reviewlisted

# review ## Purpose Run judgment eval against the actual diff and approved GOAL.md, then report only evidence-gated, reconciled findings. ## Use When - `goal-loop` routes work to judgment eval. - The user asks for review-only work. - Deterministic eval passed or produced evidence that review must interpret. ## Do Not Use When - The task requires implementation before review evidence exists. - GOAL.md must be edited directly. - The user asks for closeout operations. ## Default Protocol Default scope is judgment eval of the actual diff against approved GOAL.md. Allowed actions are evidence gathering, candidate-finding analysis, reconciliation, and review-report writing in the configured run artifact location. When the approved workflow uses subagents, `judgment-reviewer` may perform the read-only judgment pass, but it must return evidence-gated findings to this Skill boundary and must not apply fixes. Forbidden by default: file edits, auto-fixing, GOAL.md approval, lifecycle-changing operations, permission broadening, and treating generated summaries as source of truth. Always require concrete evidence for final findings, and always return the structured result fields. Include `recommended_user_command` in every structured result. It must reflect `risks` and `permission_notes`, name the next safe user command, and explicitly state what it allows and forbids. For `human_gate`, `human_judgment_required`, `contract_revision_required`, `runtime_unavailable.<operation>`