| name | sf-flow-auto |
| description | Use when a defined flow (.pi/workflows/<name>.yaml) must be run end-to-end with no human gates. Input may be a prompt, a markdown file, a PRD, or a Jira story. |
sf-flow-auto
Purpose
Run a defined flow end-to-end, no human gates. Input forms: inline prompt, markdown file, PRD, or Jira story (resolved via @pi-stef/atlassian).
Process
Phase 1: Resolve the flow
Read .pi/workflows/<workflow>.yaml. Validate it (validateFlowYaml). Generate the pi-dw script (idempotent). If the flow isn't registered, error with the create hint.
Phase 2: Resolve the input
prompt → use verbatim as the flow's args.input
md-file → read the file, pass contents as args.input
prd → parse the PRD file, pass as args.input
jira → resolve the story via @pi-stef/atlassian (Jira), pass description+acceptance as args.input
Phase 3: Run the flow
Execute the generated pi-dw script with args.input. Phases run sequentially; intra-phase fan-out via parallel(). Loops (until_dry / until:approved) run to completion. tmux visualization (if enabled) shows per-agent panes.
Phase 4: Terminal state
Each phase exits success / no-op / blocked / exhausted. On blocked/exhausted, stop and report. No human gates; on completion return the flow's result.
Phase 5: Telegram
Send a completion summary via notify-telegram.sh.