AI Verify

Verify

Purpose

Evidence before claims. This skill has two faces: (1) a verification protocol that proves claims with commands, and (2) a multi-mode scanner for quality, security, and governance. Both share the same principle: run the command, read the output, check the exit code. No guessing.

When to Use

• Before claiming "it works" (run the test, show the output)
• Before claiming "it's secure" (run the scan, show the findings)
• Before claiming "Done!" (verify every acceptance criterion with evidence)
• When running quality/security/governance scans on a codebase

Process

Verification Protocol (claim mode)

For every claim, follow IRRV:

I -- IDENTIFY: What command proves this claim?

• "Tests pass" -> uv run pytest tests/ -v
• "No lint errors" -> ruff check .
• "No secrets" -> gitleaks protect --staged
• "File exists" -> ls -la path/to/file

R -- RUN: Execute the FULL command. Not a subset. Not from memory. Fresh execution.

R -- READ: Read the FULL output. Check:

• Exit code (0 = success, non-zero = failure)
• Warning lines (even with exit code 0)
• Actual numbers (test count, coverage %, finding count)

V -- VERIFY: Does the output CONFIRM the claim?

• If yes: report with evidence (exact command + key output lines)
• If no: report the discrepancy. Do not claim success.

Forbidden words (never use these without evidence):

• "should work", "probably fine", "seems to", "looks good"
• "Done!", "Perfect!", "All set!"
• "I believe", "I think", "most likely"

Scan Modes (7 parallel modes)

Auto-detect: when invoked without a mode, infer from context.

Scan Output Contract

Every scan mode produces:

## Score: N/100
## Verdict: PASS | WARN | FAIL

## Findings
| # | Severity | Category | Description | Location | Remediation |

## Gate Check
- Blocker findings: N (threshold: 0)
- Critical findings: N (threshold: 0)

Scan Thresholds

Verification Checklist (use before claiming DONE)

- [ ] Every acceptance criterion verified with a command
- [ ] All tests pass (exact count reported)
- [ ] Lint/format clean (zero warnings)
- [ ] No secrets in staged files
- [ ] Coverage maintained or improved (exact % reported)
- [ ] No forbidden words used in the completion report

Common Mistakes

• Claiming success without running the command
• Running a subset of tests instead of the full suite
• Ignoring warnings when exit code is 0
• Using forbidden words ("should work") instead of evidence
• Not checking exit codes
• Reporting coverage from memory instead of from the tool output

Integration

• Called by: /ai-dispatch (post-task review), ai-build agent (after implementation), user directly
• Calls: stack-specific tools (pytest, ruff, gitleaks, etc.)
• Read-only: never modifies source code -- produces findings with remediation

$ARGUMENTS