[GH-ISSUE #2454] fix: reconcile documentation and MCP contracts with current behavior #821

Closed
opened 2026-10-15 18:28:45 +00:00 by bsakaguchi · 0 comments
Owner

Originally created by @N0zoM1z0 on GitHub (Oct 13, 2026).
Original GitHub issue: https://github.com/morluto/rea/issues/2454

Originally assigned to: @N0zoM1z0 on GitHub.

Problem

A documentation/behavior audit at main f254f5ac2b804b353a711264bb178f43fdf07867 confirmed the following 12 inconsistencies. Findings were checked against source owners and, where applicable, actual public CLI/SDK-connected stdio MCP workflows. Root READMEs/translations and website/ are outside this work.

Runtime contracts and tool guidance

  • D09: Harmony archive session outputs. The domain recognizes .hap, .hsp and .app (app-pack), but the shared session output-format enum omits all three. Actual MCP open_binary and binary_session return output-validation errors even though the target becomes active and inspect_artifact succeeds. CLI inventory/projection succeeds. The Harmony verifier currently only submits CLI-produced Evidence to MCP projection and misses the opener/status journey. Fix the owning format contract and exercise the complete public lifecycle.
  • D12: Unsupported native summary recommendation. The live inspect_analysis_view description recommends view: {kind: "summary"} for native function Evidence, while the native projector accepts kind: "native" facets. Actual current SDK calls reject summary and accept a procedure facet for the same authentic parent. Correct the description to match supported behavior.

Documentation accuracy

  • D05: docs/installation.md incorrectly groups Qoder, Trae, ZCode and Hermes under ~/.agents/skills; isolated public setup dry-runs show dedicated client roots. Reuse the authoritative client configuration guide.
  • D02: docs/raw-binary.md says Windows Ghidra P0 accepts PE applications only, although native PE DLLs are admitted too.
  • D06: docs/javascript-runtime-reconciliation.md omits supported target-only partial capture_electron_scenario Evidence from its exclusive producer list.
  • D07: The native/artifact skill reference says extract_artifact takes no arguments; it accepts optional integrity_policy, confirmed with real ZIP extraction.
  • D04: The ADB guide says am is not exposed while fixed-vector app/activity lifecycle operations use am start -W and am force-stop.
  • D11: SECURITY.md describes every owned provider bridge as a Unix socket, omitting Windows Ghidra's authenticated loopback TCP and private NTFS descriptor boundary.
  • D03: The RISC-V guide calls RVE a 15-register ABI. RV32E/RV64E define 16 architectural GPRs, x0–x15, including the zero register. See the official ISA.
  • D01: The managed guide's #managednative-boundary-rules link does not match installed VitePress's #managed-native-boundary-rules anchor. Preserve navigation in both repository and rendered docs.

Records and operational prerequisites

  • D10: License records claim complete locked dependency coverage but list 561 entries versus 586 in the audited lockfile, omitting 25 Frida-related optional runtime entries. detect-libc is also still classified as development; runtime notices lack a Frida entry. Refresh from the actual lockfile/archives and preserve upstream texts. This finding is a record-coverage gap, not a legal conclusion.
  • D08: The release synchronization guide requires a PR merge commit, while the repository currently disables that merge method. Document the maintainer prerequisite and ancestry requirement without silently changing repository settings or substituting squash/rebase.

Acceptance

  • Recheck every finding against updated main before editing; already-fixed items need verification rather than duplicate changes.
  • Correctness first; concise English guidance, shared rules maintained at their owning source, necessary limits and recovery steps retained.
  • Preserve CLI/MCP request semantics, Evidence provenance, target selection and lifecycle ownership. Do not add a native summary feature just to match inaccurate wording.
  • Run relevant source/schema checks and real public CLI/MCP reproductions. Validate generated docs, rendering/anchors and regenerated notice coverage. Record unavailable Windows/macOS/licensed-provider coverage honestly.
  • Keep heavy verification sequential with bounded CPU/memory and isolated client configuration. Do not edit root READMEs, translations, website docs or unrelated code.
  • Address this issue in a reviewed PR, follow CI/review feedback to completion, and merge after the required checks pass.

The audit also noted ambiguous Hopper shutdown wording and an unverified macOS sleigh prerequisite. They are not part of the 12 confirmed fixes; uncertain behavior should not be changed without evidence.

Originally created by @N0zoM1z0 on GitHub (Oct 13, 2026). Original GitHub issue: https://github.com/morluto/rea/issues/2454 Originally assigned to: @N0zoM1z0 on GitHub. ## Problem A documentation/behavior audit at main `f254f5ac2b804b353a711264bb178f43fdf07867` confirmed the following 12 inconsistencies. Findings were checked against source owners and, where applicable, actual public CLI/SDK-connected stdio MCP workflows. Root READMEs/translations and `website/` are outside this work. ### Runtime contracts and tool guidance - [x] **D09: Harmony archive session outputs.** The domain recognizes `.hap`, `.hsp` and `.app` (`app-pack`), but the shared session output-format enum omits all three. Actual MCP `open_binary` and `binary_session` return output-validation errors even though the target becomes active and `inspect_artifact` succeeds. CLI inventory/projection succeeds. The Harmony verifier currently only submits CLI-produced Evidence to MCP projection and misses the opener/status journey. Fix the owning format contract and exercise the complete public lifecycle. - [x] **D12: Unsupported native summary recommendation.** The live `inspect_analysis_view` description recommends `view: {kind: "summary"}` for native function Evidence, while the native projector accepts `kind: "native"` facets. Actual current SDK calls reject summary and accept a procedure facet for the same authentic parent. Correct the description to match supported behavior. ### Documentation accuracy - [x] **D05:** `docs/installation.md` incorrectly groups Qoder, Trae, ZCode and Hermes under `~/.agents/skills`; isolated public setup dry-runs show dedicated client roots. Reuse the authoritative client configuration guide. - [x] **D02:** `docs/raw-binary.md` says Windows Ghidra P0 accepts PE applications only, although native PE DLLs are admitted too. - [x] **D06:** `docs/javascript-runtime-reconciliation.md` omits supported target-only partial `capture_electron_scenario` Evidence from its exclusive producer list. - [x] **D07:** The native/artifact skill reference says `extract_artifact` takes no arguments; it accepts optional `integrity_policy`, confirmed with real ZIP extraction. - [x] **D04:** The ADB guide says `am` is not exposed while fixed-vector app/activity lifecycle operations use `am start -W` and `am force-stop`. - [x] **D11:** `SECURITY.md` describes every owned provider bridge as a Unix socket, omitting Windows Ghidra's authenticated loopback TCP and private NTFS descriptor boundary. - [x] **D03:** The RISC-V guide calls RVE a 15-register ABI. RV32E/RV64E define 16 architectural GPRs, x0–x15, including the zero register. See the [official ISA](https://docs.riscv.org/reference/isa/unpriv/rv32e.html). - [x] **D01:** The managed guide's `#managednative-boundary-rules` link does not match installed VitePress's `#managed-native-boundary-rules` anchor. Preserve navigation in both repository and rendered docs. ### Records and operational prerequisites - [x] **D10:** License records claim complete locked dependency coverage but list 561 entries versus 586 in the audited lockfile, omitting 25 Frida-related optional runtime entries. `detect-libc` is also still classified as development; runtime notices lack a Frida entry. Refresh from the actual lockfile/archives and preserve upstream texts. This finding is a record-coverage gap, not a legal conclusion. - [x] **D08:** The release synchronization guide requires a PR merge commit, while the repository currently disables that merge method. Document the maintainer prerequisite and ancestry requirement without silently changing repository settings or substituting squash/rebase. ## Acceptance - Recheck every finding against updated main before editing; already-fixed items need verification rather than duplicate changes. - Correctness first; concise English guidance, shared rules maintained at their owning source, necessary limits and recovery steps retained. - Preserve CLI/MCP request semantics, Evidence provenance, target selection and lifecycle ownership. Do not add a native summary feature just to match inaccurate wording. - Run relevant source/schema checks and real public CLI/MCP reproductions. Validate generated docs, rendering/anchors and regenerated notice coverage. Record unavailable Windows/macOS/licensed-provider coverage honestly. - Keep heavy verification sequential with bounded CPU/memory and isolated client configuration. Do not edit root READMEs, translations, website docs or unrelated code. - Address this issue in a reviewed PR, follow CI/review feedback to completion, and merge after the required checks pass. The audit also noted ambiguous Hopper shutdown wording and an unverified macOS `sleigh` prerequisite. They are not part of the 12 confirmed fixes; uncertain behavior should not be changed without evidence.
bsakaguchi 2026-10-15 18:28:45 +00:00
  • closed this issue
  • added the
    bug
    label
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#821
No description provided.