## Create pipeio-flow-create skill and pixecog-flow-setup skill¶

Background¶

Goal¶

(promoted from note)

Context¶

(see source note)

Prompt¶

Fix the issue described in /storage2/arash/projects/projio/docs/log/issue/issue-arash-20260410-150432-267829.md. Read it first, then implement the proposed fix.

Acceptance Criteria¶

[ ]

Result¶

Completed in 367.6s (1 steps, model: opus)

Session: e3933927-ea0e-4c8c-9b79-b71c3239e6dd Result note: /storage2/arash/worklog/workflow/captures/20260410-151114-5349c1/note.md

Output¶

The skill is discoverable. Both tasks are done:

1. **`pipeio-flow-create`** — Created at `docs/prompts/skills/pipeio-flow-create/SKILL.md` in projio. Verified with `skill_read()` — it's immediately discoverable as an ecosystem skill. Covers the full lifecycle: survey → scaffold → config → Snakefile → scripts → datalad → verify, with concrete templates and guardrails for each friction point from the issue.

2. **`pixecog-flow-setup`** — Dispatched to pixecog as task `74026b8005d0`. It layers pixecog-specific conventions (cogpy env, raw layout, datalad subdatasets, cogpy/labpy imports, sub-test mock paths, BidsPaths adapter patterns) on top of the generic skill.

Both skills encode all five friction points from the issue:
- Don't overwrite scaffolded Makefile (step 2 hard rules)
- Keep `configfile: "config.yml"` (step 4 key patterns)
- Call `pipeio_registry_scan` after `pipeio_flow_new` (step 2)
- Use `pipeio_run` MCP tool, not manual conda (step 7 + guardrails)
- Pipeline env is `cogpy` (pixecog skill, conventions section)

## Next Steps

1. Run `projio sync` to generate the `.claude/commands/pipeio-flow-create.md` slash command wrapper so it's available as `/pipeio-flow-create` in Claude Code.
2. Check the pixecog dispatch status with `tail_task(queue_id="74026b8005d0")` and verify the skill was written correctly.
3. Test the skill end-to-end by invoking `/pipeio-flow-create` in a pixecog session to create a new flow (e.g., `preprocess_motion`).