verify-impl

Audit implementation diffs for completed milestones and gate critical findings.

6|1|Updated Apr 5, 2026

Search Tags:#Claude Code #security review #code audit #multi-agent orchestration #adversarial review #release gating #TDD forensics

Ready to run

npx skills add https://github.com/snewhouse/aa-ma-forge --skill verify-impl-snewhouse

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: verify-impl
Source: https://github.com/snewhouse/aa-ma-forge/tree/main/claude-code/skills/verify-impl
Command: npx skills add https://github.com/snewhouse/aa-ma-forge --skill verify-impl-snewhouse

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires uv, python, git.

What problem does it solve?

It prevents regressions and hidden risks by running adversarial checks on the completed implementation of a milestone, so critical issues can’t silently slip past after execution.

Core Features & Use Cases

Post-implementation adversarial review: Runs an audit slate on the implementation diff after a milestone closes.
Profile-based agent dispatch: Selects which auditors run based on the milestone’s declared Audit-Profile (e.g., full, code-only, infra, docs-only, custom).
User-gated critical findings: Surfaces CRITICAL findings through an accept/dispute/defer panel before authorization steps continue.

Quick Start

Run the skill for a closed milestone revision by asking an AI agent to execute: Skill(verify-impl) --task <name> --milestone <id>.

Frequently Asked Questions about verify-impl

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

FAQPage Schema

How do I run an adversarial code audit on a completed milestone diff?▼

Run an adversarial code audit on a completed milestone diff by executing the skill with the target task and milestone ID. It computes SHA boundaries, runs agent suites based on your audit profile, and aggregates findings into a consolidated report.

How does post-implementation security review gate unsafe code releases?▼

Post-implementation security review gates unsafe releases by surfacing CRITICAL audit findings through an interactive accept, dispute, or defer panel before authorization steps continue, preventing hidden risks from slipping past execution.

Can I select specific auditors for a code audit instead of running a full suite?▼

You can select specific auditors by declaring a custom Audit-Profile for the milestone. The skill supports profile-based agent dispatch, allowing you to choose full, code-only, infra, docs-only, or custom audit configurations.

Do I need git and python to run multi-agent orchestration for release gating?▼

You need git, python, and uv installed to run multi-agent orchestration for release gating. These dependencies are required to compute milestone boundaries and execute the adversarial audit suites.

What is the best way to handle CRITICAL findings during a TDD forensics audit?▼

The best way to handle CRITICAL findings during a TDD forensics audit is through the built-in user-gated decision panel. The skill pauses to present an accept, dispute, or defer choice, blocking further workflow execution until resolved.

Why does the audit agent dispatch mode change between parallel and sequential?▼

The audit agent dispatch mode changes based on the AA_MA_AUDIT budget configuration. This setting controls whether agent suites run in parallel or sequential modes to optimize the adversarial review process for resource constraints.