AgentSkillsCN

verification-before-completion

在声称工作已完成、问题已修复或测试已通过之前使用此指南。先用证据佐证,再作出结论——先运行验证,读取输出结果,然后才正式宣布成果。

SKILL.md
--- frontmatter
name: verification-before-completion
description: Use before claiming work complete, fixed, or passing. Evidence before assertions - run verification, read output, THEN claim result.

Verification Before Completion

Claiming work is complete without verification is dishonesty, not efficiency.

Iron Law: No completion claims without fresh verification evidence.

The Gate

code
BEFORE claiming success/completion:

1. IDENTIFY: What command proves this claim?
2. RUN: Execute the FULL command (fresh, complete)
3. READ: Full output, check exit code, count failures
4. VERIFY: Does output confirm the claim?
   - NO → State actual status with evidence
   - YES → State claim WITH evidence
5. ONLY THEN: Make the claim

Skip any step = lying, not verifying

Common Claims and Requirements

ClaimRequiresNot Sufficient
"Tests pass"Test command output: 0 failuresPrevious run, "should pass"
"Build succeeds"Build command: exit 0Linter passing
"Bug fixed"Test original symptom: passes"Code changed"
"Linter clean"Linter output: 0 errorsPartial check
"Requirements met"Line-by-line checklistTests passing
"Agent completed"VCS diff shows changesAgent reports "success"

Forbidden Patterns

Using before verification:

  • "Should work now"
  • "Probably passes"
  • "Seems to be working"
  • "Great!", "Perfect!", "Done!"
  • "I'm confident it works"

Instead: Run verification → read output → state result with evidence.

Verification Patterns

Tests:

code
✅ [Run npm test] "47/47 pass" → "All tests pass"
❌ "Should pass now"

Build:

code
✅ [Run cargo build] "exit 0" → "Build succeeds"
❌ "Linter passed so build should work"

Regression test (TDD):

code
✅ Write test → Run (pass) → Revert fix → Run (MUST FAIL) → Restore → Run (pass)
❌ "I've written a regression test"

Requirements:

code
✅ Re-read plan → Checklist each item → Report with evidence
❌ "Tests pass, must be complete"

Agent delegation:

code
✅ Agent reports → Check VCS diff → Verify changes exist → Report actual state
❌ Trust agent report blindly

Common Rationalizations

ExcuseReality
"Should work now"RUN the verification
"I'm confident"Confidence ≠ evidence
"Just this once"No exceptions
"Partial check is enough"Partial proves nothing
"Agent said success"Verify independently
"I'm tired"Exhaustion ≠ excuse

When to Apply

Always before:

  • Any success/completion claim
  • Any expression of satisfaction
  • Committing, PR creation, task completion
  • Moving to next task
  • Trusting agent delegation results

Integration

  • tdd skill: Verify RED-GREEN cycle
  • debugging skill: Verify fix actually works
  • implement skill: Verify each task completion
  • code-review skill: Verify fixes before claiming done