First Steps — factory

Factory is env plus one real task driven through the pipeline: you write a kickoff, ask the pipeline what to start, dispatch it, and harvest the finished branch. Experimental. The sequence's goal, from the source: env + one task driven through the pipeline.

Run it with the profile flag --profile factory (legacy equivalents: --with-aif-suite, --all). One precondition before you start: pick this only if this machine runs the aif-handoff operator runtime — the factory payload dead-ends without it.

The eight steps

1. Install at factory depth — bash <getff>/install.sh <stack> --profile factory. As with env: already on a shallower depth, re-running the same command adds the deeper payload on top.

2. Verify the payload landed — ls .claude/skills/. On top of env you should see dispatcher, aif-doctor, harvest, story and claude-glm-executor-handoff. (pipeline and night-mode are NOT factory additions — they arrive with env+ and are already there.) These are the pipeline's moving parts; the sequence below uses four of them by name.

3. Fill the project passport — replace every <…> placeholder field in .ai-factory/DESCRIPTION.md. Dispatched workers read this passport too — a vague one sends every dispatched task out with a vague brief.

4. Prove the rules are not inert on your layout — bash scripts/check-rule-globs.sh, then bash scripts/check-fences-fire.sh to see a rule go RED on planted input. On an empty skeleton check-rule-globs.sh fails by design (zero source files to match); re-run it once you have some. Dispatched workers inherit these gates on every branch they touch, which is why they are proven here and not assumed.

5. Read the tier + degradation SSOT — open .ai-factory/tier-home.md: it decides which tier a task routes to, and what degrades when a capability is absent. The public render of its degradation matrix lives on our Degradations page.

6. Write your first kickoff — create .ai-factory/orchestrator-prompts/<work-item>/kickoff.md: a Type: line (fix / research / feature), the goal, and — if the work splits into parallel sub-steps — a ## §1 Sub-wave section with one table row per sub-step. The kickoff is the whole brief the worker gets; write it so that a stranger could act on it.

7. Ask the pipeline what to start next — invoke /pipeline. It reads your kickoffs plus .ai-factory/orchestrator-prompts/plan.md (created on first run), ranks them, and emits a launch table. An empty backlog just renders the overview with zero open umbrellas — that is normal, not an error, and it means you are one kickoff away from a launch table.

8. Dispatch the top row and read the result — dispatch the launch table's top row, then bring the finished branch back with /harvest. If a task stalls or the runtime misbehaves, /aif-doctor is the diagnostic entry point.

Where this goes next

The same loop as a working day: Daily cycle — factory. What each tier means and what degrades when a piece is missing: Overview — multi-model pipeline and Degradations.