emperor-verify

Emperor Time — VERIFY / review / critique. Claim audit, eight-count self-critique, isolated hetero-critique, request-review HARD-GATE, receive-review HARD-GATE, verification-before-completion / eviden

SKILL.md

GitHub

Emperor Verify (Judgment wrapper)

MUST — critique eight-count HARD-GATE before G4

Before opening G4, run scripts/emperor critique <task-dir> (or self-critique). Doctrine: chains/judgment-chain/self-critique.md. Missing axes / empty Checked cells / critique-file-present theater → exit 1. G4 calls scripts/lib/critique.py; a critique file alone is theater.

MUST — claim-audit HARD-GATE before G4

Before opening G4, run scripts/emperor claim-audit <task-dir> (or judgment-audit). Doctrine: chains/judgment-chain/claim-audit.md. Missing CLAIM AUDIT line / unfinished HYPOTHESIS|TESTED → exit 1. G4 calls scripts/lib/claim_audit.py; a critique file alone is theater.

MUST — request-review checklist first

Before merge to main, after a major feature, and after each subagent-driven task, open skills/emperor-verify/request-review-checklist.md (Chain Jail leaf from Superpowers requesting-code-review → When / How / Act-on-feedback only) and/or run scripts/emperor review (prints the mechanical REVIEW / STEP / MUST card).

No author self-review in place of dispatch. Do not load whole requesting-code-review; ET + emperor-verify orchestrate.

MUST — receive-review checklist when acting on feedback

When receiving code-review feedback (human partner, external reviewer, or hetero-critique) — before implementing suggestions — open skills/emperor-verify/receive-review-checklist.md (Chain Jail leaf from Superpowers receiving-code-review → The Response Pattern / Forbidden Responses / When To Push Back only) and/or run scripts/emperor receive (prints the mechanical RECEIVE / STEP / MUST card).

No blind implement. No performative agreement. Do not load whole receiving-code-review; ET + emperor-verify orchestrate.

MUST — evidence checklist before completion claims

Before claiming tests pass, bug fixed, build green, requirements met, agent done, or any satisfaction/completion phrasing — and before commit/PR/deliver — open skills/emperor-verify/verification-checklist.md (Chain Jail leaf from Superpowers verification-before-completion → The Iron Law / The Gate Function only) and/or run scripts/emperor evidence (prints the mechanical EVIDENCE / STEP / MUST card).

No completion claim without a fresh proving command in this turn. Do not load whole verification-before-completion; ET + emperor-verify orchestrate.

Hard rules

  1. Run scripts/emperor evidence → quote EVIDENCE checklist=yes before any completion / pass / fixed / done claim. Advance with scripts/emperor evidence --advance N N+1 (skips fail). Unverified claim → scripts/emperor evidence --reject-unverified (HARD-GATE exit 1).
  2. Run scripts/emperor review → quote REVIEW checklist=yes. Advance with scripts/emperor review --advance N N+1 (skips fail). Self-review skip → scripts/emperor review --reject-self-review (HARD-GATE exit 1).
  3. When acting on review feedback, run scripts/emperor receive → quote RECEIVE checklist=yes. Advance with scripts/emperor receive --advance N N+1 (skips fail). Blind implement → scripts/emperor receive --reject-blind-implement (HARD-GATE exit 1).
  4. Read chains/judgment-chain/SKILL.md and select one aspect: claim-audit.md | self-critique.md | hetero-critique.md | gatekeeping.md | verdicts-and-breaches.md.
  5. Claim ledger: every VERIFIED row needs quoted evidence. Cross-agent rows start CONJECTURE.
  6. Self-critique uses templates/critique.md. "No findings" without named commands/paths is a fail.
  7. Hetero-critique: run scripts/emperor review-pack <task-dir> <base> <head> (Python core scripts/lib/review_pack.py; thin review-pack.sh / review-pack.ps1; isolation HARD-GATE --reject-unisolated / --reject-author-diary / --check-isolation) and dispatch the pack to a different context (Steal Chain worker or a fresh subagent). The builder does not write the hetero verdict.
  8. Act on Critical immediately; Important before proceed; Minor noted; pushback only with quoted evidence (run emperor receive when implementing feedback).
  9. Run scripts/gate.sh g4 <task-dir> and quote the tail.
  10. Deliver only after scripts/emperor verdict <task-dir> and scripts/gate.sh g5 <task-dir> (verdict.py: no empty/theater Breach Register; Verdict cites claim audit / critique / hetero).

Proportionality / anti-loop: honor effort_class caps via scripts/emperor proportionality --check-proportionality <task-dir> (HARD-GATE --reject-over-verify; G4 records+checks). Missing class on critique/finish/grill → tiny hard cap (bump_and_check). Tiny asks do not re-run the museum of gates. Idle Steal/Jail/Holy SKIP (vacuous — no activity) is separate.