From 23f24d3665be52da9f9438449d2087b6e5de5672 Mon Sep 17 00:00:00 2001 From: Cuarroc <47299143+Cuarroc@users.noreply.github.com> Date: Thu, 15 Oct 2026 20:59:46 +0200 Subject: [PATCH 1/2] docs: require PLAN currency checks in reviews No-Test: documentation of a review rule --- .agents/skills/projecta-workflow/SKILL.md | 5 +++++ .claude/skills/projecta-workflow/SKILL.md | 5 +++++ AGENTS.md | 6 ++++-- 3 files changed, 14 insertions(+), 2 deletions(-) diff --git a/.agents/skills/projecta-workflow/SKILL.md b/.agents/skills/projecta-workflow/SKILL.md index 9997d56b..bfeccf4e 100644 --- a/.agents/skills/projecta-workflow/SKILL.md +++ b/.agents/skills/projecta-workflow/SKILL.md @@ -97,6 +97,11 @@ test first, then the fix. (initial dispatch: 300); record the overrun and forcing finding IDs. Current eligible reviewers and credit limits: coordinator's private record; follow AGENTS.md's recorded exceptions, never infer live availability. +- Every review checks `docs/PLAN.md` at the head: the PR's `Plan-ID:` row must + state what is true after merge, and the PR must leave no other row, user + section or decision inbox stale; cross-check against `gh`. A mismatch is a + medium finding (high if the user section asks for something already done) + and must be fixed in the same PR before `review-ok`. - Use recursive prompting selectively for complex plans/dependencies, hard diagnosis or demanding designs. Keep requirements/acceptance, necessary subquestions, first solution, one neutral extra check (no defect is valid), diff --git a/.claude/skills/projecta-workflow/SKILL.md b/.claude/skills/projecta-workflow/SKILL.md index 9997d56b..bfeccf4e 100644 --- a/.claude/skills/projecta-workflow/SKILL.md +++ b/.claude/skills/projecta-workflow/SKILL.md @@ -97,6 +97,11 @@ test first, then the fix. (initial dispatch: 300); record the overrun and forcing finding IDs. Current eligible reviewers and credit limits: coordinator's private record; follow AGENTS.md's recorded exceptions, never infer live availability. +- Every review checks `docs/PLAN.md` at the head: the PR's `Plan-ID:` row must + state what is true after merge, and the PR must leave no other row, user + section or decision inbox stale; cross-check against `gh`. A mismatch is a + medium finding (high if the user section asks for something already done) + and must be fixed in the same PR before `review-ok`. - Use recursive prompting selectively for complex plans/dependencies, hard diagnosis or demanding designs. Keep requirements/acceptance, necessary subquestions, first solution, one neutral extra check (no defect is valid), diff --git a/AGENTS.md b/AGENTS.md index 12d8063a..af703f1f 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -113,11 +113,13 @@ when it touches tier-A files. Never let the author's model family judge its own candidate. Record every finding and its disposition (accepted with commit, rejected with reason, follow-up) in the PR text. A second round only when round one had a high-severity finding; after -round two the coordinator (Root) decides merge, split or drop and records it in -the PR report; no third review round. Evidence is bound to the actual candidate; later + round two the coordinator (Root) decides merge, split or drop and records it in + the PR report; no third review round. Evidence is bound to the actual candidate; later changes invalidate the affected evidence, so review the delta again. A finding without `file:line` counts as unproven. +Every review checks `docs/PLAN.md` at the head: the PR's own row (identified by `Plan-ID:`) states what is true after merge, and changes in the PR leave no other row, the user section or decision inbox stale; cross-check against `gh`. A mismatch is a medium finding, or high if the user section asks for something already done, and must be fixed in the same PR before `review-ok` (user order 2026-10-15: the plan must never be stale). + Use recursive prompting selectively for complex plans, dependencies, hard diagnosis or designs with checkable criteria (user order 2026-10-08), following `docs/development/prompting.md`. Preserve requirements and acceptance; ask only From aecad6d15b80e70b7352127de434a56fe5b90493 Mon Sep 17 00:00:00 2001 From: Cuarroc <47299143+Cuarroc@users.noreply.github.com> Date: Thu, 15 Oct 2026 21:25:02 +0200 Subject: [PATCH 2/2] docs: fix PR 1588 review findings No-Test: documentation-only review follow-up --- .github/pull_request_template.md | 1 + AGENTS.md | 4 ++-- docs/PLAN.md | 1 + 3 files changed, 4 insertions(+), 2 deletions(-) diff --git a/.github/pull_request_template.md b/.github/pull_request_template.md index 860feccd..a692b440 100644 --- a/.github/pull_request_template.md +++ b/.github/pull_request_template.md @@ -6,6 +6,7 @@ starts with "## Report" on package branches. --> ## Report +Plan-ID: **Package:** · **Review tier:** A / B / C (AGENTS.md, rule 5) diff --git a/AGENTS.md b/AGENTS.md index af703f1f..d27bcfc3 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -113,8 +113,8 @@ when it touches tier-A files. Never let the author's model family judge its own candidate. Record every finding and its disposition (accepted with commit, rejected with reason, follow-up) in the PR text. A second round only when round one had a high-severity finding; after - round two the coordinator (Root) decides merge, split or drop and records it in - the PR report; no third review round. Evidence is bound to the actual candidate; later +round two the coordinator (Root) decides merge, split or drop and records it in +the PR report; no third review round. Evidence is bound to the actual candidate; later changes invalidate the affected evidence, so review the delta again. A finding without `file:line` counts as unproven. diff --git a/docs/PLAN.md b/docs/PLAN.md index e77d2f91..5570526d 100644 --- a/docs/PLAN.md +++ b/docs/PLAN.md @@ -1446,6 +1446,7 @@ Keine zweite Roadmap neben dem Plan; der Plan bleibt die einzige Quelle. | ID | Ziel | Abhängigkeit | Status | |---|---|---|---| | PLAN-AUTO-1 | PLAN-Modell + `npm run plan:check` (Befunde: PR ohne Planzeile, Status widerspricht GitHub, unbekannte Abhängigkeit, Job ohne Zeile, Entscheidung ohne Zeile, Plan veraltet) | – | läuft: Draft-PR #1564, 15.10. (`gh pr list --search "PLAN-AUTO-1"`) | +| AGENTS-PLAN-REVIEW | Reviews prüfen PLAN-Aktualität (eigene Zeile über Plan-ID, Nutzerbereich, Entscheidungs-Eingang gegen gh) | Root | Erledigt mit Merge von PR #1588 | | PLAN-AUTO-2 | Gate: jeder Paket-PR nennt eine `Plan-ID:`, die im Plan steht | PLAN-AUTO-1 | geplant | | PLAN-AUTO-3 | Status-Abgleich als Skript statt LLM: stündlich, Docs-PR, automatisch eingereiht | PLAN-AUTO-1 | geplant | | PLAN-NORM-n | Pakettabellen auf feste Spalten (ID, Ziel, Richtung, Version, Abhängigkeit, Status, Owner), je ≤300 Zeilen | PLAN-AUTO-1 | geplant |