Optimization Workflow Orchestrator

Orchestrates the complete performance optimization workflow. Single entry point that manages state, transitions, and termination.

Quick Start

Step 1: Determine current state (see State Detection below)

Step 2: Execute the appropriate phase skill

Step 3: After phase completes, re-evaluate state and continue or terminate

State Detection

Check files in this order to determine current phase:

1. BENCHMARK_BASELINE.md exists?
   NO  → Execute PHASE 1 (establishing-benchmark-baseline)
   YES → Continue to check 2

2. PERFORMANCE_THEORIES.md exists?
   NO  → Execute PHASE 2 (broad-performance-investigation)
   YES → Continue to check 3

3. Any theory with status "verified"?
   YES → Execute PHASE 4 (implementing-performance-fixes) for highest severity verified theory
   NO  → Continue to check 4

4. Any theory with status "pending"?
   YES → Select highest severity pending theory:
         - If "Deeper Investigation: SKIP" → Execute PHASE 4 (implementing-performance-fixes)
         - If "Deeper Investigation: REQUIRED" → Execute PHASE 3 (deep-performance-investigation)
   NO  → Continue to check 5

5. All theories have status "fixed" or "fix-failed" or "falsified"?
   YES → Evaluate termination (see below)
   NO  → Error state, ask user

Priority Order for Theory Selection:

1. •Prioritize by Severity: Critical > High > Medium > Low
2. •If multiple theories share the same severity, select the first one
3. •Respect "Deeper Investigation" decision (SKIP→Phase 4, REQUIRED→Phase 3)
4. •This ensures critical performance issues are addressed before easy fixes

Phase Skills

Termination Criteria

After all theories are processed (fixed/fix-failed/falsified), evaluate:

Continue optimization (loop to PHASE 2) if:

• •User requests more optimization

Stop optimization if:

• •One optimization iteration completed
• •User didn't request more optimization
• •Performance meets or exceeds baseline expectations
• •User indicates satisfaction
• •No more theories can be generated (diminishing returns)
• •All recent fixes failed (indicates deeper issues)

Edge Cases

Workflow Visualization

flowchart TD
    Start([Start]) --> P1[PHASE 1<br/>Baseline]
    P1 --> P2[PHASE 2<br/>Theories]

    P2 -->|Deeper Investigation:<br/>SKIP| P4[PHASE 4<br/>Fix]
    P2 -->|Deeper Investigation:<br/>REQUIRED| P3[PHASE 3<br/>Verify]

    P3 --> P4

    P4 -->|Gaps remain| P3
    P4 -->|Done| End([Complete])

Key paths:

• •Phase 2 → Phase 4 directly if "Deeper Investigation: SKIP" (definitive evidence + low-risk fix)
• •Phase 2 → Phase 3 → Phase 4 if "Deeper Investigation: REQUIRED" (circumstantial evidence OR high-risk fix)

Orchestrator Checklist

Copy this to track progress:

Optimization Progress:
- [ ] State determined
- [ ] PHASE 1: Baseline established (BENCHMARK_BASELINE.md exists)
- [ ] PHASE 2: Theories generated (PERFORMANCE_THEORIES.md exists)
- [ ] PHASE 3: Theory verified (at least one verified)
- [ ] PHASE 4: Fix implemented and validated
- [ ] Termination evaluated
- [ ] Loop or complete

Integration

This skill CALLS the phase skills:

• •establishing-benchmark-baseline (PHASE 1)
• •broad-performance-investigation (PHASE 2)
• •deep-performance-investigation (PHASE 3)
• •implementing-performance-fixes (PHASE 4)

This skill does NOT call tool skills directly. Tool skills are called by PHASE 3.