Your First Report

Once Better Harness is installed, open the repository you want to review, start a new session, and run:

/better-harness review this project's AI coding workflow and generate a report

Better Harness scopes behavior claims to relevant Task Episodes and the surrounding project mechanisms. Qoder produces a Canvas report; Claude Code, Codex, and Cursor produce self-contained HTML with paired Markdown. Missing or partial evidence remains explicit.

See the live demo report for what the HTML output looks like.

Reading the report

The report combines the five-part Agent Work Loop overview, prioritized findings, detected agent assets, and an evidence brief. Read it through the five work-loop questions; session evidence changes confidence and coverage, not the model.

From report to action

A report is the start of a loop, not a verdict. Each finding is a row with a next step, so a score turns into a change:

1. Draft a bounded fix. Run /better-harness repair-plan to validate one finding and draft a scoped repair plan, without writing new report artifacts.
2. Give recurring work an owner. When a finding looks like repeated work, route it through Loop Discovery to pick the smallest durable owner: a skill, hook, script, automation, or rule.
3. Schedule follow-up. A schedule-ready finding renders a row-scoped /schedule /better-harness handoff with cadence, validation, and a stop condition.
4. Confirm movement. Re-run the review to check the change landed and the capability signal moved.

Static-only inspection

From a source checkout, you can inspect repository evidence without reading local sessions:

node scripts/better-harness.mjs report --no-sessions