handoff-audit

Generates a self-contained adversarial audit prompt for independent verification in a fresh session.

Updated Aug 30, 2026
One-click install
npx skills add https://github.com/ryanportfolio/threejs-interview-test --skill handoff-audit-ryanportfolio
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: handoff-audit
Source: https://github.com/ryanportfolio/threejs-interview-test/tree/main/.claude/skills/handoff-audit
Command: npx skills add https://github.com/ryanportfolio/threejs-interview-test --skill handoff-audit-ryanportfolio

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? When you finish a code change, asking the same session to review its own work produces rubber-stamping rather than real verification. This Skill drafts a complete, self-contained audit prompt you paste into a separate fresh session, so an independent auditor with zero shared context can genuinely try to falsify your change. ## Core Features & Use Cases - Claim Extraction: Pulls concrete, falsifiable assertions from the diff and conversation — what changed, what invariants must hold, and why risky decisions were safe. - Fidelity References: Pins pre-change git handles (e.g. git show <base-sha>:path) so the auditor can byte-diff moved or refactored code instead of trusting your summary. - Structured Output Contract: Produces one fenced markdown block with tagged sections (<audit_target>, <claims>, <checks>, <constraints>, <output_contract>) and PASS/FAIL reporting rules with file:line evidence. - Use Case: After completing a refactor that moved nine symbols across files, invoke this Skill to get a ready-to-paste prompt instructing a fresh session to re-grep importers, verify no dangling consumers, and diff the moved code against its original. ## Quick Start Ask the assistant to draft a handoff audit prompt for the work just completed on this branch, optionally naming a specific commit, PR, or file as the audit target.

Frequently Asked Questions about handoff-audit

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

FAQPage Schema
How do I get an independent review of AI-generated code changes?▼

Draft an audit prompt that carries the branch name, commit SHAs, and specific falsifiable claims, then paste it into a fresh session with no shared context. The fresh session re-derives everything from the diff and its own greps rather than trusting the original session's summary.

How to verify a code refactor didn't break anything?▼

Pin the pre-change state with git handles like `git show <base-sha>:path` and byte-diff moved code against its new location. Then grep the whole repo with word boundaries for dangling consumers of deleted or moved symbols, watching for substring false positives.

Why not run the audit in the same session that made the change?▼

A session that made a change shares context and assumptions with the change itself, which biases it toward confirmation. Independence requires a separate session with zero memory of what changed or why, so every verdict comes from its own investigation.

What makes a code audit prompt effective?▼

Effective prompts state specific falsifiable claims, an adversarial falsify-first role, concrete pass/fail checks with file:line evidence, and environment constraints. Vague 'please review my changes' prompts produce vague audits because there is nothing specific to disprove.

What are common false alarms in grep-based code audits?▼

Substring matches are the main culprit, such as `User` matching inside `UserProfile` or `get` inside `getCached`. Pointer comments like `// X moved to ...` also look like real references. Auditors should use word-boundary greps and re-verify every FAIL before reporting it.