Keyword reference

Every repo myai init scaffolds ships a CLAUDE.md (and GEMINI.md) that teaches your agent a small vocabulary of short phrases. Type one instead of a full prompt, and the agent runs a defined, repeatable protocol instead of improvising. This page documents that vocabulary as shipped by the framework's project template — an individual repo's CLAUDE.md may extend or reorder it.

Keywords fall into two tiers:

  • Core — kept inline in CLAUDE.md because the agent needs the full

protocol in context every session (session start, shipping, autonomy).

  • Extended — loaded on demand from AI/documentation/KEYWORDS_REFERENCE.md

when you type one; kept out of the always-loaded context to save tokens.

Core keywords

KeywordWhat it does
helloShow all available keywords as a table.
start workSync git, read state + handoff, report what's done/in-progress/blocked, identify the next priority.
agent modeFull session activation: git sync, multi-machine + Docker checks, mobile-branch merge, state boot (brain-first, file fallback), dispatch relevant specialist agents in parallel. The heaviest, most thorough boot.
agent mode -minLightweight boot — sync + one file read (or a brain delta) + report, under a minute. Use for quick check-ins; use full agent mode before shipping or after a machine switch.
agent mode -resumeJoin the last scheduled/headless run for this repo and continue its work from the runner's transcript.
jam / jam <idea>Collaborative brainstorming mode — the agent becomes a thought partner, pushes back, weighs trade-offs. No code or commits until you say build it.
ship itThe full deploy pipeline: commit → push to test → CI → Vercel preview → PR → review → merge to main → confirm production.
merge itMerge test → main via PR once CI is green, without the full ship it walkthrough.
wrap upSession close: summarize, update state + handoff, commit + push, print a traffic-light dashboard, end with a status banner.
yolo godFull autonomous mode — no clarifying questions — until the current plan finishes or the next commit lands.
yolo [minutes]Timed autonomous mode, e.g. yolo 10 for the next ten minutes.
yolo offDeactivate autonomous mode immediately.

Extended keywords (on demand)

KeywordWhat it does
status / review / plan [feature] / audit / handoffQuick status summary, code review dispatch, feature planning, security+QA+standards audit, or a detailed handoff write-up.
scaffold [thing]Generate boilerplate — API, page, schema, Docker, tests — via the matching specialist.
schedule <description>Queue one autonomous task in the gateway's work queue for the off-hours runner to pick up.
schedule listShow what's queued for this repo or the whole fleet.
schedule planRefresh the repo's improvement plan and 10-day schedule, register it with the runner, then ship + wrap up.
brain status / brain commit / brain stash / brain pop / brain merge / brain log / brain deltaGit-flavored control of the brain — status, append an atom, freeze/resume context, merge a session, view history, or boot a diff-only catch-up.
check bugs / fix bug [id] / check features / build feature [id]Pull from the Connect Hub bug/feature tracker, triage, and implement.
make preview / make prodProductionize a project — provision Vercel, MongoDB Atlas, and optionally a Render service.
remote / remote start [repo…] / remote stop [repo…]Manage phone-drivable remote sessions across the fleet.

Why keywords instead of prompts

Two reasons this beats writing the same instructions from scratch each time:

  1. Consistency — ship it always means "test branch first, CI must pass,

never push straight to main," whether it's 9am or 2am and whether you or the autonomous runner typed it.

  1. Token economy — the extended tier only loads its protocol into

context when actually invoked, instead of every keyword's full instructions sitting in every session's system prompt.

Next: the Concepts page for the ideas behind agent mode's boot sequence, or the Gateway overview for the queue that schedule writes into.