audit

name	audit
description	Fresh-context verification of a phase deliverable or wave outcome. Spawns a fresh auditor subagent that re-reads the diff against the success criteria.
argument-hint	<phase-id|wave-id|commit-range>
user-invocable	true
disable-model-invocation	false

/audit

Canonical algorithm

1. Resolve target: phase id, wave id, or commit range.
2. Identify success criteria from the plan / phase spec and cite evidence with dense [N] references.
3. Dispatch the auditor subagent with paths, line numbers, criteria.
4. Parse the verdict; convert refutations into TODOs or new waves.
5. Render audit evidence through eawf audit show --md.

v0.4 cross-links

The auditor emits one EvidenceRecord per criterion (evidence_kind = gate | claim | decision). The aggregate verdict folds into the target wave / iter CloseReadiness projection — /ship reads the same projection so audit and ship share one source of truth on "is this iter actually closable?". Roles are pinned via RoleSpec (role="auditor"); fresh-context isolation stays a RoleSpec invariant, not an ad-hoc dispatch flag.

Pre-flight checklist

• The auditor must NOT have access to the parent conversation.
• Every quantitative claim must include source evidence and dense citation refs.
• The target iter is NOT yet closed. Iter close is gated on audit + polish + ship CI + PR review pass per the iter-phase-close-timing rule in AGENTS.md; /audit runs before that close.

Decision surfaces

On pass-with-followups: present the follow-up disposition (open backlog, open wave, defer) through AskUserQuestion. On fail: ask whether to halt the flow or open a remediation wave.

Output contract

Skill envelope with a per-criterion verdict table and an aggregate status (pass | pass-with-followups | fail).