FAQ

Do I need bb to use Stelow? No. Path B (skills-only) runs the identical workflow on any Agent Skills-compatible host. bb adds the visual layer (board, inbox, questions, presets) — see overview.md.

Which path should I pick? Pick bb if you want visual review gates, a quiet inbox instead of chat pings, and GitHub issue automation. Pick skills-only if you already live in another agent and are comfortable with slash commands and receipts on disk.

Who decides run knobs and Review Mode? You do, once, at setup. The LLM never widens knobs and never skips a gate you asked for. Experimental + Auto is the fastest path (~6 stages, no questions); production quality with full review runs all 18.

Does the AI estimate the work? No. Knobs are constraints you declare ("how much may this consider?"); the workflow only checks fit (fits / cuts_needed / reshape) and proposes cuts. There is no estimation step to be overconfident about.

What happens when the agent needs me? It opens a structured question and waits — in bb as a form on the card, on other hosts via the ask protocol. Unanswered questions stay answerable; the agent does not guess ahead.

Is parallel scope execution safe? Sequential is the default and stays safe by construction. Parallel dispatch is opt-in with file locks plus a post-hoc overlap audit, and any overlap routes to you (merge, re-run, or rework). See scopes-tasks-records.md.

Does it work on brownfield codebases? Yes, with a caveat: Tech Preview and Feature Recon surface architecture and constraints before shaping, but a codebase with years of undocumented conventions will still surprise the agent. Greenfield is the home turf.

Where is my workflow state? stelow.json (tracking), state.md (per-workflow frontmatter), .stelow/ (artifacts, approvals, locks). Hosts read these; no host keeps a second workflow database. scripts/stelow doctor detects drift.

Can I turn off the GitHub automation? Yes: STELOW_GITHUB_ISSUES=0 on the host disables the scheduler, its RPCs, and the panel button without touching anything else.

Is the API stable? No — pre-1.0 (0.x). Skills, contracts, and the CLI may change incompatibly before 1.0.0. See status.md.

Who reviews the reviewers? Critique stages use fresh-context reviewers (same model, clean session), never the worker reviewing its own output in place. Independent cross-lineage review is opt-in (bb stelow review on bb).