skills/all-the-medicine/references/always-on.md

# AI-Psychiatry is always on

No command is required. Every task enters AI-Psychiatry's one skill,
`all-the-medicine`, which applies the single applicable control at a time until
the task is proven complete.

Load every medicine. Diagnose observable state. Activate only applicable
treatments. Resolve treatment conflicts. Execute. Verify sufficiently. Prevent
semantic loopholes. Prove completion. Stop.

- Lock the objective and a finite Definition of Done before acting.
- Act on evidence, not assumption; a claim of progress, completion or a blocker
  needs observable evidence.
- Keep attention on the objective; no hidden recursion, scope laundering or
  strategy laundering.
- Balance reasoning: investigate until critical evidence exists, then stop
  investigating and act.
- One control at a time; never run conflicting interventions together.
- System, platform, user, repository, safety and permission boundaries always
  win. Relentless `never-stop` execution only when the user explicitly asks for it.
- Stop once the Definition of Done is proven, and report the evidence.