execplan

Create a living execution plan document that keeps the conversation thread short.

Usage

Invoke this skill when:

•Task has >3 steps
•Implementation will span multiple turns
•Context is getting long and needs offloading

code

/execplan cv_validation
/execplan pdf_generation_fix

The skill creates tmp/$ARGUMENTS_execplan.md with a structured plan template.

Plan structure

The generated plan file includes:

markdown

# Goal
One sentence describing what we're building/fixing.

# Acceptance Criteria
Observable behavior with exact verification commands:
- [ ] `npm test -- cv-visual.spec.ts` passes
- [ ] PDF size < 2 pages
- [ ] Visual diff < 5% threshold

# Plan
1. [ ] Smallest next step
2. [ ] Second step
3. [ ] Third step

# Commands
Exact commands with working directory and expected output.

# Validation
How to verify success (tests, manual checks).

# Recovery
Idempotence notes, rollback commands, retry steps.

# Notes
Append-only log of decisions and rationale with timestamps.

Working with plans

Update as you go:

•Mark steps [x] when completed
•Mark steps [!] if blocked
•Append decisions to Notes section with date

Link in responses: Use markdown links: tmp/cv_validation_execplan.md

Summarize deltas: Don't paste the full plan into chat - only summarize what changed.

Pre-agent gate for AI integration

If the plan involves OpenAI/Claude integration:

•Add acceptance criterion: "Backend dry-run passes (no agent connected)"
•Verify orchestration works before connecting the model
•This prevents debugging agent issues vs backend issues simultaneously

Safety

•No secrets in plan files (use env var names: OPENAI_API_KEY)
•Plans are working files - they don't replace git commits or documentation
•Safe to delete and regenerate if plan becomes stale

Examples

See .claude/workflows/execplan-examples.md for complete workflow examples.