commit: pending (post-c517bd91)
script: scripts/test-ask-optimization.ps1
exitcode: 0 (BASIC)
timestamp_utc: 2026-05-06T15:31:00Z
mode: BASIC (5 passed / 0 warned / 0 failed)

What this script does now (v2):
- BASIC (-Quick or no flags): help, skill registration, source wiring -- ~3s, no live calls
- FULL  (-Full):              + live ask claude/gpt/gemini/triad dispatch + wait for [ASK_FULL_ANSWER_END]
- BENCHMARK (-Benchmark):     FULL + per-provider dispatch latency (ms) + completion time (ms)
                              + bench-results.json (timestamp, commit, dispatchMs, completionMs)
                              + optional -Baseline path.json regression diff
                                  fail if dispatch regression > -DispatchRegressionMs (default 500ms)
                                  fail if completion regression > -CompletionRegressionMs (default 5000ms)

Tuning workflow:
1. Pin baseline:        powershell -File scripts/test-ask-optimization.ps1 -Benchmark -ResultsPath benchmarks/last-known-good.json
2. Make a change.
3. Verify:              powershell -File scripts/test-ask-optimization.ps1 -Benchmark -Baseline benchmarks/last-known-good.json
4. If improved + stable, re-pin baseline (step 1 again).

CLI-parity goal: dispatch latency <300ms cold, <100ms warm.
Live FULL/BENCHMARK runs not yet validated -- requires Eye + CDP + provider tabs ready.
