Prepare a code branch and pull/merge request for human review through safe rebasing, repository-specific validation, multiple independent subagent review rounds, evidence-based fixes, conventional commits, pushing, PR creation, and CI babysitting. Use when asked to prepare, polish, or finalize a PR/MR; obtain several reviews; iterate until clean; or make a branch ready for human review.
Debug an `imprint teach` run — env vars, CLI flags, log files, tracing, and diagnostic commands. Helps diagnose slow, stuck, or failing teach runs.
Refresh checked-in examples from generated Imprint site tools and cut an Imprint patch release. Use when the user asks to commit new generated tools to examples, replace stale example tool snapshots, update example docs/tests after generated tool repairs, or publish a patch release after example/runtime fixes.
Audit and repair generated Google Flights Imprint tools. Use when validating Google Flights search, calendar, booking, airline/bag filters, one-way, round-trip, multi-city, or open-jaw behavior; when investigating `selection_token` or `selected_flights` producer-consumer contracts; or when live audit results are slow, bot-sensitive, under-tested, or incorrectly waived as infrastructure.
Investigate a site's auth-tool compile — ongoing or completed. Locate the real agent transcript (not just the hook log), read the live 2FA verification, inspect the emitted workflow.json, confirm the login actually completed, and diagnose "no auth tool compiled".
Cut a new semver release — bumps package.json, previews changelog, commits, tags, pushes, and verifies npm publish.
Create a conventional commit from staged changes — picks type, scope, and writes the message.
Re-teach one or more imprint sites from their existing recordings and verify them with `imprint audit`, including the concurrency rules, background-monitoring recipe, and the known failure modes.