← ClaudeAtlas

add-review-disciplinelisted

Use when a command dispatches a reviewer, a cold reader or the prompt reviewer over a document or a delivery — how many times each runs, why none of them writes a file, and the rule that a finding is judged before it is applied.
brabos-ai/code-addiction · ★ 9 · Code & Development · score 76
Install: claude install-skill brabos-ai/code-addiction
# Review Discipline <!-- uses: - agent: plan-review-agent - agent: plan-readback-agent - agent: prompt-review-agent - mention: /add-framework--build - mention: /add-framework--plan - mention: /add-framework--brainstorm - mention: add-plan-authoring - mention: add-build-ledger --> <!-- A sibling skill of the same name now lives in the PRODUCT layer, at framwork/.codeadd/skills/add-review-discipline/. The duplication is DELIBERATE: that layer ships to users' projects, where this directory does not exist, so nothing there can reference anything here — and neither `uses:` block may name the other, because a cross-layer target resolves inside its own layer and dangles, which fails the build. The two files carry the same subject and REACH OPPOSITE CONCLUSIONS in two places, both on purpose: 1. The COUNT. This layer allows no second dispatch. The product layer allows one, after `apply then re-gate` — it has a deterministic schema gate between the two reads, which re-approves the changed document and makes the second read a different question. Nothing here does that, so a second read here would only produce new opinions. 2. DISK. Nothing here may write a report. The product layer permits exactly two, review-NNN.md and qa-validation-NNN.md, because qa-evidence.sh and converge-gates.sh consume them deterministically. Those are inputs to a script, not stored opinions a human must find. DO NOT merge them, and DO NOT add a check that keeps them