `e1716846d` above and no new ones. - After merging main at `235e8bc05` (`4c4069178`; one import conflict in `src/contracts/functionWorkflowToolContracts.ts`, both imports kept; `npm ci` for zod 4.6.6) and at `65f1c4473` (`7dbfd6b54`, no conflicts): `npm run build`, `npm run test:prepare`, `npm run docs:check`, typecheck, lint, format, knip, architecture guards, `knip:production` and `lint:typed` passed. The alias, output-schema-profile and v-mode tests, plus the four files that failed CI on `4c4069178` and were fixed on main by #7479 and #7349 (`WindowsNativeLoader`, `safeOutputTreeCloseFailures`, `electronMcp`, `providerBinding`): 58 passed. `npx vitest run src/contracts src/server src/config tests/boundary/mcp`: 2,015 passed and 22 failed, all among the 25 that fail on clean main at `235e8bc05` on this host (symlinks, Unix sockets and a file name with a tab), none new. ## Compatibility, safety, and release - Breaking changes or migration steps: none. Every previously valid call keeps its result. - Real Hopper/Ghidra, browser, or OS coverage: none needed for validation, which runs before any provider call; checked on Windows here and on Linux by @CassieFox23. - Package or release metadata impact: none. - Security, privacy, process, or containment review: the alias list is fixed and small, and both spellings together are rejected, so validation never chooses between two values. ## Author readiness Human owner: @pablodiazjorge - [x] I reviewed the final diff and relevant existing code, understand the change and its edge cases, and take responsibility for the complete PR, including AI-assisted work. - [x] I checked the evidence above, confirmed it supports the current change, and recorded meaningful validation gaps. - [x] I declare this PR ready for initial maintainer review. ## Review checklist - [x] This PR addresses a concrete problem or an agreed enhancement. - [x] The PR has one focused outcome and the title follows `type(scope): outcome`. - [x] Related issue is linked, or the reason for not linking one is stated above. - [x] Tests cover changed observable behavior and meaningful failure paths. - [x] Owning docs, contracts, and generated metadata are updated where needed. - [x] User-visible CLI/MCP changes include representative output. - [x] I checked the final diff for secrets, unrelated cleanup, and unsupported claims. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
morluto/rea#5059
No description provided.