dev-optimizelisted

# Optimize Autonomous improvement loop: define metric → spawn parallel subagents in worktrees → keep improvements, discard regressions → loop until stall. **Trigger**: `/dev-optimize` or `/dev-optimize [auto]` Inspired by the `evo-hq/evo` autoresearch pattern, integrated with the `.project/` convention. No external dependency. ## Input No required input. All config is collected interactively in PHASE 1 or carried over from a previous run (resume). ## Output ``` .project/optimize/{run-id}/ ├── spec.json # metric, gate, scope, baseline, parameters ├── tree.json # nodes (id, parent, branch, score, status, hypotheses) ├── runs/{NNNN}.json # per-round result └── branches.txt # cleanup list for abort ``` Update on completion: `.project/project.json` → push to `optimization_runs[]` (see `shared/DASHBOARD.md`). ## Process **Phase tracking** — first action of the skill: call `TaskCreate` with these 7 items (status `pending`), then use `TaskUpdate` to set each phase `in_progress` at the start and `completed` at the end. During context compaction the task list remains visible — no risk of forgotten phases. 1. PHASE 0: Pre-flight 2. PHASE 1: Define Metric 3. PHASE 2: Instrument Benchmark 4. PHASE 3: Baseline Run 5. PHASE 4: Optimize Loop 6. PHASE 5: Pick Winner 7. PHASE 6: Sync + Report ### PHASE 0: Pre-flight > **Todo**: call `TaskCreate` with the 7 phase items (see above). Mark PHASE 0 → `in_progress` via `TaskUpdate`. **Capture git baseline:**