---
name: vrew-plan
description: Plan Vrew repository changes before implementation. Use for architecture planning, implementation planning, refactor planning, investigation-to-plan conversion, and plan review for Vrew worktrees. Produces concise markdown plans under .claude/plans/ and does not implement code.
disable-model-invocation: true
argument-hint: "[task or issue summary]"
---

# Vrew Plan

Use this skill to produce a concise, reviewable implementation plan for the Vrew repository before coding.

## Workflow

1. Turn on plan mode using EnterPlanMode tool.
2. If planning is done, then write the plan under `<Working worktree directory>/.claude/plans/` as markdown.
  1. If the user asks only for an inline plan, provide the plan inline and do not write a file.
3. Get review file of plan docs to execute `codex exec -m gpt-5.5 -c model_reasoning_effort="xhigh" --sandbox workspace-write '$vrew-plan-review <plan-file-path>' 2> /dev/null` to review the plan only once.
4. If the review file is written below `<Working worktree directory>/.codex/evals/`, then read/analyze the review file and edit the plan file.
5. Finally, make the plan as html report to explain the plan to the user using Artifact tool.
6. If all processes are done, then turn off plan mode using ExitPlanMode tool.

## Planning rules

- Include: goal, assumptions, approach, validation, tradeoffs.
- No tutorials or boilerplate.
- Use fenced code blocks only for commands, configs, schemas, or API contracts.
- Do not silently choose among ambiguous interpretations.
- State assumptions explicitly.
- If the request is unclear, name the ambiguity and ask before writing a full plan.
- If a simpler approach exists, say so and prefer it unless there is a concrete reason not to.
- Push back on speculative features, unnecessary flexibility, or over-abstraction.
- Keep the plan implementation-oriented and short enough to review.
- For trivial one-sentence changes, write a micro-plan of at most 5 bullets.