v1.4.0 — The Plan Intelligence Release

Dedicated red-team step extracted from monolithic validation, CK-style finding adjudication, plan-specific adversarial personas, and two new workflow modes (--parallel, --two).

What Changed

Red Team Extraction (Steps 04–07)

The monolithic step-04-validate-and-gate.md is replaced by four focused steps:

step-04-semantic-checks.md      ← semantic content + structural validation
step-05-red-team.md             ← dedicated adversarial review (NEW)
step-06-validation-interview.md ← 3-5 critical questions
step-07-gate.md                 ← Gate 1 presentation

Old step-05-hydrate-tasks.md → step-08-hydrate-tasks.md. Workflow is now 9 steps (00–08), up from 6 (00–05).

CK-Style Adjudication in step-05

Findings follow a 7-field format:

Field	Purpose
Severity	CRITICAL / HIGH / MEDIUM / LOW
Location	[PHASE:SECTION] reference
Flaw	One-line problem statement
Failure Scenario	What breaks if this is ignored
Evidence	Quote from plan text
Suggested Fix	Concrete remediation
Category	assumption / scope / architecture / risk / dependency

Agent adjudicates each finding: Accept or Reject with rationale. After adjudication, user sees a 3-option review gate:

1. Apply all accepted — batch apply without review
2. Review each — per-finding granular control (Apply / Skip / Modify)
3. Reject all — proceed without changes

Deduplication by location + core_flaw, severity-sorted, capped at 15 findings. Dispositions are written to a Red Team Review section in plan.md — auditable, persistent.

Plan-Specific Personas

Two new personas in prompts/personas/:

Persona	Focus
plan-assumption-destroyer.md	Unvalidated assumptions, fragile dependencies
plan-scope-complexity-critic.md	YAGNI violations, scope creep, over-engineering

These reference [PHASE:SECTION] (not [FILE:LINE] like code-review personas). Dynamic scaling by phase count:

Phase count	Personas active
1–3 phases	2
4–5 phases	3
6+ phases	4

Security Adversary + Failure Mode Analyst are gated on future A/B testing results.

Fast-Mode Workflow

New workflow-fast.md for the compact path:

step-00 → step-03 → step-04 → step-07 → step-08

Skips: research (step-01), codebase analysis (step-02), red-team (step-05), validation interview (step-06).

--parallel Mode

Generates an Execution Strategy section in plan.md with:

• File ownership matrix (which agent owns which files/directories)
• Parallel group assignments (which phases can run concurrently)
• Per-group task hydration with file-scope boundaries

Requires --hard internally. .plan-state.json schema bumped to v1.1 with optional parallel_groups field.

--two Mode

Generates 2 competing approach files (approach-A.md, approach-B.md) plus a trade-off matrix. User selects one before red-team runs — red-team reviews only the selected approach.

Requires --hard internally. .plan-state.json schema v1.1 adds optional selected_approach field.

Step Workflow Diagram

v1.3.2 (6 steps)                  v1.4.0 (9 steps)
─────────────────                  ────────────────────────────────────
step-00: scope challenge           step-00: scope challenge
step-01: bounded research          step-01: bounded research
step-02: codebase analysis         step-02: codebase analysis
step-03: draft plan + phases       step-03: draft plan + phases
step-04: validate + red team       step-04: semantic checks          ← split
         + interview + Gate 1      step-05: red-team (CK adjudication) ← NEW
step-05: hydrate tasks             step-06: validation interview     ← split
                                   step-07: Gate 1 presentation      ← split
                                   step-08: hydrate tasks

Fast mode path: 00 → 03 → 04 → 07 → 08

Comparison: v1.3.2 → v1.4.0

Dimension	v1.3.2	v1.4.0
Workflow steps	6 (00–05)	9 (00–08)
Red-team step	Embedded in step-04	Dedicated step-05
Finding format	Informal	7-field structured
Finding adjudication	None	Accept/Reject + rationale
User review gate	Accept/reject each	3-option batch + per-finding
Max findings	10	15 (deduplicated, severity-sorted)
Personas	Reused from meow:review	Plan-specific (PHASE:SECTION)
Persona scaling	Fixed 2	Dynamic by phase count
Auditable record	None	Red Team Review table in plan.md
Fast mode	Via scope gate	Dedicated workflow-fast.md
Parallel planning	None	--parallel mode
Competing approaches	None	--two mode
.plan-state.json	v1.0	v1.1 (parallel_groups, selected_approach)

Files Changed

File	Change
.claude/skills/meow:plan-creator/SKILL.md	Bumped to v1.4.0, updated step list
.claude/skills/meow:plan-creator/workflow.md	9-step sequence
.claude/skills/meow:plan-creator/workflow-fast.md	NEW — compact fast path
.claude/skills/meow:plan-creator/step-04-semantic-checks.md	Renamed from old step-04 (validation only)
.claude/skills/meow:plan-creator/step-05-red-team.md	NEW — dedicated red-team step
.claude/skills/meow:plan-creator/step-06-validation-interview.md	Split from old step-04
.claude/skills/meow:plan-creator/step-07-gate.md	Split from old step-04
.claude/skills/meow:plan-creator/step-08-hydrate-tasks.md	Renamed from old step-05
.claude/skills/meow:plan-creator/prompts/personas/plan-assumption-destroyer.md	NEW
.claude/skills/meow:plan-creator/prompts/personas/plan-scope-complexity-critic.md	NEW
.claude/skills/meow:plan-creator/docs/parallel-mode.md	NEW
.claude/skills/meow:plan-creator/docs/two-approach-mode.md	NEW
.claude/skills/meow:plan-creator/docs/gotchas.md	+6 gotchas
.claude/rules/step-file-rules.md	meow:plan-creator added to step-file enabled list