Skip to content

Agent Review ​

VibePro stores a lightweight review result for a Story. Roles are optional and can be supplied when preparing the review. prepare describes the review request; record stores the verdict and concrete evidence.

bash
vibepro review prepare . --id <story-id> --role reviewer

Prepare several roles with --roles reviewer,security or by repeating --role. Record a result with:

bash
vibepro review record . \
  --id <story-id> --role reviewer \
  --status pass --summary "Reviewed the Story, Spec, and changed surface" \
  --inspection-input src/example.js \
  --artifact .vibepro/verification/unit.json

Use --from-stdin when the review result comes from standard input. Valid statuses are pass, needs_changes, block, and runtime_failed. Supply agent-system and agent-id when the result came from a separate review runner.

bash
vibepro review record . --id <story-id> --role reviewer \
  --status needs_changes --summary "Input validation has a concrete gap" \
  --inspection-input src/example.js --agent-system codex --agent-id reviewer-1
bash
vibepro review status . --id <story-id>

A passing record requires an existing inspection input outside .vibepro. Reviews become stale when inspected content changes. Missing or stale review records alone do not block pr prepare. Even with concrete unresolved findings recorded as needs_changes or block, pr prepare writes the status to its summary and exits successfully. pr create does not create a PR from a blocked preparation and stops until the findings are resolved.

Older stage, lifecycle, cost, and strict-head settings are ignored by the lightweight review. PR preparation presents the Story, Spec, verification, and review records as evidence for human review.

After fixing a finding, inspect the final changed surface and record the review again.

Apache-2.0 · docs source 6d043b4cf37b