launch-charter
launch-charter
The boundary the whole launch crew operates inside. Capability is the variable; trust is the constant. An agent may become more capable — it may NOT become less governed.
Why this exists
Autonomous building drifts toward "self-deregulating" if nothing pins the rails down. An agent that can edit its own guardrails is not autonomous — it is unsupervised, which is different and dangerous. This charter is the one file the crew loads before doing anything, and the one file no agent may rewrite.
It does not invent new rules or new controls — it binds the generic launch-crew rails to the rails and controls already enforced in this repo: AGENTS.md, CLAUDE.md, the kill-switch-binding, pii-redactor, the tao-judge/tier-evaluator gate, and the meta-curator self-improvement loop — so there is a single source of truth.
Triggers
- Step 1 of
ship-it, always, before any other launch skill. - The start of any cron / autonomy-loop run.
- Before any builder subagent is dispatched to a sandbox.