plan-critics — Claude Code skill: what it does, how to install

What plan-critics does

This workflow reviews an implementation document or issue ticket from distinct viewpoints. The coordinator scans specification files inside critics/*.md. It always applies general checks for edge cases, architecture bloat, security weaknesses, and excessive complexity. When the text references specific frameworks like Next.js, Postgres, or Vercel, corresponding tech reviewers join the active pool.

Reviews run across bundled subagents in waves capped at four concurrent jobs. Inside each subagent, write access is blocked: Write, Edit, and Bash cannot be executed, leaving only inspection utilities such as Read, Grep, and Glob. Findings are parsed, deduplicated, and organized into an aggregated table that flags direct disagreements between different critics.

You must run this skill at the session root. Invoking it inside an existing subagent fails because it relies directly on the Task tool. When no explicit target path or issue number is passed, it falls back to the active plan session or the latest file saved in ~/.claude/plans/.