ac-adversary
01Agent Claude Code
Attacks acceptance criteria — testability, completeness, edges, concurrency. Phase-1 gate.
Agent Claude Code
Attacks acceptance criteria — testability, completeness, edges, concurrency. Phase-1 gate.
Agent Claude Code
Proposes the stack and technical design, with justification and the simplest alternative considered. Input to the ADR.
Agent Claude Code
Tears down the proposed design — robustness vs convenience, failure modes, scale. Phase-2 gate.
Agent Claude Code
Implements the backend slice under the approved spec. Owns backend files. Loads python + fastapi skills.
Agent Claude Code
Proves it is NOT done — cut scope, missing deliverable, TODO, happy-path only. P4 checker.
Agent Claude Code
Proves the problem/spec isn't ready — finds every ambiguity and forced assumption. Adversarial intake gate.
Agent Claude Code
Implements the frontend slice under the approved spec. Owns frontend files. Loads the typescript skill.
Agent Claude Code
Confirms each AC is met with executable evidence. P4 checker (never the maker).
Agent Claude Code
Scheduled codebase-health auditor (the discovery heartbeat). Scans for drift/debt/latent bugs and reports deduped findings.
Agent Claude Code
Turns a raw brief/overview into a structured problem statement; with split mode, decomposes it into work items. Separates problem from solution.
Agent Claude Code
Ensures the change doesn't break what already worked. Runs the full suite, checks callers. P4 checker (always on).
Agent Claude Code
Actively ATTACKS the system — injection, auth abuse, secret leakage. P4 checker; mandatory for API/auth surfaces.
Agent Claude Code
Breaks a work item into stories with testable acceptance criteria (Gherkin). Turns "guarantees" into stress tests.
Agent Claude Code
Tries to BREAK the system — concurrency, races, edges, tests that pass for the wrong reason. P4 checker.
Agent Claude Code
Writes the failing tests (the RED step of TDD) from the locked acceptance criteria, before any implementation exists. Maker of tests only — never writes production code.
Command Claude Code
Render the AI Interaction Log deliverable.
Command Claude Code
Run the 6-phase Harness Studio engagement on a work item.
Command Claude Code
Register the project brief and produce a review-gated plan via hssd.
Command Claude Code
Show Harness Studio dev-time, token, and cost analytics.
Command Claude Code
List Harness Studio templates, or register/remove your own.
Command Claude Code
Manage work items via the hssd PM (list / add / claim).
Skill Claude CodeCodex
Load when designing, building, or reviewing any HTTP/REST API — response shapes, HTTP status codes, filtering/range semantics, pagination, and error bodies. The blessed contract so every endpoint answers the same recurring questions the same way.
Skill Claude CodeCodex
Load when designing, building, or reviewing APIs — versioning, auth patterns, rate limiting, idempotency, webhooks, breaking vs non-breaking changes, and long-running operations. Extends api-conventions.
Skill Claude CodeCodex
Load when designing, building, or reviewing any feature — guards against overengineering, unnecessary abstraction, and speculative generalization. The "do you actually need this?" filter.