optimize-implement-queuelisted

# Optimize Implement Queue Daily skill that measures implement-queue efficiency, logs trends, and files `ready` issues when a real regression is detected. Phase-2 auto-tuning (model-routing tier adjustment) is documented as a future seam but NOT yet built. ## Flags | Flag | Effect | | ----------- | -------------------------------------------------------- | | `--dry-run` | Print what would happen; write nothing to disk or GitHub | ## Step 0: Reconcile Telemetry Outcomes Fill the outcome fields (`merged`, `ci_first_pass`, `rework_cycles`, `merged_at`) that workers cannot know at write time — the sensor's precise-cost path only activates when rows are reconciled: ```bash node scripts/reconcile-queue-telemetry.mjs ``` Idempotent and capped at 50 GitHub lookups; safe to run every day. ## Step 1: Collect the Queue-Efficiency Sensor Run the sensor-report to collect the `queueEfficiency` sub-report: ```bash node scripts/sensor-report.mjs --json | jq '.sensors.queueEfficiency' ``` Or capture the whole report for later regression detection: ```bash node scripts/sensor-report.mjs --json > /tmp/sensor-report.json ``` Read `regressions[]` from the output. Any entries with `sensor: "queueEfficiency"` are internal regressions detected by the rolling-7-day-median baseline baked into the collector (`collect-queue-efficiency.mjs`). The `distribution` field provides difficulty-normalized context (size tiers): a session domin