harness

Run the evaluator agent to independently assess the current state of the build against feature_list.json. Produces a structured evaluation report with pass/fail per feature and actionable feedback. Use after each /harness-run session or whenever you want a quality gate.

Initialize a long-running build project. Decomposes a high-level goal into a granular feature_list.json, sets up claude-progress.txt, creates an init.sh bootstrap script, and makes an initial git commit. Run once before your first /harness-run session.

Run one coding session of the long-running harness. Reads claude-progress.txt and feature_list.json to pick up exactly where the last session left off, implements the next batch of features, commits progress, and updates the log. Safe to invoke repeatedly across fresh context windows.

Show a quick summary of the current harness state — features completed vs remaining, recent git commits, last session log entry, and any blockers noted. Use to orient yourself before starting a new session or checking progress.

Incremental coding agent for one session of a long-running build. Reads claude-progress.txt and feature_list.json to orient itself, selects the next batch of failing features, implements them one at a time, verifies each, commits per-feature, and updates the progress log. Designed to run safely across context resets — each session is fully self-contained.

Independent evaluator agent for the generator-evaluator loop. Assesses the current build state against feature_list.json without bias toward the work it is reviewing. Produces a structured HARNESS_EVAL_REPORT.md with pass/fail verdicts per feature, quality observations, and prioritized feedback for the next coding session.

One-time harness setup agent. Decomposes a high-level goal into 50–200 granular, testable features written to feature_list.json (all starting as "failing"). Creates claude-progress.txt for cross-session state handoff, an init.sh environment bootstrap script, and makes the first git commit. Run once per project before any harness-run sessions.

3 hooks across 3 events