LaunchKit · 2026
Back to Skills

adversarial-coach

Adversarial implementation review based on Block's g3.

1
1.4k downloads
by @killerapp

Setup & Installation

openclaw skills install @killerapp/adversarial-coach

Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:

npx clawhub install adversarial-coach

What This Skill Does

A dialectical review loop where an adversarial coach independently validates implementation completeness against a requirements file. Based on Block's g3 research, it separates the implementing agent from the reviewer to catch gaps the implementer would rationalize away. Returns either IMPLEMENTATION_APPROVED or a concrete list of missing items.

Unlike self-review, the coach discards prior context and evaluates code against requirements as written, catching gaps the implementing agent would overlook or excuse.

When to use it

  • Catching missing auth endpoints before code review
  • Verifying a spec is fully implemented after a coding session
  • Looping on fixes until all requirements are met
  • Reviewing JWT/bcrypt security gaps in auth implementations
  • Validating edge case test coverage against a written spec

Example Workflow

Here's how your AI assistant might use this skill in practice.

INPUT

User asks: /coach SPEC.md

AGENT
  1. 1Locate and parse SPEC.md to extract all requirements
  2. 2Review the current implementation independently, ignoring prior conversation context
  3. 3Check each requirement for implementation status, compilation, tests, and security gaps
  4. 4Return compliance report with IMMEDIATE ACTIONS NEEDED listing specific files and fixes
  5. 5After user applies fixes, re-run review until IMPLEMENTATION_APPROVED is issued
OUTPUT

IMPLEMENTATION_APPROVED with each requirement verified and test count confirmed