v1.3.2 — The Plan Quality Release

Complete redesign of meow:plan-creator to match/exceed ClaudeKit-Engineer's ck-plan. Comparative red-team analysis across 15 dimensions drove every design decision.

What Changed

Step-File Architecture

Replaced monolithic 7-step SKILL.md with proper step-file workflow:

SKILL.md (thin entry, 102 lines)
  → workflow.md (step sequence + variable table)
    → step-00 (scope challenge + user input)
    → step-01 (bounded research)
    → step-02 (codebase analysis)
    → step-03 (draft plan + phase files)
    → step-04 (red team + validation + Gate 1)
    → step-05 (task hydration + checkpoint)

Each step loaded JIT — only the active step consumes context tokens.

Multi-File Phase Output

Plans are no longer single monolithic files:

tasks/plans/YYMMDD-name/
├── plan.md              ← overview (≤80 lines, richer YAML frontmatter)
├── research/            ← bounded researcher reports (hard mode)
├── phase-01-name.md     ← 12-section detail template
├── phase-02-name.md
├── .plan-state.json     ← checkpoint for cross-session resume
└── ...

12 required sections per phase: Context Links, Overview, Key Insights, Requirements, Architecture, Related Code Files, Implementation Steps, Todo List, Success Criteria, Risk Assessment, Security, Next Steps.

Scope Challenge (Step 0)

Trivial (≤2 files, ≤1h)  → "Use /meow:fix" → EXIT
Simple  (≤5 files, ≤4h)  → fast mode → plan.md only
Complex (>5 files, >4h)  → hard mode → full pipeline + phases

Hard mode asks user to choose scope: EXPANSION (research deep, up to 7 phases) / HOLD (standard) / REDUCTION (strip to MVP, 2-3 phases).

Plan Red Team (Hard Mode)

2 adversarial personas review the plan BEFORE validation interview:

• Assumption Destroyer — finds unstated assumptions, fragile dependencies
• Scope Complexity Critic — finds over-engineering, YAGNI violations

Reuses v1.3.1 persona prompts from meow:review. Max 10 findings. User accepts/rejects each. Accepted findings update affected phase files.

Research Integration

Research findings are no longer archived and forgotten:

• Researcher reports saved to {plan-dir}/research/
• step-03 reads reports and cites findings in phase Key Insights
• Mandatory research links: phase Context Links MUST reference research reports (verified by step-03)
• Bounded: 2 researchers max, 5 tool calls each, reports ≤150 lines

Sync-Back + Cross-Session Resilience

.plan-state.json checkpoint tracks phase completion:

{
  "phases": {
    "phase-01-setup": {
      "status": "complete",
      "tasks_completed": 5,
      "tasks_total": 5
    },
    "phase-02-impl": {
      "status": "in-progress",
      "tasks_completed": 2,
      "tasks_total": 7
    }
  }
}

New session reads checkpoint → skips completed phases → re-hydrates only pending work.

Critical-Step Tasks

Todo items marked [CRITICAL] or [HIGH] get their own Claude Tasks (not just phase-level), enabling finer-grained progress tracking for high-risk work.

Richer Frontmatter

plan.md YAML now includes: description (card preview), tags (filtering), issue (GitHub link), blockedBy/blocks (cross-plan deps).

Comparative Position

After v1.3.2, meow:plan-creator matches or exceeds ck-plan on all critical dimensions while staying 10% lighter (1,280 vs 1,423 total lines):

Dimension	v1.3.1 (before)	v1.3.2 (after)
Plan output	Single file	Multi-file with 12-section template
Scope gating	None	Trivial exit + user scope input
Plan red team	None	2 adversarial personas (hard mode)
Research	Archived, disconnected	Integrated, linked, verified
Session resilience	Plan files only	Checkpoint + sync-back
Token efficiency	High (but wasted on trivial)	Higher (scope gate + JIT steps)