Optimize
When this skill is invoked, the agent should do the following:
What to look for
Scan the full codebase (src/) for the following categories of issues, ordered from easiest wins to bigger refactors:
- File size / split candidates — files over ~300 lines that contain multiple distinct responsibilities; components that mix data-fetching with presentation; large utility files where functions could be grouped by domain.
- Component complexity — components with too many props, deeply nested JSX, or logic that belongs in a custom hook.
- Duplicate / near-duplicate logic — repeated patterns across files that could be extracted into a shared hook, util, or DAO function.
- Unnecessary re-renders — missing
memo, missing useCallback/useMemo, derived state computed in render body instead of memoized, objects/arrays created inline as props.
- Performance bottlenecks — N+1 queries in DAOs, missing
Promise.all for independent async calls in server components, expensive operations not cached.
- Dead code — unused exports, unreachable branches, commented-out code blocks.
- Trivial readability — deeply nested ternaries that could be a variable or early return; magic numbers/strings without a named constant.
Workflow
Scan first. Read the relevant files — do not guess. Prioritize files flagged as large or complex by the file listing.
Rank by effort vs. impact. Pick the single highest-impact, lowest-effort issue (the "lowest hanging fruit").
Present the finding in this format:
📁 File: <path>
🔍 Issue: <one-sentence description>
💡 Fix: <one-sentence plan>
Write 'go' to apply this fix, or 'skip' to find the next one.
Wait for the user to write go. Do not make any edits until the user confirms.
Apply the fix. Make the smallest, most focused change that addresses the issue.
After the fix, immediately find the next lowest-hanging-fruit issue and present it in the same format. Return to step 4.
If the user writes skip, find the next issue without making changes.
Keep looping until the user says stop, done, or exit, or there are no more issues to report.
Constraints
- One change at a time. Never batch multiple fixes into a single edit.
- No speculative changes. Only fix things you can see are actually a problem after reading the code.
- Preserve behavior. Refactors must not change observable behavior. If unsure, note the risk in the finding.
- Respect AGENTS.md rules. All existing architecture rules (data flow layers, critical file patterns, etc.) apply.
1---2name: optimize3description: Scans the repository for optimization opportunities — complexity, performance issues, large files, and maintainability problems — and fixes them one at a time with user confirmation. Use only when the user explicitly invokes this skill.4---56# Optimize78When this skill is invoked, the agent should do the following:910## What to look for1112Scan the full codebase (`src/`) for the following categories of issues, ordered from easiest wins to bigger refactors:1314- **File size / split candidates** — files over ~300 lines that contain multiple distinct responsibilities; components that mix data-fetching with presentation; large utility files where functions could be grouped by domain.15- **Component complexity** — components with too many props, deeply nested JSX, or logic that belongs in a custom hook.16- **Duplicate / near-duplicate logic** — repeated patterns across files that could be extracted into a shared hook, util, or DAO function.17- **Unnecessary re-renders** — missing `memo`, missing `useCallback`/`useMemo`, derived state computed in render body instead of memoized, objects/arrays created inline as props.18- **Performance bottlenecks** — N+1 queries in DAOs, missing `Promise.all` for independent async calls in server components, expensive operations not cached.19- **Dead code** — unused exports, unreachable branches, commented-out code blocks.20- **Trivial readability** — deeply nested ternaries that could be a variable or early return; magic numbers/strings without a named constant.2122## Workflow23241. **Scan first.** Read the relevant files — do not guess. Prioritize files flagged as large or complex by the file listing.25262. **Rank by effort vs. impact.** Pick the single highest-impact, lowest-effort issue (the "lowest hanging fruit").27283. **Present the finding** in this format:2930```31📁 File: <path>32🔍 Issue: <one-sentence description>33💡 Fix: <one-sentence plan>3435Write 'go' to apply this fix, or 'skip' to find the next one.36```37384. **Wait for the user to write `go`.** Do not make any edits until the user confirms.39405. **Apply the fix.** Make the smallest, most focused change that addresses the issue.41426. **After the fix**, immediately find the next lowest-hanging-fruit issue and present it in the same format. Return to step 4.43447. **If the user writes `skip`**, find the next issue without making changes.45468. **Keep looping** until the user says `stop`, `done`, or `exit`, or there are no more issues to report.4748## Constraints4950- **One change at a time.** Never batch multiple fixes into a single edit.51- **No speculative changes.** Only fix things you can see are actually a problem after reading the code.52- **Preserve behavior.** Refactors must not change observable behavior. If unsure, note the risk in the finding.53- **Respect AGENTS.md rules.** All existing architecture rules (data flow layers, critical file patterns, etc.) apply.
kovrichard