ai-engineering-harness

AI-native software engineering organization harness that turns a project from idea → PRD → MVP → polished product. Multi-agent (Coordinator + 17 specialists (Explore, Plan, Frontend, Backend, Database, QA, Bug Hunter, Behavior Reviewer, Architecture Reviewer, Security Reviewer, UI Reviewer, Conflict Resolver, Release, Context Assembly, Memory Curator, Review Aggregator, Scope)) operating on an Issue-driven, Worktree-isolated, PR-carried, Adversarial-reviewed, Evidence-Gated loop. Maintains CLAUDE.md/AGENTS.md/DESIGN.md/ENGINEERING.md/TESTING.md/CONTRIBUTING.md/PROJECT_STATUS.md plus docs/, memory/, sessions/, tasks/ with dynamic L0–L3 context control. Use when an engineering effort needs a persistent multi-agent organization that produces verifiable, reviewable, shippable software instead of one-shot coding.

Design, take over, or refactor an agent app. Use when (a) the user has a PRD/idea and the answer is or could be an LLM-driven agent, (b) the user wants Codex to wire an existing agent app into a new harness / workflow / eval, or (c) the user hands over a "broken" agent app that needs diagnosis and a targeted rebuild. Acts as the Agent App Architect: writes the Agent Contract + Harness Contract, picks the entry workflow (new / takeover / refactor), then hands implementation off to $ai-engineering-harness. Pairs with that skill -- build-agent-app answers "should we even build an agent, and what should it look like"; ai-engineering-harness answers "how do we ship it as a working, evidence-gated piece of code."

Awwwards-grade creative web UI design + frontend engineering for AI agents. Generates non-Dashboard, asymmetric, motion-rich websites (Next.js + React + TS + Tailwind + Framer Motion + GSAP + React Three Fiber). Use when the user wants a creative landing page, design portfolio, or experimental product page — not a standard SaaS dashboard. Adjacent to $build-agent-app and $ai-engineering-harness: hand off implementation to the harness after the design brief is approved.