メニュー
Assess whether the harness components are still appropriate for the current model and project. Audits each component for relevance, checks evaluator accuracy, measures feature velocity, and recommends simplifications.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Run a goal-bounded autonomous loop over the harness — select, build, verify, and commit features one at a time with a maker/checker split, an explicit stop-contract, and a per-run token/commit budget. Best run in the orchestrator agent.
Sync FEATURES.json and PLAN.md after architecture or scope changes mid-project. Preserves completed feature status while updating not-started features to align with the latest ARCHITECTURE.md and PLAN.md.
Initialize a new harness-managed solution project from a brief idea. Use when starting a brand new project. Creates docs/IDEA.md, docs/PLAN.md, docs/FEATURES.json, docs/ARCHITECTURE.md, docs/PROGRESS.md, AGENTS.md, and .github/copilot-instructions.md.
Design a feature before building it — research latest patterns, evaluate approaches, design APIs/data models/UI, and produce a technical design document. Use after /add-features and before the builder starts implementing.
Provide feedback on a design document and request changes. Use after /design-feature to iterate on the technical design before implementation begins. Supports multiple rounds of review.
Plan new features to add to an existing onboarded project. Analyzes impact on existing modules, designs features that fit the current architecture, and ensures backward compatibility.
| name | harness-health |
| description | Assess whether the harness components are still appropriate for the current model and project. Audits each component for relevance, checks evaluator accuracy, measures feature velocity, and recommends simplifications. |
Every harness component encodes an assumption about what the model can't do on its own. Those assumptions expire. Periodically re-evaluate.
List every harness component and assess if it's still needed:
| Component | Load-Bearing? | Evidence |
|---|---|---|
| Planner agent | Yes/Maybe/No | [Why] |
| Builder agent | Yes/Maybe/No | [Why] |
| Evaluator agent | Yes/Maybe/No | [Why] |
| Reviewer agent | Yes/Maybe/No | [Why] |
| Janitor agent | Yes/Maybe/No | [Why] |
| SessionStart hook | Yes/Maybe/No | [Why] |
| Stop hook | Yes/Maybe/No | [Why] |
| PreToolUse hook | Yes/Maybe/No | [Why] |
| PostToolUse hook | Yes/Maybe/No | [Why] |
Review docs/EVALUATION.md if it exists. Were findings legitimate or false positives?
Review docs/PROGRESS.md. Are logs detailed enough? Gaps? Consistent format?
From FEATURES.json and PROGRESS.md: features per session? Stuck items?
