Main-side verification checks — pre-commit test / sensitive-file blocking, discipline conformance, novel-trap detection. Pure verify-and-flag; persistence of findings is scribe's job, intent-artifact match is witness's.
범용 자율 루프 — 기획→정의→실행→평가→판단→기록→반복. 에스컬레이션 판단 포함.
Punch list — completeness and residue check before delivery. Finds what's missing that should exist (omission) and what's left behind that shouldn't (refactor residue). Reconstructs intent, overlays domain checklist, scans for stale references.
Post-work persistence layer — observes completed work and writes what should outlive the session. Bridge notes, journal, progress, project rules discovered from user corrections, novel traps detected by guard, witness verdicts.
Devil's advocate for reasoning — discovery-led counter-arguments against interpretations, plans, and irreversible decisions. Locates the load-bearing assumption via CCT, then drives depth into the one axis where the claim is actually thin.
Trigger-reactive discovery-led depth — when a System 1 signal surfaces (blast radius, shared code, infra, auth, migration), pull depth into side-effect, dependency, assumption, and rollback checks. Not a mode; a reaction.