/plan

Interview-based validation to surface assumptions, risks, and edge cases before implementation.

Syntax

/plan:validate                # Validate active plan
/plan:validate [plan-path]    # Validate specific plan

What It Does

Systematically reviews plan for:

Architecture decisions - Design patterns, database choices, API structures
Assumptions - Implicit expectations that may be wrong
Tradeoffs - Comparing alternatives and their implications
Risks - Dependencies, blockers, potential failures
Scope - MVP boundaries, future phases, out-of-scope items

Question Categories

Category	Detects
Architecture	”approach”, “pattern”, “design”, “structure”, “database”, “API”
Assumptions	”assume”, “expect”, “should”, “will”, “must”, “default”
Tradeoffs	”tradeoff”, “vs”, “alternative”, “option”, “choice”
Risks	”risk”, “might”, “could fail”, “dependency”, “blocker”
Scope	”phase”, “MVP”, “future”, “out of scope”, “nice to have”

Interview Process

Reads Plan - Analyzes plan.md and all phase-*.md files
Generates Questions - 3-8 critical questions with 2-4 concrete options each
User Interview - Presents questions via interactive prompts
Documents Answers - Adds ## Validation Summary to plan.md
Action Items - Flags any plan revisions needed

Configuration

Set in .claude/.ck.json:

{
  "validation": {
    "mode": "auto",           // auto|prompt|off
    "questions": "3-8"        // min-max range
  }
}

Example Output

## Validation Summary

**Validated:** 2026-01-28
**Questions asked:** 5

### Confirmed Decisions
- Validation persistence: Save to plan.md frontmatter
- Rate limiting: Add basic implementation now
- Database: PostgreSQL with TimescaleDB extension

### Action Items
- [ ] Update phase-02 to include rate limiting middleware
- [ ] Add PostgreSQL setup to phase-01

Best Practice

After validation, run /clear then /cook to start implementation with fresh context.

This prevents planning context pollution and improves plan adherence.

/plan
- Research-based planning
/cook - Implementation from validated plan