Workflow overview
Hyrax's product surface is a small set of jobs you can run against any repo. Each one has a specific job to do; most teams use all of them in a regular cadence.
| Workflow | What it does | Cost | Page |
|---|---|---|---|
| Audit | Find what's worth fixing (findings + a synthesis pass of suggestions) | $1–35 | Two depths — fast (scanner) or standard (scanner + reasoning agents) |
| Improve | Find what could be better (suggestions only) | ~$7–11 | Same engine as audit, deeper standalone pass focused on opportunities |
| Fix | One finding → one PR | ~$1–10 | 14-step fix gauntlet, isolated worktree, reviewer agent, tests |
| Task | Plain-English directive → PR | ~$1–10 | Same gauntlet as fix, but driven by an instruction instead of a finding |
| PR review | AI on every pull request | ~$1–2.50 | Scanner mode (inline comments) or LLM mode (rolling checklist) |
| Discovery | Knowledge bundle into the repo | ~$5–10 | HYRAX.md + .hyrax/discovery/ — read by Hyrax and your IDE AI |
| Learn | Hyrax getting better at your code | ~$0.50–1.00 | Background — proposes virtual patches you review |
Cadence most teams use
| When | What |
|---|---|
| Once, on adding a repo | Discovery (auto-runs) |
| Weekly or after each release | Audit (standard) |
| Whenever you spot a finding worth acting on | Fix |
| When you have a goal that doesn't map to a finding | Task |
| Quarterly or per planning cycle | Improve |
| On every PR (automatic) | PR review |
| Quarterly or after a refactor | Re-run Discovery |
| Continuously (automatic) | Learn |
How they fit together
Findings are the unit of work — produced by audit (kind=finding) and improve (kind=suggestion), consumed by fix, tracked through tickets, closed by re-run. Discovery is the context layer — its output makes every other workflow on the repo sharper. Related repos extend that context across repo boundaries. Learn is the meta layer — it watches the others run and proposes improvements to Hyrax itself.