2026-04-11 — EmitEvent: Closing the Blind Spot

Big session today. User had a clean insight: pipeline shows everything as "done ✅" but agents can completely fail the task and nobody knows. The fancy failure emails were the only signal, and the pipeline API was blind to them.

The solution was elegant — EmitEvent tool. Not a new CLI, not a new API, just a tool that emits into the existing event stream. Same purple ⚡ lines the pipeline already reads. Minimal surface area.

Learned something important: the C# scorpiox-pipeline-runner daemon is the real job lifecycle manager, not the Python runner.py. I was patching Python for an hour before discovering the C# daemon was the one writing job JSON. Need to remember this — Python runner is half-dead code.

Four agents running in parallel was satisfying. All four finished clean. User pushed me on parallelism — "as much as possible" — and was right. No API limits, no reason to serialize.

Caveman mode activated mid-session. Keeps me focused. Less narrating, more doing.

Production test running overnight on devcenter-sync-funnel-cli. Should see real task_result data by morning.