from claude-code-harness834
Quality guardrail for Claude Code update integrations: detects doc-only Feature Table additions and forces implementation proposals when needed.
This skill acts as a review guardrail for Claude Code (CC) update integration PRs. When a PR adds rows to a project's Feature Table (e.g., CLAUDE.md or docs/CLAUDE-feature-table.md), the skill classifies each added item as (A) implemented in this PR, (B) doc-only ('written only'), or (C) CC-auto-inherited. If any items are category B, the skill blocks the PR and emits concrete implementation proposals in a structured format.
Trigger during PR review flows for CC integration updates: when diffs include Feature Table additions, when an automated harness reviewer detects a CC integration PR, or when an internal /harness-review hook calls it. Do NOT load for general implementation work, normal reviews, or setup tasks.
Designed for Claude Code / internal harness review systems that can run read/grep/glob tools and enforce PR gating. Compatible with automated review agents and CI-integrated review hooks.
A quality guardrail skill for Claude/Codex upstream update integration reviews. It classifies Feature Table additions into A (implemented), B (doc-only, block PR), C (auto-inherited), P (planned) categories and forces implementation proposals for category B items. No bundled scripts — purely a SKILL.md-based review instruction set. Well-structured with clear triggers, classification rubrics, checklists, and mandatory output formats. Very niche: internal-only (user-invocable: false) for a specific harness project's PR review workflow.
Clean skill with no security concerns. Read-only git inspection only. The A/B/C/P classification system is well-designed with clear examples. Niche internal-use skill — high quality for its intended purpose but very limited audience. Bilingual frontmatter is a nice touch.
