| name | audit-progress |
| description | Audit implementation or project progress against the approved XMemo plan, checkpoint, decisions, and current evidence. Use when Claude must verify another agent's completion report, reconcile multi-agent claims, identify drift or blockers, and determine the exact next action. Never promote a claim to verified state without evidence. |
Audit Progress with XMemo
Measure reality against the approved baseline. Follow baseline → claims → evidence → reconcile → verdict → checkpoint.
1. Recover the governing baseline
Use recall_context, get_project_context, search_memory, and relevant todo reads to recover the approved plan, acceptance criteria, settled decisions, last verified checkpoint, and known blockers. State which artifact or memory is the baseline and flag ambiguity before auditing.
2. Normalize the progress claims
Break the report into atomic claims: changed, tested, deployed, reviewed, documented, blocked, or deferred. Preserve the source agent and artifact/version when known, but do not treat source identity as evidence.
3. Verify against current evidence
Inspect the live artifacts available to Claude: files, diffs, commits, tests, logs, deployments, or explicit user confirmation. Classify each claim as:
- verified — direct evidence satisfies the criterion;
- partial — some evidence exists, but the criterion is incomplete;
- claimed — asserted without adequate evidence;
- stale — once true but superseded by newer state;
- conflicting — sources disagree and evidence does not resolve them;
- blocked — a named dependency prevents verification or completion.
Absence of evidence is not proof of failure. A passing local test is not proof of deployment. A commit is not proof of runtime behavior.
4. Reconcile drift and conflicts
Resolve disagreement by exact scope, recency, provenance, and verifiable evidence. Current explicit user decisions override older memories. If a saved baseline is wrong, propose a correction; do not silently rewrite history.
5. Return a formal audit
Use this structure:
- Audit scope and baseline
- Verdict —
on_track, at_risk, blocked, or not_verified
- Claim matrix — claim, status, evidence, gap
- Plan drift
- Risks and blockers
- Acceptance gaps
- Exact next action
When the user requests a durable review artifact, write the report in the
repository's established documentation location using
assets/progress-audit-template.md. Keep the baseline, evidence, status, gaps,
and exact next action independently reviewable. Do not create or overwrite a
file when the request is report-only.
6. Update XMemo only from reconciled truth
With user authorization or an active save workflow:
- replace the working checkpoint with verified status via
update_state;
- complete or update TODOs only when their conditions are met;
- record a material milestone or handoff via
record_event;
- correct a durable claim via
update_memory;
- preserve a new durable lesson via
remember.
Never save the audit narrative as a raw transcript. Do not mark completion from an agent's confidence alone. Return a receipt that distinguishes verified updates from unresolved claims.
Read ../memory-steward/references/review-playbooks.md, ../memory-steward/references/workflows.md, and ../memory-steward/references/tool-routing.md for detailed review and routing contracts. Use assets/progress-audit-template.md for a requested Markdown audit artifact.
