flow-plan-review_skill

Overview

This skill coordinates Carmack-level plan reviews using rp-cli and RepoPrompt. It is designed to orchestrate context building and delegate the actual review to a RepoPrompt chat or export context for an external LLM. Use it when you want structured, repeatable review runs without the coordinator writing the review itself.

How this skill works

The skill checks for rp-cli and, if available, finds the active RepoPrompt window, builds focused context around a plan file or beads ID, and triggers a chat-based review session. It never reviews the plan itself; it issues rp-cli commands to select windows, assemble context, and start the review conversation in RepoPrompt or export the assembled context file for other LLMs.

When to use it

• Reviewing implementation plans, architecture docs, or design specs.
• When you want a reproducible, context-rich review run via RepoPrompt chat.
• If you need to export plan context for external LLMs like ChatGPT or Claude web.
• When coordinating multi-model review gates or night-run Ralph-style passes.
• When you require strict separation between coordinator actions and reviewer output.

Best practices

• Provide a clear input: plan file path (plans/<slug>.md) or beads ID plus focus areas.
• Do not expect the coordinator to critique — it only builds context and triggers reviewers.
• If rp-cli is missing, install it before invoking this skill; the skill will report that requirement.
• Re-use prior chosen review mode if already set in the conversation; avoid re-prompting the user.
• Follow the workflow phases in order: window selection, context build, then review execution.

Example use cases

• Invoke with: /flow:plan-review docs/plan/auth-refactor.md focus on security and error handling
• Automate overnight Ralph-mode reviews that run with fresh context and re-anchoring.
• Export context files to run the same review prompt in a web-based LLM for cross-model comparison.
• Coordinate multi-stage review gates where RepoPrompt chat runs the detailed checks and records receipts.
• Run a focused architecture review that collects codemaps and related files before launching the chat.