How Zeno Works

Core workflow

1. Scan the local project — Zeno reads the files in the current directory and builds a compact source summary. Nothing is written at this stage.
2. Build a source summary — Zeno compresses the project into a structured representation that fits the context window of the AI provider.
3. Identify risky and high-value areas — Zeno surfaces files, paths, and patterns that carry meaningful production consequence.
4. Decide the next step — Based on the scan, Zeno determines whether the most useful action is a report, a test recommendation, a skip, a cleanup, a split, or a manual review flag.
5. Apply safety gates before changing files — For any action that modifies files, Zeno requires explicit confirmation and applies conservative scoping.
6. Keep diffs small and reviewable — Zeno prefers a focused 30-line change over a sweeping 300-line rewrite.
7. Let the human decide — Zeno’s output is a recommendation. You choose whether to keep or reject changes.

Operating principles

Read before touching. Zeno never modifies files without first analyzing them. The ship-readiness report and security scan are always read-only.

Prefer small safe improvements. When cleanup is warranted, Zeno extracts helpers, removes duplication, or improves naming in contained areas. It avoids broad rewrites.

Reserve Critical for severe consequence. A file is not Critical because it is large, untested, or uses browser globals. Critical means a bad change can cause auth bypass, payment failure, data loss, or production outage. See Risk Levels.

Treat missing tests as a risk multiplier. An untested auth path is riskier than an untested utility function. Test coverage affects how much Zeno trusts a change, not whether something is automatically Critical.

Skip files when the safest change is no change. Some files are messy but cohesive. Recommending a skip is a feature, not a failure.

Make rollback and history clear. Zeno keeps a per-project history so you can see what changed and when. See Config Files.