From brainstorming-tools
Use immediately after the superpowers writing-plans skill produces an implementation plan (the brainstorm → spec → plan flow), before the execution handoff. An independent fresh-eyes second pass that reviews the plan against its spec and applies fixes. Do NOT use for arbitrary or external plans, generic "make this better" requests, or once execution has started.
How this skill is triggered — by the user, by Claude, or both
Slash command
/brainstorming-tools:optimizing-writing-plansThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
`writing-plans` ends with its own **Self-Review** — but that is the *author* checking their own work, blind to its own assumptions. This skill is the **independent second pass**: fresh eyes that re-review the finished plan against the spec and **apply fixes in place**, then hand off to execution.
writing-plans ends with its own Self-Review — but that is the author checking their own work, blind to its own assumptions. This skill is the independent second pass: fresh eyes that re-review the finished plan against the spec and apply fixes in place, then hand off to execution.
REQUIRED SUB-SKILL: Use andrej-karpathy-skills:karpathy-guidelines as the reviewing lens — it names the failure modes this pass hunts for (over-engineering, unsurfaced assumptions, non-surgical scope, weak success criteria).
Core discipline — fix vs flag:
Stay a reviewer, not a re-author. Review the plan against the spec. You MAY check the codebase to verify a path/type/convention, but surface corrections as scoped fixes with a one-line reason — do not rewrite the whole plan or invent new scope.
writing-plans writes the plan, before it offers the execution choice. Runs once per plan.Run all four. For each finding, decide fix or flag per the discipline above.
run: commands or expected FAIL/PASS. Fill in the concrete content the author should have written. (These are exactly the failures writing-plans forbids.)**Parallel:** independent of Tasks N–M); do NOT build a dependency graph or named tracks.Every file path, code block, exact run command, expected FAIL/PASS, error code, and status code that is already correct survives verbatim. Length comes down by merging and de-boilerplating, not by summarizing away executable detail.
writing-plans execution handoff (subagent-driven vs inline).Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.
npx claudepluginhub bartekbp/claude-skills --plugin brainstorming-tools