Harness Score

An Archon workflow that audits a repository's readiness for AI-agent-driven development.

What it does

Runs a deterministic scan across 10 MVP checks — agent instructions (AGENTS.md/CLAUDE.md), README, build/test commands, CI workflows, test directories, lint/typecheck config, .archon/workflows, security scanning, sensitive paths — and scores the repo out of 100. Claude then synthesizes a markdown report with a rating, the top 3 fixes ranked by score gain, and remediation suggestions, written to $ARTIFACTS_DIR/harness-score.md.

Ratings: Agent-ready (90+) · Mostly ready (75+) · Partially ready (60+) · Risky (40+) · Not harnessed (<40)

The scoring is fully deterministic (a bun script node) — the AI only narrates the results, it never changes the numbers.

Install

Copy harness-score.yaml into your repo's .archon/workflows/ directory, or into ~/.archon/workflows/ to make it available in every project.

Usage

archon workflow run harness-score

Or from chat: ask Archon to "score this repo" / "run a harness diagnostic".

Requirements

• Archon >= 0.3.0
• Claude provider configured (the report node uses claude-opus-4-7; override model: to taste)

Not for

Implementing fixes, modifying code, or the full 10-category rubric — this is the MVP diagnostic only.