Skip to content

Retro skill should offer to dispatch follow-up tasks

Observed: The retro skill ends with a handoff note (task note in the local project), but there's no prompt or mechanism to dispatch cross-project follow-up work. For instance, after the pixecog preprocess_motion session, the retro identified a skill creation task that belongs in projio — but the agent had to manually figure out how to get that work dispatched.

Expected: The retro skill should: 1. After filing issues and creating the handoff task, identify which follow-ups target other projects (e.g., projio skill creation, cogpy bug fix) 2. Offer to create a task and dispatch it via schedule_queue to the target project 3. Include the full prompt with context from the retro findings

This closes the loop: retro identifies the work, files the issue, AND queues the execution — rather than leaving a dangling idea note that requires a human to pick up.


Source context: pixecog

PixEcog (pixecog): Neuropixels and ECoG dataset and analysis

Recent commits:

27a75e5 Preprocess_ieeg DAG redesign: ttl_removal before interpolate, two-tier badlabel, auto-skip TTL, pre-lowpass LNR
42cf0d9 Preprocess_ieeg redesign: DAG reorder, two-tier badlabel, auto-skip TTL, pre-lowpass LNR, cross-subject TTL analysis, notebooks published
6b295b2 Update badlabel audit note: full pipeline comparison (85 new vs 7 legacy vs 71 TTL), zarr fix confirmed, 97.2% TTL catch rate

README:


type: readme


Quick Start for Collaborators

Follow this checklist to get started with Pixecog documentation and workflows.

🐀 Pixecog Project — Compact Overview

Core principles

  • One immutable BIDS raw dataset (raw/) as the canonical baseline
  • Each analysis pipeline ha
  • issue-arash-20260409-034440-983440.md — Both concern reducing agent friction in multi-step workflows — note_create round-trips and retro dispatch are both about closing automation gaps in the agent loop
  • issue-arash-20260410-145955-796957.md — Same theme: tools should auto-prompt or auto-trigger follow-on actions (registry_scan after flow_new; dispatch after retro) rather than leaving manual steps for the agent