| name | optimize |
| description | 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. |
| disable-model-invocation | true |
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.
