learning-looplisted

# Learning Loop Closed-loop improvement system: collect sensor data → detect regressions → create issues → verify fixes → learn from results. ## Workflow ### Step 1: Collect Sensor Data Run the unified sensor report to gather metrics from all available sensors: ```bash node scripts/sensor-report.mjs ``` Read the output. The script queries 7 sensors (ACMM, PR metrics, agent cost, CI health, Lighthouse, GitHub issues, session logs) and persists the report to `metrics/sensor-report.json`. It also detects regressions by comparing against the previous report. If the script exits with code 1, regressions were detected. Note them for Step 3. ### Step 1b: Sentry Triage If the Sentry MCP is available, run production error triage: Invoke `/sentry-triage` to query Sentry for new/regressed production errors and create GitHub issues for any that pass the severity/frequency/deduplication filters. This step is optional — if Sentry is not authenticated or unavailable, skip with a note in the summary. ### Step 2: Verify Past Fixes Run fix verification on recently-closed issues: ```bash node scripts/verify-fixes.mjs ``` This finds issues closed in the last 48 hours with sensor labels (`ci-fix`, `audit`, `acmm`, `sentry`, `bug`), queries the originating sensor, and: - Comments on the issue with verification evidence - Reopens issues where the fix didn't improve the metric Note any reopened issues for the summary. ### Step 3: Triage Regressions Read the sensor report from `met