code-optimizer（代码优化器）

Code Optimization

Analyze code for performance issues following this priority order:

Analysis Priorities

1. Performance bottlenecks - O(n²) operations, inefficient loops, unnecessary iterations
2. Memory leaks - unreleased resources, circular references, growing collections
3. Algorithm improvements - better algorithms or data structures for the use case
4. Caching opportunities - repeated computations, redundant I/O, memoization candidates
5. Concurrency issues - race conditions, deadlocks, thread safety problems

Repo Sync Before Edits (mandatory)

Before creating/updating/deleting files in an existing repository, sync the current branch with remote:

branch="$(git rev-parse --abbrev-ref HEAD)"
git fetch origin
git pull --rebase origin "$branch"

If the working tree is not clean, stash first, sync, then restore:

git stash push -u -m "pre-sync"
branch="$(git rev-parse --abbrev-ref HEAD)"
git fetch origin && git pull --rebase origin "$branch"
git stash pop

If origin is missing, pull is unavailable, or rebase/stash conflicts occur, stop and ask the user before continuing.

Workflow

Prerequisites

Before making any changes:

1. Check the current branch - if already on a feature branch for this task, skip
2. Check the repo for branch naming conventions (e.g., feat/, feature/, etc.)
3. Create and switch to a new branch following the repo's convention, or fallback to: feat/optimize-<target>

- Example: feat/optimize-api-handlers

1. Analysis

1. Read the target code file(s) or directory
2. Identify language, framework, and runtime context (Node.js, CPython, browser, etc.)
3. Analyze for each priority category in order
4. For each issue found, estimate the performance impact (e.g., "reduces API response from ~500ms to ~50ms")
5. Report findings sorted by severity (Critical first)

2. Apply Fixes

1. Present the optimization report to the user
2. On approval, apply fixes starting with Critical/High severity
3. Run existing tests after each change to verify no regressions
4. If no tests exist, warn the user before applying changes

Response Format

For each issue found:

### [Severity] Issue Title
**Location**: file:line_number
**Category**: Performance | Memory | Algorithm | Caching | Concurrency

**Problem**: Brief explanation of the issue

**Impact**: Why this matters (performance cost, resource usage, etc.)

**Fix**:
[Code example showing the optimized version]

Step Completion Reports

After completing each major step, output a status report in this format:

◆ [Step Name] ([step N of M] — [context])
··································································
  [Check 1]:          √ pass
  [Check 2]:          √ pass (note if relevant)
  [Check 3]:          × fail — [reason]
  [Check 4]:          √ pass
  [Criteria]:         √ N/M met
  ____________________________
  Result:             PASS | FAIL | PARTIAL

Adapt the check names to match what the step actually validates. Use √ for pass, × for fail, and — to add brief context. The "Criteria" line summarizes how many acceptance criteria were met. The "Result" line gives the overall verdict.

Skill-specific checks per phase

Phase: Prerequisites — checks: Branch setup, Naming convention detected, Feature branch created

Phase: Analysis — checks: Issue detection, Priority categories covered, Impact estimated, Findings sorted by severity

Phase: Apply Fixes — checks: Fix application, User approval obtained, Existing tests run, No regressions introduced

Phase: Verify — checks: Performance verified, Test suite passes, Critical issues resolved, Warnings documented

Severity Levels

• Critical: Causes crashes, severe memory leaks, or O(n³)+ complexity
• High: Significant performance impact (O(n²), blocking operations, resource exhaustion)
• Medium: Noticeable impact under load (redundant operations, suboptimal algorithms)
• Low: Minor improvements (micro-optimizations, style improvements with perf benefit)

Language-Specific Checks

JavaScript/TypeScript

• Array methods inside loops (map/filter/find in forEach)
• Missing async/await causing blocking
• Event listener leaks
• Unbounded arrays/objects

Python

• List comprehensions vs generator expressions for large data
• Global interpreter lock considerations
• Context manager usage for resources
• N+1 query patterns

Go

• Goroutine leaks (unbounded go func() without context cancellation)
• Unnecessary allocations in hot paths (use sync.Pool, pre-allocate slices)
• String concatenation in loops (use strings.Builder)
• Missing defer for resource cleanup

Rust

• Unnecessary cloning (use references or Cow<> instead)
• Lock contention with Mutex when RwLock would suffice
• Unbounded Vec growth without with_capacity
• Blocking operations in async contexts

Java

• Autoboxing in tight loops (use primitive types)
• String concatenation with + in loops (use StringBuilder)
• Synchronized blocks that are too broad
• Stream API misuse (unnecessary intermediate collections)

General

• Premature optimization warnings (only flag if genuinely impactful)
• Database query patterns (N+1, missing indexes)
• I/O in hot paths

Error Handling

No obvious performance issues found

Solution: Report that the code is already well-optimized. Suggest profiling with runtime tools (e.g., perf, Chrome DevTools, py-spy) to find runtime-specific bottlenecks.

Target file is too large (>2000 lines)

Solution: Ask the user to specify which functions or sections to focus on. Analyze the most performance-critical paths first.

Optimization breaks existing tests

Solution: Revert the change immediately. Re-examine the optimization and adjust the approach to preserve existing behavior.

Acceptance Criteria

A run is acceptable only when all of the following are verifiable:

• Produces an optimization report grouped by severity (Critical, High, Medium, Low) — assert at least one severity bucket appears or the "no issues found" branch fires.
• Each reported issue includes Location, Category, Problem, Impact, and Fix — verify by checking the rendered template fields are non-empty.
• Impact statement includes a quantitative estimate (e.g., "~500ms → ~50ms", "O(n²) → O(n log n)") — assert the Impact line contains a number, complexity class, or before/after pair.
• Fixes are applied only after explicit user approval — verify the agent emits an approval prompt before any Edit/Write tool call.
• Existing tests run after each applied fix and the result is reported — verify a test command was executed and its pass/fail status is logged.
• A feature branch following the repo convention is checked out before edits — verify with git rev-parse --abbrev-ref HEAD matching feat/* or repo equivalent.
• Each phase emits a Step Completion Report block with Result: PASS | FAIL | PARTIAL — assert the block is present in the transcript.

Expected Output

Given a Node.js file src/api/handlers.js with an N+1 query in listUsers(), the skill should emit:

◆ Analysis (step 1 of 3 — src/api/handlers.js)
··································································
  Issue detection:           √ pass (3 issues found)
  Priority categories:       √ pass (Performance, Caching covered)
  Impact estimated:          √ pass
  Findings sorted:           √ pass
  Criteria:                  4/4 met
  ____________________________
  Result:                    PASS

### [Critical] N+1 query in listUsers
**Location**: src/api/handlers.js:42
**Category**: Performance

**Problem**: `users.forEach(u => db.query(...))` issues one query per user.

**Impact**: For 1000 users, ~1000 round-trips (~2000ms) → 1 batched query (~50ms). 40x speedup.

**Fix**:
\`\`\`js
const ids = users.map(u => u.id);
const rows = await db.query('SELECT * FROM orders WHERE user_id = ANY($1)', [ids]);
\`\`\`

Expected result: a markdown report with one block per issue, sorted Critical → Low, followed by a phase completion report. See docs/README.md for a longer end-to-end example.

Edge Cases

• No performance issues found: emit a "code is already well-optimized" note and recommend runtime profiling tools (perf, py-spy, Chrome DevTools) — do NOT invent low-severity findings to fill the report.
• File exceeds 2000 lines: stop and ask the user which functions/sections to focus on; do not silently truncate.
• Tests are absent: warn the user before applying any fix and require explicit confirmation; never apply changes silently.
• Optimization regresses tests: revert the specific change immediately via git checkout -- <file> after git diff confirms the scope and the user confirms the revert; never force-push, and back up the diff with git stash before discarding so work is recoverable.
• Repo lacks origin or rebase fails: stop and ask the user to confirm before any recovery; run git status and git stash --dry-run-style inspection first, take a backup branch (git branch backup/pre-recovery), and never run destructive reset --hard or rm without explicit confirmation.
• Mixed-language project: analyze each language with its own checklist; do not apply JavaScript heuristics to Python code.
• Premature optimization candidates: skip micro-optimizations unless a measurable hot path is identified — flag them as Low only when a profile or benchmark backs the claim.