ln-810-performance-optimization-coordinator（ln 810 性能优化协调器）

Paths: File paths (shared/, references/, ../ln-*) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.

ln-810-performance-optimization-coordinator

Type: L2 Domain Coordinator Category: 8XX Optimization

Coordinates performance optimization by delegating to L3 workers: ln-811 (algorithm), ln-812 (query), ln-813 (runtime). Workers run in parallel when inputs are independent.

Overview

Workflow

Phases: Pre-flight → Analyze Input → Delegate → Collect → Verify → Report

Phase 0: Pre-flight Checks

MANDATORY READ: Load shared/references/ci_tool_detection.md for test/build detection.

Phase 1: Analyze Input

Input Sources

Worker Selection

*ln-811 activated only if specific algorithmic inefficiency identified in audit.

Phase 2: Delegate to Workers

CRITICAL: All delegations use Agent tool with subagent_type: "general-purpose" and isolation: "worktree" — each worker creates its own branch per shared/references/git_worktree_fallback.md.

Delegation Protocol

FOR each selected worker:
  Agent(description: "Optimize via ln-81X",
       prompt: "Execute optimization worker.

Step 1: Invoke worker:
  Skill(skill: \"ln-81X-{worker}\")

CONTEXT:
{delegationContext}",
       subagent_type: "general-purpose",
       isolation: "worktree")

Delegation Context

Parallelism

Rules:

• If workers target the SAME file, run sequentially (ln-811 first, then ln-812/813).
• Dependent workers share branch: If worker B depends on worker A's output (e.g., ln-813 needs ln-812's query changes), launch worker B in worker A's branch — so B sees A's changes.

Phase 3: Collect Results

Each worker produces an isolated branch. Coordinator aggregates branch reports.

Worker Branches

Result Schema

Phase 4: Aggregate Reports

Each worker verified independently in its branch (tests, build, lint run by worker itself). Coordinator does NOT rerun verification or revert worker changes.

On Failure

1. Branch with failing tests logged as "failed" in report
2. User reviews failed branch independently

Phase 5: Report Summary

Report Schema

Configuration

Options:
  # Input
  audit_report: "docs/project/persistence_audit.md"
  target_file: ""
  target_function: ""

  # Workers
  enable_algorithm: true
  enable_query: true
  enable_runtime: true

  # Verification
  run_tests: true
  run_build: true
  run_lint: true

  # Safety
  revert_on_build_failure: true

Error Handling

Recoverable Errors

Fatal Errors

References

• ../ln-811-algorithm-optimizer/SKILL.md
• ../ln-812-query-optimizer/SKILL.md
• ../ln-813-runtime-optimizer/SKILL.md
• shared/references/ci_tool_detection.md

Definition of Done

• Input analyzed (audit report or target file/function)
• Appropriate workers selected based on input type
• Workers delegated with worktree isolation (isolation: "worktree")
• Each worker produces isolated branch, pushed to remote
• Coordinator report aggregates per-worker results (branch, fixes, status)

Phase 6: Meta-Analysis

MANDATORY READ: Load shared/references/meta_analysis_protocol.md

Skill type: optimization-coordinator. Run after all phases complete. Output to chat using the optimization-coordinator format.

Version: 1.0.0 Last Updated: 2026-03-08