onboard

Install or update the skills toolkit (skills, rules, templates, prompts, CLAUDE.md) in the current project from the canonical GitHub repo.

Implement a milestone's tasks one at a time using BDD then TDD, with live progress tracked in three places and a Review Contract filled per task.

Implement a milestone autonomously inside a Conductor worktree without stop-and-ask checkpoints, strictly enforcing scope-first and logging anything out of…

Build and refine a personal writing-style profile that /humanizer auto-loads, by sampling real writing (Slack, Linear), interviewing the user, and calibrating…

Cut a list of N candidates down to the M most-worth-doing using RICE or a simpler heuristic, with the user making the final call.

Author a new skill in this toolkit that conforms to the contract — strict frontmatter, 100-line cap, explicit trigger language.

Go up one layer of abstraction from the current code or file and map the modules around it plus who calls into them.

Interview the user one question at a time about a plan, decision, or design until a shared understanding emerges.

Compact the current Claude session into a cold-readable handoff doc so a fresh session can resume with no context loss.

Reproduce a bug, capture evidence to a cleared log, propose 2-3 fix options for the human to pick, implement, validate by re-running the repro, and add a…

Survey a codebase or product surface in PM-mode and produce a research note with hot picks per dimension, without making decisions or writing code.

Validate a completed milestone in two phases — agentic QA (run tests, hit endpoints, screenshot visuals) then human-eyes pass (manual test guide, feedback…