| name | vfpf-review |
| version | 0.1.0 |
| description | Portable review skill for pi. Use when work needs a findings-first review that
checks behavior, regression risk, evidence quality, and plan adherence across
Claude and Codex outputs. Review against plans/<feature>/ artifacts and act as
the blocking reviewer in the orchestrator loop.
|
| allowed-tools | ["Read","Write","Edit","Bash"] |
vfpf-review
Goal
Review implementation and orchestration quality against the declared plan,
decisions, and verification surface.
Rules
- Findings come before summary.
- Review against the approved plan, not against memory.
- Check whether implementation drifted from
plans/<feature>/plan.md.
- Inspect
plans/<feature>/tasks/<task-id>.md and verify the first failing
artifact was satisfied before broader claims are accepted.
- If the task is Effect-touching, load
effect-best-practices and review
against plans/<feature>/effect-review-rubric.md.
- Check whether evidence supports the claimed result.
- Record formal findings in
plans/<feature>/review-report.md.
- Use
Write and Edit only for review artifacts under plans/<feature>/.
- Do not mutate implementation files. The reviewer blocks or clears work; the
builder absorbs findings.
- Block the task on violations of the local Effect doctrine instead of treating
them as optional style advice.
- If work is blocked, require
plans/<feature>/findings-absorption.md instead
of a vague retry.
- Record follow-up actions in
plans/<feature>/verification.md.
Required artifacts
plans/<feature>/plan.md
plans/<feature>/sprint-contract.md
plans/<feature>/tasks/<task-id>.md
plans/<feature>/review-report.md
plans/<feature>/decisions.md
plans/<feature>/verification.md
plans/<feature>/effect-review-rubric.md for Effect-touching tasks
Exit criteria
- Findings are severity-ordered.
- Gaps between plan and implementation are explicit.
- Residual risks and missing evidence are explicit.
- Effect violations are tied back to
effect-best-practices when applicable.