docs(planning): extension survey & task planning (post-D-class-closur…#155
Conversation
…e roadmap) Survey the implemented surface and the design docs' declared seams, judge extension candidates, and shape them into executor-ready waves: - New docs/extension_survey_planning.md (PLANNING): capability map, survey findings S1-S8 (schema/extractor ghost facets, output-format asymmetries, stale ROADMAP D-class line, unrecorded open-question resolutions), design verdicts per candidate, and a CSCI-56..61 task plan sized to the AGENTS.md 5.2 envelope (1 brief = 1 PR = <=1 day). - Wave 1 (hygiene, Sonnet): docs consistency sweep + H-5 P3 x3. - Wave 2 (asymmetries, Opus/Sonnet): ghost-facet authoring advisory (ADVISORY-D9 candidate, advisory not error to preserve 23.1 hand-built states) + SARIF/gh-actions format parity. - Wave 3 (gated on Phase X-2 evidence): 19 spec quality metrics (Advisor-surface only, never seats the verdict) + 10.3 round-trip log. - Declared-asymmetry rejections pinned: compare/sensor parity, mute-vs-suppression unification, init recipe expansion. - E-3 (X-2 pilot) remains the primary queue axis; waves run alongside. - Register the doc in the CLAUDE.md Design Documents table (298/300 lines) and README planning list. Survey execution was delegated to two parallel Sonnet agents per docs/model_delegation_policy.md; design verdicts and planning are the design model's. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: e6c211e2cc
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…ths in D9 scope
- W1-a / S6: the stale D-class status also lives in README.md:28
("5 of 7 D-class findings resolved"), not only ROADMAP.md; widen the
sweep to a repo-wide grep for stale wording so public-facing status
stays consistent.
- W2-a / CSCI-57: ADVISORY-D9 scope covers both state paths
(type_relations / control_flow / data_flow / coverage) and their
delta counterparts (type_changes / cfg_delta / coverage_delta etc.) —
constraints can target delta paths too, so warning on only one side
leaves a hole. Derive the state<->delta mapping from path_schema, no
hardcoding.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0601d9db67
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
… (Codex P2) The stale D-class wording also survives in code_semantic_ci_design.md:1295 and dogfooding_real_pr_complexity.md (4 spots), not only ROADMAP/README. Widen the W1-a grep term set (5 of 7 / 未解決 / both open), cite the verified remaining locations, and pin the principle: current D-class status claims live in the findings tracker alone — other docs carry cross-links plus the resolving PR number. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: fe93ba2d8f
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…scope (Codex P2) S2/W2-b covered ssp and compile-repair/validate-plan but missed check's own declared deferrals: sensor-enabled runs reject --format gh-actions (G-4 slice) and advisory-enabled runs reject sarif/gh-actions (H-4 slice), check.py:426-438. Add them to the S2 finding and the CSCI-58 scope under the shared escape hatch (declare exclusion where no natural mapping exists), with the requirement that 'later slice' wording is rewritten to a declared exclusion either way — so W2-b closes S2 with zero undecided asymmetries. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: dbc5d38383
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…ilot (Codex P2) Internal inconsistency: 2.3 says 19 spec-quality design is informed by C-lite mis-verdict tag data, but the W3-a gate said 'after the X-2 pilot' — the pilot is plumbing-only (5 cases, numbers explicitly banned from performance claims per pr_validation_planning.md) and produces no C-lite tags; those come from the N>=48 production step. Move the W3-a gate to full X-2 production analysis completion, align W3-b and the Wave 3 heading, and state explicitly that pilot completion does not open the gate. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 690cad6444
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…origin D9 (Codex P2 x2) - S8/W1-a: the staleness also lives in the indexes, not just planning docs — CLAUDE.md Design Documents rows still call Phase G 'CSCI unstarted / active queue main axis' and Phase H 'gated on G-5 / not queued' (both phases completed 2026-06-10), README planning list mirrors it. Add index refresh to the sweep. - W2-a/CSCI-57: target-doctor sees the compiled target after template prepending, and the stock refactor template already targets the type_relations ghost facet (compiler/templates.py:48) — unscoped D9 would fire on every default refactor target for a constraint users did not author. Pin: D9 limited to user-authored constraints (reuse D3's origin distinction); the template itself gets updated, removed, or explicitly rationalized in the same brief. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 41002eff7f
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…plate (Codex P2) Rationale-only closure would leave every default refactor target carrying an inert hard constraint (type_relations is never extracted) with no advisory at all, since D9 is pinned user-authored-only. Valid closures are now: update the template, remove the constraint, or explicitly surface a template-origin warning. Recommendation recorded: drop the ghost-facet constraint from the stock template and point hand-built-state users to explicit authoring. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 5149789536
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
Archiving docs/doc_refactor_planning.md per its own self-archive step would leave the README planning-list link broken and the archive index unaware of the move; W1-a's S7 now requires updating the CLAUDE.md table status, the README link, and the archive index together — zero broken or stale links after the move. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 0a2a64bbff
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
…nk sweep (Codex P2 x2) - W2-a/CSCI-57: adding D9 to the in-use advisory enum is a mandatory advisory-3 bump under the json_schema.md compatibility policy (same reasoning as the D6/D7 advisory-1->2 bump) — schema artifact, docs, and sync test updates go into the AC; 'decide later' was too weak. - W1-a/S7: inbound references to doc_refactor_planning.md live in 13 files including AGENTS.md and a discipline test, so the archive move requires a repo-wide inbound-link grep, not an enumerated list; live docs/tests get updated, dated memory logs stay immutable as history. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 24db54a735
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".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".
… (Codex P2) llm_sensor_adapter_planning.md declares H-1..H-5 complete at the top while its premise section still reads 'G-5 remaining / gated on G-5 / queue insertion to be decided' — an internal contradiction in a first-party planning doc. W1-a's S8 now covers completed-phase planning doc bodies, not only open-question sections and index rows. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
…査 planning (#155) 反映 - 2026-07-07.md reflection 新設 (体制ルール 2 層化 / 委譲調査→S1-S8→ Wave 1-3 / Codex 8 巡 P2 x10 / 委譲ループで生ペイロード直撃ゼロの実測) - _index.md 1 行追記 - 30 日 TTL: dated log 11 file を archive/2026-05, 2026-06 へ移送 (2026-06 dir 新設)、_index Detail path 更新、archive/INDEX.md 反映 - STATUS.md: Phase 1 段落更新 / 直近 merged に #154+#155 追加・#148 を STATUS_MERGED_LOG へ移送 / 次の発行順序を Wave 1 参照に更新 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01CmnG4powRuRvD37unq7i1m
2 participants
Completion Summary: extension-survey-planning
Phase
Repository planning docs (post-D-class-closure extension roadmap). Product code untouched.
What Changed
docs/extension_survey_planning.md(PLANNING): capability map of the implemented surface, survey findings S1–S8 (ghost facets without extractors, output-format asymmetries, stale D-class wording, unrecorded open-question resolutions), per-candidate design verdicts, and a CSCI-56–61 task plan in three waves (hygiene → asymmetry fixes → §19/§10.3 gated on Phase X-2 evidence).docs/model_delegation_policy.md; design verdicts and planning are the design model's.code_semantic_ci_design.md+dogfooding_real_pr_complexity.mdwith the tracker-single-source principle (fe93ba2).Acceptance Criteria Status
advisory-2version constant were verified against the code by an independent executor passTests
python -m pytest tests/discipline/ -q --no-cov→ 36 passed;ruff check .→ cleanDogfooding
python -m pytest tests/discipline/ -q --no-covgrep -rn "5 of 7|未解決|both open" README.md ROADMAP.md docs/(verified the S6 stale-wording finding is real before planning W1-a)Files Changed
Deviations from Brief
None (no formal brief — planning doc authored from a two-agent delegated survey; this PR is itself the brief source for Waves 1–3).
Open Questions / Deferred
Next Handoff
Merge, then issue the Wave 1 briefs (W1-a docs sweep / W1-b CSCI-56) via /new-brief.
🤖 Generated with Claude Code