iot-ailisted

# iot-ai Use one natural-language goal. Do not make the operator translate normal language into internal flags. Examples: ```text /iot-ai finish all critical PMD tasks and keep working until each is complete, awaiting Founder, or externally blocked /iot-ai prüfe die Fehler, repariere sie vollständig und liefere den Abschlussbericht ``` The runtime compiles the conversation into a versioned intent contract, resolves prior task references, selects the authoritative task backend, and runs this loop: ```text Task intake → validation → full hybrid Meeting → Multi-Coder implementation → deterministic tests → failure Meeting → bounded repair → independent review → audit → re-plan when evidence is incomplete → terminal report ``` Rules: - Execution verbs mean “continue to a truthful terminal state” by default. - Use every eligible required coder/model seat at material planning, review and release gates; record outages honestly. - One successful model is not Multi-Coder consensus. - Never silently read PMD/PRCS databases; use the authenticated versioned PMD API adapter. - Progress is telemetry, not completion authority. - Never move audit-failing work to `awaiting_founder`. - Founder Accept/Reject/Rework remains human-only. - Stop only for a real human, safety, authority, external-service or bounded-budget gate. - Always produce task, provider, iteration, evidence and blocker tables.