Getting Started · 04
Every session starts with /startwork and ends with /handoff. This is the single most important habit in HQ.
Session Lifecycle
/startwork
work
/handoff
Without /startwork, your AI tool starts cold — no company context, no policies, no knowledge of your prior work. With it, the session loads everything: company policies (including hard-enforcement rules), available workers and skills, active projects with incomplete stories, and your last session's handoff state.
The slug you pass determines what context loads. Use a company name for broad work, a project name to pick up where you left off, a repo name for focused coding, or describe a task in plain English and HQ classifies it and proposes a worker pipeline.
Session started — Acme Corp
Pick a story, describe a task, or run a worker.
/handoff is how HQ gets smarter. The foreground pass is minimal — it commits pending changes, writes a thread file capturing everything about this session (what you did, what's next, what you learned), and saves a small handoff pointer. Index regeneration, learning capture, document updates, and search reindexing all run as background follow-ups.
The next session reads handoff.json (a small pointer file) and picks up exactly where you left off. No context lost. No re-explaining. No "where was I?"
On a long session, run /checkpoint mid-session to save state without ending — the in-flight counterpart to /startwork…/handoff.
What /handoff saves
Thread file
Full session summary, next steps, decisions made
Git state
Branch, commit hash, dirty flag, files touched
Worker state
Which worker was active, skill, completion status
Learnings
Mistakes caught, patterns that worked, corrections
Indexes
Thread + orchestrator INDEX — regenerated in background
handoff.json
Pointer to thread — small file, instant resume
Handoff complete
Start a new session and run /startwork to resume.
Each /startwork → work → /handoff cycle makes HQ smarter. Learnings accumulate as policies. Knowledge indexes grow richer. Thread history gives every future session a map of your team's decisions, mistakes, and patterns. After a few weeks, your AI tools know your codebase, your conventions, and your team's preferences — not because they were trained on it, but because you built the context.