Pwrl review

PWRL Review

Review code changes through a deterministic 4-phase pipeline: validate scope, prepare for review, analyze code quality/security/tests, and generate approval decision.

Interaction Method

• Use platform's ask_user_question, ask_user, ask_user_input, vscode/askQuestions or any available extension/tool for user interaction for all decisions
• Ask one question at a time
• Use multiple-choice questions when possible
• If input is empty, ask: "What code would you like to review? Provide a branch name, PR number, or leave blank for the current branch."
• Provide clear recovery suggestions when errors occur

Purpose

Provide comprehensive code review through systematic analysis:

• Validate Scope — Ensure changes match requirements, detect scope creep
• Prepare Review — Gather artifacts, establish baselines, configure analysis
• Analyze Code — Review quality, security, tests, documentation, integration
• Generate Report — Compile findings, determine approval status

Usage

/pwrl-review
/pwrl-review feature/U2-email-validation
/pwrl-review origin/dev

Architecture

Pure Skill Pipeline — Direct sequence of 4 micro-skills:

Input (branch/PR)
  ↓
Phase 1: pwrl-review-scope
  ├ Input: Source branch, requirements
  ├ Output: Scope artifact (scope_verdict, files_analyzed)
  ↓
Phase 2: pwrl-review-prepare
  ├ Input: Scope artifact
  ├ Processing: Diff gathering, baseline setup, tool configuration
  ├ Output: Prepare artifact (review_scope, tools_configured)
  ↓
Phase 3: pwrl-review-analyze
  ├ Input: Prepare artifact
  ├ Processing: Code quality, security, tests, docs, integration checks
  ├ Output: Analyze artifact (findings, issues_found by severity, integration_check)
  ↓
Phase 4: pwrl-review-report
  ├ Input: Analyze artifact
  ├ Processing: Format report, calculate verdict, get user approval
  ├ Output: Report artifact (verdict: approved/request-changes/rejected)
  ↓
COMPLETE

Phase Summary

Each phase is orchestrated sequentially. The orchestrator calls the micro-skill, receives the output artifact, validates it with a quality gate, and passes it to the next phase.

Phase 1: Scope (pwrl-review-scope) — Validate branch scope and file categorization. Output: Scope artifact.

Phase 2: Prepare (pwrl-review-prepare) — Setup review environment and configure analysis tools. Output: Prepare artifact.

Phase 3: Analyze (pwrl-review-analyze) — Review code quality, security, tests, docs, integration. Output: Analyze artifact.

Phase 4: Report (pwrl-review-report) — Compile findings and determine approval verdict. Output: Report artifact.

Phase 5: Sync Status (pwrl-review-sync-status) — Post findings to GitHub PR with comment, formal review, and labels. Output: Sync confirmation.

Quality Gates: Run /pwrl-phase-checkpoint review N [artifact-path] to validate each phase. See pwrl-phase-checkpoint for validation rules.

Interaction Mode Propagation

Interaction mode (detailed | smart | yolo) is set in Phase 1 (via pwrl-review-scope Step 0) and read at the start of each subsequent phase. The mode is stored in the scope artifact's interactionMode field and propagated through every downstream phase.

• detailed — Pause at every phase transition (Scope → Prepare → Analyze → Report → Sync); show generated artifacts; require explicit approval to proceed. Best for complex reviews, unfamiliar code, and high-stakes changes.
• smart — Phases run automatically; pause only when the next phase produces a HIGH-risk operation (e.g., posting a formal GitHub review, labeling a PR with a release-blocker). v1 simplification: behaves like Yolo with a single confirmation prompt at workflow start.
• yolo — Every phase runs automatically; only the final report is shown. Fastest. Best for routine, well-understood reviews.

Future refinement: In Yolo mode, pwrl-review-report should auto-approve the review unless CRITICAL issues are found (currently it always asks). This behavior change is out of scope for plan 2026-06-29-001; see docs/learnings/pattern/interaction-mode-three-mode-propagation-2026-06-29.md §"Future Refinements".

Exception: Error recovery steps always pause the pipeline, regardless of mode.

Verdict Criteria

See verdict-criteria.md for comprehensive approval matrices, quality score calculations, and decision logic for all verdicts (APPROVED, REQUEST CHANGES, REJECTED).

Review Lenses

• Correctness — Logic errors, edge cases, error handling
• Security — Input validation, injection, auth, secrets
• Maintainability — Clarity, complexity, duplication, abstraction
• Testing — Coverage, scenarios, assertions, isolation
• Performance — Efficiency, resource usage, slow operations
• Integration — No broken imports, circular deps, regressions

Support Files

• pwrl-review-scope/references/scope-validation-protocol.md — Phase 1 specification
• pwrl-review-prepare/references/prepare-review-protocol.md — Phase 2 specification
• pwrl-review-analyze/references/analyze-code-protocol.md — Phase 3 specification
• pwrl-review-report/references/report-generation-protocol.md — Phase 4 specification

When to Use

• After completing implementation work
• Before creating a pull request
• To review someone else's code changes
• When you need confidence before merging
• After completing a task from pwrl-work

Control Tokens

Optional tokens control review depth and execution mode:

• depth:fast — Quick scan, minimal findings, no artifacts
• depth:standard — Balanced review (default), prefer subagents if available
• depth:deep — Thorough audit with parallel subagents and artifacts
• subagents:on|off — Force subagent mode (default: auto)

Tokens are stripped before interpreting remainder as branch/PR/commit/URL. Error if conflicting depth tokens provided.

Micro-Skill References

For detailed implementation and workflow specifications for each phase, see:

• pwrl-review-scope — Phase 1: Scope validation
• pwrl-review-prepare — Phase 2: Environment preparation
• pwrl-review-analyze — Phase 3: Code analysis
• pwrl-review-report — Phase 4: Report generation

Each micro-skill is complete and self-contained with its own workflow, error handling, and testing coverage.

Output

• Code review report with findings grouped by severity
• Verdict: APPROVED | REQUEST CHANGES | REJECTED
• Artifacts from each phase (scope, prepare, analyze, report)