What you'll get

agent-harness/

SKILL.md
references/claude-md-best-practices.md
references/code-architecture-guidelines.md
references/design-harness-guide.md
references/implementation-guide.md
references/ios-swift-guide.md
references/research-foundation.md
references/scoring-rubric.md

A ready-to-install folder. Unzip it, drop it in your agent's skills directory, run /agent-harness.

Engineering /agent-harness Claude Code · Codex · any agent

Agent Harness

Audit and optimize your agent harness — the scaffolding that makes AI coding agents effective in your repo.

The skill I run on a repo before I trust agents to ship in it. It scores your harness — CLAUDE.md / AGENTS.md structure, documentation routing, definition of done, automated verification, knowledge persistence, task hygiene — against a rubric, shows you the scorecard, then fixes what you approve. This is the difference between agents that flail and agents that operate.

Scores six harness dimensions against a rubric
Audits CLAUDE.md / AGENTS.md structure + doc routing
Checks definition-of-done + automated verification
Fixes what you approve — not a lint dump
Ships with Rails, iOS/Swift + design-harness reference guides