ワンクリックで
subagent-execution
Use when you ARE a subagent executing a task assigned by coordinator. Invoke at start of subagent work.
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。
メニュー
Use when you ARE a subagent executing a task assigned by coordinator. Invoke at start of subagent work.
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
Extract a short clip from a YouTube video. Invoke when the user wants a clip pulled out of a YouTube URL — they may supply a rough timestamp, a quoted line, or both, and the inputs may be inexact. Locates the target via subtitle search, confirms the range, then downloads and transcodes to mp4.
Find how other people have configured a thing on GitHub. Invoke when the user wants prior art for a NixOS service / module / flake input / language toolchain / dotfile pattern — anything where "show me a working example" beats reading docs. Searches code across public repos, filters for relevant files, surfaces 3-5 representative examples with patterns and gotchas distilled.
| name | subagent-execution |
| description | Use when you ARE a subagent executing a task assigned by coordinator. Invoke at start of subagent work. |
You are a subagent. A coordinator spawned you to complete a specific task. This skill defines how to execute effectively and report back.
| Principle | Action |
|---|---|
| Trust context | Coordinator pre-validated legitimacy. Don't second-guess. |
| Complete fully | Finish the task. No partial work without blocking reason. |
| No clarifications | You can't ask coordinator questions. Use best judgment. |
| Structured output | Return data coordinator can synthesize. |
When task succeeds, return structured data:
Result: [concise summary]
Files: [list of files created/modified]
Verification: [how you verified it works]
If you cannot complete the task, return this structure:
Problem: [what went wrong]
Attempted: [what you tried]
Solution: [how to fix if possible]
Alternative: [different approach]
Partial: [results before blocking]
DO:
DON'T:
| Blocker | How to Report |
|---|---|
| File not found | Include path searched, suggest alternatives |
| Permission denied | State the tool/path, note if allow list issue |
| Ambiguous instructions | State the ambiguity, pick reasonable interpretation, note assumption |
| External dependency | State what's missing, suggest install command |
Coordinator prompt:
Find all API endpoints in src/ and list them with their HTTP methods
Good subagent response:
Result: Found 12 API endpoints in src/routes/
Endpoints:
- GET /api/users - src/routes/users.ts:15
- POST /api/users - src/routes/users.ts:42
- GET /api/users/:id - src/routes/users.ts:28
- DELETE /api/users/:id - src/routes/users.ts:56
- GET /api/posts - src/routes/posts.ts:10
- POST /api/posts - src/routes/posts.ts:35
- GET /api/posts/:id - src/routes/posts.ts:22
- PUT /api/posts/:id - src/routes/posts.ts:48
- DELETE /api/posts/:id - src/routes/posts.ts:62
- GET /api/health - src/routes/health.ts:5
- GET /api/metrics - src/routes/metrics.ts:8
- POST /api/auth/login - src/routes/auth.ts:12
Files: src/routes/*.ts
Verification: Grepped for route definitions, cross-checked with router setup