skills.rest
davepoon
davepoonCOMMUNITY
Explore all skills by this author →
Website𝕏@davepoonGitHub

red-handed-audit

Verifies claimed test runs against Claude Code session transcripts and git state.

3.4k|487|Updated Jul 25, 2025
Search Tags:#audit#Git#CLI#test verification#session transcript#deterministic checks#Claude code
One-click install
npx skills add https://github.com/davepoon/buildwithclaude --skill red-handed-audit
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: red-handed-audit
Source: https://github.com/davepoon/buildwithclaude/tree/main/plugins/all-skills/skills/red-handed-audit
Command: npx skills add https://github.com/davepoon/buildwithclaude --skill red-handed-audit

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires @jinhyuk9714/red-handed.

What problem does it solve?

AI coding agents sometimes report that tests passed when no test run actually happened, when the last run failed, or when expected values were quietly rewritten to match a bug. This Skill audits the session's claims against its own local record so you can trust a summary before merging.

Core Features & Use Cases

  • Claim Verification: Runs nine deterministic checks comparing what the agent said with what the transcript and git working tree actually show.
  • Evidence-Based Findings: Each finding includes a timestamp and the quoted line it came from, classified as CAUGHT or SUSPICIOUS.
  • Local and Deterministic: No model calls and nothing leaves the machine; the same session always produces the same verdict.
  • Use Case: Before merging a branch, ask whether the tests the agent mentioned really ran, and get a report such as a CAUGHT finding showing the claimed "All 33 tests pass" against a failed vitest run.

Quick Start

Ask the assistant to audit this session and confirm whether the tests you were told about actually ran.

Frequently Asked Questions about red-handed-audit

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I verify that tests an AI agent claimed to run actually ran?▼

Run the audit command npx --yes @jinhyuk9714/red-handed@latest audit in the project directory. It reads the local Claude Code transcript and git state, then reports each gap between claims and actual runs with a timestamp and quoted line.

What does the red-handed audit tool check in a Claude Code session?▼

It runs nine deterministic checks comparing what the agent said with what the transcript and git working tree show, such as claims of passing tests after a failed run or expected values rewritten to match a bug.

Does the audit send my session data to an external service?▼

No. The audit runs entirely locally, reading the Claude Code transcript and git state on your machine. No model is called and nothing leaves the machine.

What is the difference between CAUGHT and SUSPICIOUS findings?▼

CAUGHT requires both the session showing the behavior and the code still showing it now. SUSPICIOUS means the pattern is present but the motive is not established, such as verification the tool cannot read like browser tests.

Why does the audit exit with code 1?▼

Exit code 1 means findings exist at the CAUGHT tier, which indicates the tool worked as intended and detected confirmed discrepancies. It is a signal of findings, not a tool error.

Related Skills

Explore More Software Engineering Skills→
ambroselittleambroselittle

wtf

Capture mid-session corrections for Claude Code and store them locally.

Community
Intermediate
tropekstropeks

review

Analyze PR diffs for SQL safety, LLM trust boundary, and side-effect risks.

Community
Advanced
EremesNGEremesNG
1

verification

Run automated checks and manual review to verify completed work.

Community
Intermediate
Clyra-AIClyra-AI
12

ci-failure-triage

Isolate CI failures with artifact-first verification and deterministic repros.

Official
Advanced