安装
claude skill add --url github.com/openclaw/skills/tree/main/skills/amdf01-debug/sw-output-driven-dev文档
Trigger
Define success criteria and verification BEFORE coding. Agents prove their work.
Trigger phrases: "define success criteria", "output-driven", "verify before done", "prove it works", "acceptance criteria"
Process
- Define output: What exactly should the result look like?
- Write verification: How will we prove it works?
- Build: Implement the solution
- Verify: Run verification, show evidence
- Ship: Only after verification passes
Template
markdown
# Task: [Description]
## Success Criteria
- [ ] [Specific, measurable criterion 1]
- [ ] [Specific, measurable criterion 2]
- [ ] [Specific, measurable criterion 3]
## Verification Plan
For each criterion, how to verify:
1. [Run command X, expect output Y]
2. [Open URL, see element Z]
3. [Check file, contains content W]
## Build Log
[What was implemented and how]
## Verification Results
- Criterion 1: ✅ PASS — [evidence]
- Criterion 2: ✅ PASS — [evidence]
- Criterion 3: ❌ FAIL — [what went wrong, fix plan]
Rules
- Never claim "done" without showing verification evidence
- "Should work" is not verification — run it and show the output
- If you can't define success criteria, you don't understand the task
- Verification should be reproducible by anyone
- Failed verification → fix → re-verify (don't skip)
- Screenshots, logs, test output > "I checked and it works"