Add review finding reception loop - #4
Merged
Conversation
Lightweight Plan for the Review Finding Receiving / Fix Loop confirmed in docs/2026-07-23-review-finding-reception-spec.md (8c984f1), aligned to the HARNESS_DESIGN_PRINCIPLES north star (c99119c). Covers both Plan and Code receiving loops: artifact owner (Plan Writer / Fix) Receiving (understand -> verify -> judge -> respond) before any change; original Reviewer re-reviews fix or evidence-based feedback; all open findings closed before gate GO; 小P only routes and guards the gate. Phrasing fixes remove pre-classification of findings as valid/confirmed/ must-fix. No heavy state machine, forced disposition table, or pre-Code Review; verification via minimal contract invariants + representative Pilot of the six behaviors. Awaiting 小P independent Plan Review; no runtime/Harness code changed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Address 小P Plan Review (4 findings): - P1: fix MD032 (blank line before list) and switch Verification to the repo's real CI entry (npm ci && npm run ci && git diff --check); drop the non-standard pytest command. Full npm run ci now passes. - P1: remove the post-implementation "小P Plan Review (文档语义)" gate from Unit 1 — Plan Review happens before implementation; post-impl semantic review is 云上C总 Code Review. Unit 1-2 self-check routes to 小P, who only does material/readiness + status routing to Code Review. - P2: drop the new Receiving-marker contract validator/tests design (DD7/Unit 2) — validate_ordered_markers is still string-existence and can't prove Receiving behavior; north star §7/§9 says no mechanical guardrail without evidence. Keep existing contract CI for structure; behavior proven by Review + Pilot. - P2: rewrite DD8 — use this delivery itself as the main real Pilot (Plan finding Receiving/复审 + subsequent Code finding/Fix if natural), minimal walkthrough for non-occurring §10.2 branches; don't prescribe at-bot count/combo. Open Points adjudicated (no marker contract; this run + minimal walkthrough; CHANGELOG aligns event-driven style). DD1-DD6 unchanged. No runtime/Harness code changed. Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Unit 1 — Role minds + protocol text: - roles/fix.md: replace pre-classify language (有效/已确认 finding, 确认应修复) with Receiving responsibility (理解→核对→判断→响应). Add finding 不成立/依据不足/含义不清/改变需求 handling. - roles/plan-writer.md: add Plan Review finding Receiving responsibility. - roles/plan-reviewer.md: add 复审修正或有依据的反馈; clarify closing on feedback evidence, not just adopting suggestion. - roles/code-reviewer.md: same 复审 feedback semantics; Fix may propose alternative repair as long as it closes same problem. - roles/coordinator.md: add Finding 路由与 Reviewer 交接 section — 不预判 finding 成立; 不替 Receiving/闭环判断; readiness 检查不扩展 为实质 Code Review; 原 Reviewer 不可用改派规则 (独立 Reviewer + 完整上下文, 不随意换人绕过分歧). - HARNESS.md L39: 处理已路由 finding → 对已路由 finding 完成 Receiving. - core-workflow-spec.md L109/L111/L159: 有效 finding → 已路由 finding; 理解已确认 finding → Receiving; 判断应进入 fix → 路由到 Fix. Unit 2 — README/CHANGELOG consistency: - README.md: 主流程 Fix 后复审 → Fix Receiving 后复审. - CHANGELOG.md: add Review Finding Receiving / Fix Loop entry in existing event-driven orchestration style. Not modified: HARNESS_DESIGN_PRINCIPLES.md (北极星 c99119c already correct), scripts/validate_repository.py, tests/, Plan checkbox/status. No new headings, fields, state machines, disposition tables, or Reviewer identities added. Verification: - npm ci: 0 vulnerabilities - npm run ci: markdownlint 0 issues, 6 contract tests OK, validate_repository.py passed (21 files, 6 role cards) - git diff --check: CLEAN Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Why
The previous Harness wording described routed findings as already valid or confirmed. That encouraged Plan Writer and Fix agents to apply review feedback without independently checking it against the confirmed Spec, current evidence, and scope.
This change establishes the intended mindset: findings are professional objections to evaluate, not unconditional modification commands. Valid findings are fixed and verified; insufficient findings receive an evidence-backed response; requirement changes return to the decision owner; the Reviewer remains responsible for closing the loop.
Validation
npm cinpm run cigit diff --check