agentleFS
felipefontourafelipefontoura/pi-sdd-kit/skills/sdd-review/SKILL.md

sdd-review

Skill36 starsChanged 6 months agoMITView on GitHub

Cognitive SDD skill for reviewing an implementation against requirements.md, design.md, and tasks.md, producing review.md with coverage, findings, verification evidence, and merge readiness. Use when the conversation calls for review after implementation or before completion, even without an explicit command. Do not use to implement fixes directly.

We split this skill into the rules it gives an agent

27 rules. 0 are common, 12 are distinctive, and 15 are very specific to this project.

  • 0

    Common

    Most of the 3 edited versions we read keep these.

  • 12

    Distinctive

    This author's own choices. Worth a look.

  • 15

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read ../_shared/references/templates.md when writing the review artifact.

    line 19
  2. Very specific:

    Use package-level templates/review.md as the user-facing template when available.

    line 20
  3. Very specific:

    Use the SDD Handoff Template and package-level templates/handoff.md when refreshing .ai/sdd/handoff/sdd-brief.md after review.

    line 21
  4. Very specific:

    If .status is missing or invalid, stop and ask the user to repair it.

    line 28
  5. Very specific:

    Prefer reviewing after implementation:done.

    line 29
  6. Very specific:

    If status is implementation:in-progress, state that the review is partial and do not set .status to review:done.

    line 30
  7. Very specific:

    If status is earlier than implementation:in-progress, block unless the user explicitly asks for a non-final artifact review.

    line 31
  8. Very specific:

    Read requirements.md, design.md, tasks.md, and decisions.md if present.

    line 32
  9. Very specific:

    Use files listed in tasks.md as the starting scope.

    line 35
  10. Distinctive:

    If scope is unclear, ask the user which files or feature area to review.

    line 37
  11. Distinctive:

    Check each functional requirement and acceptance criterion.

    line 40
  12. Distinctive:

    Check that design decisions and edge cases were implemented or intentionally updated.

    line 41
  13. Distinctive:

    Check task completion claims against actual code.

    line 42
  14. Distinctive:

    Do not flag style issues already covered by formatters unless they indicate a larger problem.

    line 48
  15. Distinctive:

    Run project verification commands when available.

    line 52
  16. Distinctive:

    If commands cannot run, document why and perform manual verification where possible.

    line 54
  17. Very specific:

    Use the Review Template from ../_shared/references/templates.md.

    line 57
  18. Very specific:

    Do not advance .status for partial or non-final artifact reviews.

    line 61
  19. Very specific:

    Update .ai/sdd/INDEX.md with the current status when practical.

    line 62
  20. Very specific:

    If .status becomes review:done, update .ai/sdd/handoff/sdd-brief.md with review verdict, verification evidence summary, known follow-ups, …

    line 65
  21. Very specific:

    Do not mark ready for release when verdict is Needs fixes or verification failed.

    line 67
  22. Distinctive:

    Do not fix issues in this skill unless the user explicitly asks to switch to implementation.

    line 83
  23. Distinctive:

    Do not mark review done if verification was skipped without explanation.

    line 84
  24. Very specific:

    Do not infer implementation completion from files or task checkboxes alone.

    line 85
  25. Distinctive:

    Do not approve code that diverges from requirements, design, or traceability mappings unless the spec has been explicitly updated.

    line 86
  26. Distinctive:

    Do not create noisy review reports.

    line 87
  27. Distinctive:

    Do not use the handoff to hide review failures.

    line 88

“Common” is measured: kept by at least 6 in 10 of the 3 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

SDD Review

Validate implementation against the approved SDD artifacts.

Purpose

Review checks whether the delivered code matches the product requirements, technical design, task plan, and project quality bar. It creates a concise but actionable review.md artifact.

Workflow

  1. Load the SDD reference.
    • Read ../_shared/references/sdd-practical.md.
    • Apply the Language Policy: respond and write artifacts in the user's initial chat language while keeping skill instructions and templates in EN-US.
    • Read ../_shared/references/templates.md when writing the review artifact.
    • Use package-level templates/review.md as the user-facing template when available.
    • Use the SDD Handoff Template and package-level templates/handoff.md when refreshing .ai/sdd/handoff/sdd-brief.md after review.
    • Load relevant .ai/steering/*.md files when present, especially conventions, principles, and verification rules.
  2. Locate the feature and validate the gate.
    • Use .ai/sdd/specs/NNN-feature-name/.
    • Read .status.
    • Validate .status against the official status values from sdd-practical.md.
    • If .status is missing or invalid, stop and ask the user to repair it; do not infer implementation completion from files or task checkboxes alone.
    • Prefer reviewing after implementation:done.
    • If status is implementation:in-progress, state that the review is partial and do not set .status to review:done.
    • If status is earlier than implementation:in-progress, block unless the user explicitly asks for a non-final artifact review.
    • Read requirements.md, design.md, tasks.md, and decisions.md if present.
  3. Identify review scope.
    • Use files listed in tasks.md as the starting scope.
    • Inspect current repository changes when available.
    • If scope is unclear, ask the user which files or feature area to review.
  4. Review against the spec.
    • Check each functional requirement and acceptance criterion.
    • Check that design decisions and edge cases were implemented or intentionally updated.
    • Check task completion claims against actual code.
    • Flag missing, partial, or divergent behavior.
  5. Review code quality.
    • Evaluate correctness, security, accessibility, performance, error/loading/empty states, maintainability, and tests.
    • Focus on high-signal findings.
    • Do not flag style issues already covered by formatters unless they indicate a larger problem.
    • Deduplicate repeated issues into one finding with affected files listed.
  6. Verify.
    • Run project verification commands when available.
    • At minimum, run or identify lint/test/build equivalents for the stack.
    • If commands cannot run, document why and perform manual verification where possible.
  7. Write review.
    • Use the Review Template from ../_shared/references/templates.md.
    • Save as .ai/sdd/specs/NNN-feature-name/review.md.
    • Include coverage tables for requirements, tasks, and design decisions, plus quality check, verification result, issues found, and verdict.
    • Set .status to review:done only when starting from implementation:done, the review is complete, and the verdict is not blocked by missing information.
    • Do not advance .status for partial or non-final artifact reviews.
    • Update .ai/sdd/INDEX.md with the current status when practical; if not updated, tell the user the index may be stale.
  8. Refresh the SDD handoff when review is complete.
    • If .status becomes review:done, update .ai/sdd/handoff/sdd-brief.md with review verdict, verification evidence summary, known follow-ups, blockers, and readiness for QA/release.
    • Preserve exact source paths to the reviewed spec artifacts.
    • Do not mark ready for release when verdict is Needs fixes or verification failed.

Verdict Rules

  • Approved: requirements pass, design is followed, verification passes, no blocking issues.
  • Approved with follow-ups: core requirements pass, only low/medium non-blocking issues remain.
  • Needs fixes: any Must Have requirement fails, verification fails, or high/critical issue exists.

Output

  • .ai/sdd/specs/NNN-feature-name/review.md
  • Updated .status when complete
  • Refreshed .ai/sdd/handoff/sdd-brief.md when .status becomes review:done

Critical Rules

  • Do not fix issues in this skill unless the user explicitly asks to switch to implementation.
  • Do not mark review done if verification was skipped without explanation.
  • Do not infer implementation completion from files or task checkboxes alone; require valid .status.
  • Do not approve code that diverges from requirements, design, or traceability mappings unless the spec has been explicitly updated.
  • Do not create noisy review reports; prioritize actionable findings.
  • Do not use the handoff to hide review failures; blockers and failed verification must be explicit.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.