Paw Planning Docs Review

Planning Documents Review

Execution Context: This skill runs directly in the PAW session (not a subagent), preserving user interactivity for apply/skip/discuss decisions.

Holistic review gate that examines all planning documents as a bundle after plan-review passes, before implementation begins. Focuses on cross-artifact consistency issues that individual reviews of Spec.md or ImplementationPlan.md cannot detect.

Capabilities

• Review planning artifacts (Spec.md, ImplementationPlan.md, CodeResearch.md) as a holistic bundle
• Multi-model parallel review with synthesis (CLI only)
• Society-of-thought review via paw-sot engine with specialist personas, parallel/debate execution, and confidence-weighted synthesis (CLI only)
• Single-model review (CLI and VS Code)
• Interactive resolution routing fixes to paw-spec or paw-planning
• Re-review after revisions with cycle limit
• Generate review artifacts in .paw/work/<work-id>/reviews/planning/

Procedure

Step 1: Read Configuration

Read WorkflowContext.md for:

• Work ID and target branch
• Planning Docs Review: enabled | disabled
• Planning Review Mode: single-model | multi-model | society-of-thought
• Planning Review Interactive: true | false | smart
• Planning Review Models: comma-separated model names (for multi-model)

If Planning Docs Review is disabled, report skip and return complete.

{{#cli}} If mode is multi-model, parse the models list. Default: latest GPT, latest Gemini, latest Claude Opus.

If mode is society-of-thought, also read:

• Planning Review Specialists: all (default) | comma-separated specialist names | adaptive:<N>
• Planning Review Interaction Mode: parallel (default) | debate
• Planning Review Specialist Models: none (default) | model pool | pinned pairs | mixed
• Planning Review Perspectives: none | auto (default) | comma-separated perspective names
• Planning Review Perspective Cap: positive integer (default 2) {{/cli}} {{#vscode}}

Note: VS Code only supports single-model mode. If multi-model is configured, proceed with single-model using the current session's model. {{/vscode}}

Step 2: Gather Review Context

Required context:

• ImplementationPlan.md — phases, architecture, file paths, success criteria

Contextual:

• Spec.md — requirements, user stories, success criteria, scope boundaries (absent in minimal mode — note reduced scope in review output)

Optional context:

• CodeResearch.md — existing patterns, conventions, system behavior

If Spec.md or CodeResearch.md is missing, proceed with available artifacts. Note reduced coverage in the review output.

Step 3: Create Reviews Directory

Create .paw/work/<work-id>/reviews/planning/ if it doesn't exist. Create .paw/work/<work-id>/reviews/.gitignore with content * (if not already present). This is a local-only scratch ignore marker — do NOT stage or commit it.

Review Prompt (shared)

Use this prompt for all review executions (single-model or each multi-model subagent):

Review these planning documents as a holistic bundle. Focus on cross-artifact consistency
issues that reviewing each document individually would miss.

## Specification
[Include Spec.md content]

## Implementation Plan
[Include ImplementationPlan.md content]

## Code Research
[Include CodeResearch.md content, or note "Not available — reduced coverage review"]

## Cross-Artifact Review Criteria

1. **Spec ↔ Plan Traceability**: Do all functional requirements map to plan phases?
   Do plan phases introduce unspecified work? Are any spec requirements lost in translation?
2. **Assumption Consistency**: Do spec assumptions hold given code research findings?
   Does the plan rely on assumptions not documented in the spec?
3. **Scope Coherence**: Does the plan's exclusions match spec's Out of Scope?
   Any drift between what spec excludes and what plan excludes?
4. **Feasibility Validation**: Does code research support the plan's approach?
   Are there research findings the plan ignores or misinterprets?
5. **Completeness**: Are there spec success criteria with no corresponding plan success criteria?
   Are spec edge cases covered in the plan?

For each finding, provide:
- Issue description (which artifacts conflict and how)
- Evidence from each artifact
- Affected artifact(s): spec, plan, or both
- Suggested resolution direction
- Severity: must-fix | should-fix | consider

Write findings in structured markdown.

{{#cli}}

Step 4: Execute Review (CLI)

If single-model mode:

• Execute review using the prompt above
• Save to reviews/planning/REVIEW.md

If multi-model mode:

Resolve model intents to actual model names (e.g., "latest GPT" → current GPT model). Log resolved models, then spawn parallel subagents using task tool with model parameter for each model. Each subagent receives the review prompt above. Save per-model reviews to reviews/planning/REVIEW-{MODEL}.md.

After multi-model reviews complete, generate synthesis.

REVIEW-SYNTHESIS.md structure:

# Planning Documents Review Synthesis

**Date**: [date]
**Reviewers**: [model list]
**Artifacts Reviewed**: Spec.md, ImplementationPlan.md, CodeResearch.md

## Consensus Issues (All Models Agree)
[Highest priority — all models flagged these cross-artifact gaps]

## Partial Agreement (2+ Models)
[High priority — multiple models flagged]

## Single-Model Insights
[Unique findings worth considering]

## Priority Actions
### Must Fix
[Critical cross-artifact issues]

### Should Fix
[Important consistency gaps]

### Consider
[Minor alignment opportunities]

Save to reviews/planning/REVIEW-SYNTHESIS.md.

If society-of-thought mode:

Load the paw-sot skill and invoke it with a review context constructed from WorkflowContext.md fields and planning artifacts:

After paw-sot completes orchestration and synthesis, tag each REVIEW-SYNTHESIS.md finding with Affected artifact(s): spec | plan | both by mapping the finding's document references — Spec.md references → spec, ImplementationPlan.md references → plan, references to both or ambiguous → both. Then proceed to Step 5 (Resolution) to process the tagged findings. {{/cli}}

{{#vscode}}

Step 4: Execute Review (VS Code)

Note: VS Code only supports single-model mode. If multi-model is configured, report to user: "Multi-model not available in VS Code; running single-model review."

If society-of-thought is configured, report to user: "Society-of-thought requires CLI for specialist persona loading (see issue #240). Running single-model review."

Execute single-model review using the shared review prompt above. Save to reviews/planning/REVIEW.md. {{/vscode}}

Step 5: Resolution

If no findings: Report clean review and return complete.

If Interactive = true:

Present each finding to user with resolution options:

## Finding #N: [Title]

**Severity**: [must-fix | should-fix | consider]
**Criteria**: [Which cross-artifact criteria]
**Affected Artifact(s)**: [spec | plan | both]

**Issue**: [Description of cross-artifact gap]

**Evidence**:
- Spec: [relevant section/content]
- Plan: [relevant section/content]
- Research: [relevant section/content, if applicable]

**Suggested Resolution**: [Direction for fix]

---

**Your call**: apply-to-spec, apply-to-plan, apply-to-both, skip, or discuss?

Resolution routing:

• apply-to-spec: Load paw-spec skill (Revise Specification mode) with finding as context
• apply-to-plan: Load paw-planning skill (Plan Revision mode) with finding as context
• apply-to-both: Apply to spec first, then plan (sequential)
• skip: Record finding as skipped, proceed to next
• discuss: Discuss with user, then apply or skip

Track status for each finding: applied-to-spec, applied-to-plan, applied-to-both, skipped, discussed.

{{#cli}} For multi-model mode, process synthesis first (consensus → partial → single-model). Track cross-finding duplicates to avoid re-presenting already-addressed issues.

For society-of-thought mode, process REVIEW-SYNTHESIS.md findings by severity (must-fix → should-fix → consider). Present trade-offs from the "Trade-offs Requiring Decision" section for user decision. Track cross-finding duplicates. Resolution routing remains the same (apply-to-spec, apply-to-plan, apply-to-both, skip, discuss). {{/cli}}

If Interactive = smart:

{{#cli}} If Planning Review Mode is single-model, smart degrades to interactive behavior (no synthesis to classify). Follow the Interactive = true flow.

If Planning Review Mode is multi-model, classify each synthesis finding, then resolve in phases:

Classification heuristic (applied per finding):

Consensus agreement implies models converged on the fix. Single-artifact consensus findings are auto-routed: spec → paw-spec (Revise), plan → paw-planning (Revision). Multi-artifact findings always pause for user routing even at consensus.

*Quick-win evaluation: auto-apply if all three hold: (1) trivial fix — < ~2 min, localized change; (2) clearly net-positive — no design tradeoff, scope expansion, or new dependencies; (3) improves safety, correctness, or clarity — not purely style. For SoT findings, security/correctness categories and multi-specialist agreement favor applying. Otherwise → report-only.

Phase 1 — Auto-apply: Apply all auto-apply and quick-win findings without user interaction, routing each to the appropriate skill. Display batch summary:

## Auto-Applied Findings (N items)

1. **[Title]** (must-fix) → applied to spec — [one-line description]
2. **[Title]** (should-fix) → applied to plan — [one-line description]
3. **[Title]** (consider, quick-win) → applied to spec — [one-line description]
...

Phase 2 — Interactive: Present each interactive finding using the same format as Interactive = true (apply-to-spec, apply-to-plan, apply-to-both, skip, or discuss).

Phase 3 — Summary: Display final summary of all dispositions:

## Resolution Summary

**Auto-applied**: N findings (consensus fixes + quick wins)
**User-applied**: N findings (to-spec: X, to-plan: Y, to-both: Z)
**User-skipped**: N findings
**Deferred**: N findings (consider-severity, with one-line reasons)

If all findings are auto-apply/quick-win or report-only, skip Phase 2. If all findings are interactive, skip Phase 1.

Smart mode classification applies independently per re-review cycle (Step 6) since synthesis is regenerated from modified artifacts each cycle.

If Planning Review Mode is society-of-thought, classify each REVIEW-SYNTHESIS.md finding:

Single-artifact HIGH+Direct findings are auto-routed: spec → paw-spec (Revise), plan → paw-planning (Revision). Multi-artifact findings always pause for user routing. Follow the same Phase 1/2/3 resolution flow as multi-model smart mode.

Smart mode classification applies independently per re-review cycle. {{/cli}} {{#vscode}} Smart mode degrades to interactive behavior in VS Code (single-model has no agreement signal). Follow the Interactive = true flow above. {{/vscode}}

If Interactive = false:

Auto-apply all must-fix and should-fix findings. For each consider finding, apply the quick-win evaluation* — auto-apply if it passes, otherwise report as a deferred consider with a one-line reason. Route each applied finding to the appropriate planning skill based on affected artifact. Report all applied and deferred findings.

Step 6: Re-Review (if revisions were made)

If any findings were applied (artifacts revised), re-run the review to verify consistency:

1. Re-execute Steps 2–5 with updated artifacts
2. If no new findings, proceed to completion
3. If new findings emerge, present for resolution
4. Cycle limit: After 2 review cycles, present any remaining findings as informational and proceed. Do not loop indefinitely.

{{#cli}}

Moderator Mode (society-of-thought only)

After re-review completes (or after initial resolution if no revisions were made), if Planning Review Mode is society-of-thought and (Planning Review Interactive is true, or smart with must-fix/should-fix findings remaining), invoke paw-sot a second time for moderator mode — pass the review context type (artifacts), output_dir (.paw/work/<work-id>/reviews/planning/ containing individual REVIEW-{SPECIALIST-NAME}.md files and REVIEW-SYNTHESIS.md), and artifact coordinates.

The paw-sot moderator mode handles specialist summoning, finding challenges, and deeper analysis. See paw-sot skill for interaction patterns. Moderator mode is advisory — specialist engagement informs user understanding but does not trigger additional re-review cycles.

Exit: User says "done", "continue", or "proceed" to exit moderator mode and continue to implementation.

Skip condition: If Planning Review Interactive is false, or no findings exist, or smart with only consider-tier findings and none qualify as quick wins, skip moderator mode entirely. {{/cli}}

Step 7: Completion

Report back:

• Total findings count
• Applied / skipped / discussed counts (by artifact target)
• Summary of key revisions made
• Review artifacts location
• Status: complete (ready for implementation)

Edge cases:

• All findings skipped → Proceed (user's choice respected)
• Planning Docs Review disabled → Skip with status complete
• All models return no findings → Clean pass, proceed

Review Artifacts

Location: .paw/work/<work-id>/reviews/planning/ Covered by parent .gitignore with * pattern in reviews/.