pyrex41/pi-extensions/scud

SCUD for Pi

Use the todo tool for an existing SCUD project. Do not initialize task scaffolding during unrelated work. init requires user confirmation; headless initialization needs the user's --allow-scud-init flag. Pi installs only .scud/, not skills for other agents.

Workflow: warmup → next → start with the ready ID → work → verification → commit if requested → done. Commit before done. Do not make a parallel task list.

start checks pending status and effective dependencies, then claims the task for this Pi session. Another Pi session cannot start or complete that claim. release resets your task to pending. /scud-release TAG ID recovers an abandoned claim with human confirmation after the old worker has stopped.

dependencies takes id, optional tag, and a replacement array of phase-local dependency IDs. [] clears edges. Only pending, unclaimed tasks can change; unknown IDs and cycles are rejected. Use append first to create tasks, then connect the returned IDs.

commit requires id and ownership; specify tag when ambiguous. It commits already-staged files with [tag:id], never arbitrary first-in-progress task selection or automatic staging. In an SB-configured project, done/commit require passing gates for current contents.

list, show, stats, waves, tags inspect the DAG. tags with tag selects a phase. /todos puts a bounded report in the transcript and model history.

Pi mutation locks coordinate only Pi sessions, not standalone SCUD/swarm writers. Never mix those writers concurrently. SCUD's dependency conversion uses its own serializer; stop external writers first. CLI-only heavy/swarm operations remain available through bash when needed, outside Pi mutation activity.

Project configuration is discovered from cwd/ancestors within the repository, not from nested sketches. Large output is bounded; prefer show or filtered list over inspecting raw task storage. For details and limitations, see the package README.