메뉴
Score and evaluate your project's codified context infrastructure — rubrics for constitution, trigger tables, failure modes, and subsystem map quality. Use when you say "check context health", "score my context".
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Ban useEffect in React components — enforce derived state, event handlers, data-fetching hooks, key resets, and useMountEffect instead.
Capture debugging insights, architecture decisions, and repeated explanations as durable project context. Implements the "explained it twice, write it down" methodology. Use when you say "write this down", "remember this", "codify this".
Three-tier codified context framework for AI coding agents — constitution, trigger tables, failure modes, and subsystem map templates. Use when you say "set up context", "create a constitution", "what is codified context".
Scaffold a Claude Code plugin from scratch — guides through directory structure, manifest creation, and best practices. Use when you say "create a plugin", "new plugin", "build a plugin", "scaffold a plugin".
Transform an agent execution plan into a team-ready proposal or RFC with adjustable depth and audience awareness. Use when you say "share this plan", "write a proposal", "create an RFC", "make this shareable".
Navigate Claude Code plugin development documentation and resources. Use when asked "where are the docs", "how do I learn plugin development", "find resources for plugins".
| name | context-health |
| description | Score and evaluate your project's codified context infrastructure — rubrics for constitution, trigger tables, failure modes, and subsystem map quality. Use when you say "check context health", "score my context". |
| version | 1.0.0 |
Quality rubrics and evaluation methodology for codified context infrastructure. Use this to assess whether your context documents are serving their purpose and identify improvement opportunities.
A context infrastructure that exists but is inaccurate or stale can be worse than no infrastructure — it actively misleads the AI agent. Regular health evaluation catches drift before it becomes a problem.
Run through these five questions to get a rough health score:
Does .claude/context/constitution.md exist and have all 10 sections filled in?
Does trigger-tables.md have routes for your main source directories?
Does failure-modes.md have entries from recent debugging sessions?
Does subsystem-map.md cover all major directories?
Do specs/ exist for your most-changed subsystems?
Total: /30
Map to health rating:
For a thorough evaluation, score each document against the rubrics in references/coverage-criteria.md. The rubrics cover:
| Score | Rating | Interpretation |
|---|---|---|
| 90-100 | Excellent | Context infrastructure is comprehensive and current |
| 70-89 | Good | Core documents solid, some gaps in coverage or freshness |
| 50-69 | Fair | Foundational docs exist but need significant enrichment |
| 30-49 | Poor | Major gaps — several core documents missing or very stale |
| 0-29 | Critical | Context infrastructure barely exists — run bootstrap setup |
When the health check reveals gaps, address them in this order:
Missing core files — Create them, even minimal versions. The paper's guideline G1 ("a basic constitution does heavy lifting") applies here: even a minimal constitution is dramatically better than none. Run through the bootstrap setup guide from codified-context-patterns.
Broken references — Links in the constitution that point to non-existent spec files, or trigger table routes pointing to missing context documents. Fix immediately — broken references cause the AI to fail silently.
Stale build commands — Build commands that no longer work are the most immediately damaging stale content. The AI will run them and fail. Verify every command in §4 actually executes successfully.
Undocumented active subsystems — Write specs for high-churn subsystems first. The areas you change most frequently have the highest leverage for improvement. Check git log to identify high-churn directories.
Incomplete failure modes — Add entries from recent debugging sessions. If you can recall debugging anything in the last month that took more than 30 minutes, it probably belongs in failure-modes.md.
Routing gaps — Add trigger table entries for file patterns and keywords that you find yourself repeatedly context-switching around. If you're loading the same spec manually every time you work in a directory, add the route.
Quality improvements — Fill in missing fields in failure mode entries (especially the Prevention column), add rationale to trigger table rows that lack it, update last-reviewed dates.
Every 2 weeks: Review failure modes. Remove entries where the root cause has been permanently fixed (move to the Archived section). Add entries from any debugging sessions since the last review.
Monthly: Check tech stack versions in §2 against your actual installed versions. Verify build commands in §4 still work. Versions drift, commands change, package managers update.
After major refactors: Review the architecture summary in §5 and the trigger tables in §8. Structural changes to the codebase mean routing rules may be pointing at moved or deleted directories.
After onboarding new agents or skills: Update the trigger tables. New agents/skills should have routing entries so the AI knows when to invoke them.
