ワンクリックで
supervisor
MANDATORY final step. Validates task completion through testing and QA before allowing stop.
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
MANDATORY final step. Validates task completion through testing and QA before allowing stop.
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
Convert files between formats (PDF to text, Markdown to HTML, JSON to CSV, CSV to JSON). Use when asked to convert, transform, or export files to different formats.
Persistent memory with semantic search - store and recall information across sessions. Use when asked to remember something, recall previous discussions, or manage long-term memory. Also use proactively to store important discoveries and compact context.
Install software packages at runtime in the Docker container. Use when a required tool or library is missing, or when asked to install packages.
Query SQLite and PostgreSQL databases. Use when asked to run SQL queries, analyze database data, or explore database tables.
Track tasks for the current work session. Use this to break down work into steps and track progress. ALWAYS use this when starting a new task - add items first, then mark them done as you complete them.
Analyze images using vision AI model. Use for UI screenshot verification, understanding downloaded images/assets, reading text from pictures (OCR), or any visual analysis task.
| name | supervisor |
| description | MANDATORY final step. Validates task completion through testing and QA before allowing stop. |
| allowed-tools | Bash, Read, Task |
You called /supervisor. Now I will verify your work is complete.
Answer these questions honestly:
# Try common test commands (use whichever applies)
npm test 2>&1 || python -m pytest 2>&1 || go test ./... 2>&1 || echo "No test runner found"
# Check if server is running on common ports
curl -s http://localhost:3000 > /dev/null && echo "Server OK on 3000" || \
curl -s http://localhost:8000 > /dev/null && echo "Server OK on 8000" || \
curl -s http://localhost:5000 > /dev/null && echo "Server OK on 5000" || \
echo "No server detected"
# Look for error patterns in recent output
echo "=== Checking for errors ==="
Call the qa-agent to review your work:
Use Task tool with subagent_type "qa-agent" to verify:
1. Does the implementation match the original request?
2. Are there any obvious bugs or missing pieces?
3. Is there adequate test coverage?
CONTINUE
Reason: [what QA found]
Action: [fix the issues]
Then go fix the issues - do NOT stop.
CONTINUE
Reason: Tests failing
Action: Fix the failing tests
Then go fix them - do NOT stop.
Output this:
ALL COMPLETE
Task: [what was requested]
Built: [what you made]
Tested: [how you verified]
QA: VERIFIED
Then write the completion marker:
echo "COMPLETE $(date -u +%Y-%m-%dT%H:%M:%SZ)" > ~/workspace/.supervisor_complete
If task/workflow files exist, you can check them too:
# Task list (if exists)
cat ~/workspace/.tasks 2>/dev/null || echo "No .tasks file"
# Workflow state (if exists)
cat ~/workspace/.workflow_state 2>/dev/null || echo "No .workflow_state file"
But these are NOT required for completion - tests and QA are what matter.
.supervisor_complete marker when truly done