Claude Code Plugins

Community-maintained marketplace

Feedback

|

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name verification
description Use before claiming work complete, fixed, or passing. Requires running verification commands and confirming output before any success claims. Triggers: about to commit, create PR, mark task done, say "it works".

Verification Before Completion

Evidence before claims. Always.

The Iron Law

NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE

If you haven't run the verification command in this step, you cannot claim it passes.

The Gate

Before ANY claim of success:

1. IDENTIFY → What command proves this claim?
2. RUN      → Execute full command (fresh)
3. READ     → Full output, check exit code
4. VERIFY   → Does output confirm claim?
5. CLAIM    → Only now, with evidence

Skip any step = lying, not verifying.

Integration with Ant-Act

Before ant-act signals completed, ALL verification must pass:

{
  "step": "Add retry logic",
  "verification": {
    "tests": { "command": "npm test", "result": "47/47 pass", "exit_code": 0 },
    "lint": { "command": "npm run lint", "result": "0 errors", "exit_code": 0 },
    "build": { "command": "npm run build", "result": "compiled", "exit_code": 0 }
  },
  "all_verified": true
}

Evidence Requirements

Claim Requires NOT Sufficient
Tests pass Test output: 0 failures "should pass", previous run
Lint clean Lint output: 0 errors partial check
Build works Build: exit 0 lint passing
Bug fixed Symptom test passes "code changed"
Feature done All acceptance tests pass "it compiles"

Red Flags - STOP If You Think These

Thought Reality
"Should work now" RUN the verification.
"I'm confident" Confidence ≠ evidence.
"Just this once" No exceptions.
"Linter passed" Linter ≠ tests ≠ build.
"I'm tired" Exhaustion ≠ excuse.
"Looks correct" Run. The. Command.

Verification Artifact

Each verification produces evidence:

{
  "timestamp": "2025-01-02T16:30:00Z",
  "claim": "All tests pass",
  "verifications": [
    {
      "type": "test",
      "command": "npm test",
      "output_summary": "Test Suites: 5 passed, 5 total",
      "exit_code": 0,
      "passed": true
    }
  ],
  "overall": "verified"
}

Signals to Ant-Act

Signal When Payload
verified All checks pass Verification artifact
failed Any check fails { "failed_checks": [...], "output": "..." }
blocked Can't run verification { "reason": "...", "needs_human": true }

Verification Checklist

Before signaling completed to ant-orient:

  • Identified all verification commands
  • Ran each command fresh (not cached)
  • Read full output (not just exit code)
  • All exit codes are 0
  • No warnings treated as acceptable
  • Evidence artifact created
  • Can quote specific output proving claim

The Bottom Line

RUN → READ → THEN CLAIM

No shortcuts. No "should work". No assumptions.

Evidence or silence.