Docs / Workflows / Recipe system

Recipe system

A recipe is a saved workflow pattern: ordered stages (research → draft → publish), which Specialist template owns each stage, default duty prompts, cron schedules, and how artifacts pass through ~/files/. Recipes remove repeated copy-paste when hiring a multi-role team. Overview: workflows hub.

What a recipe contains

Each recipe step specifies:

• Template — Scout, Analyst, Watchdog, custom employee, or SEO audit roles.
• Role label — optional rename (for example “Researcher” instead of Scout).
• Duties — scheduled jobs with prompts; preset recipes ship with tested defaults.
• Skills — optional overrides (browser, files, integrations) per step when the playbook needs them.
• Minimum plan — the whole recipe is locked until your account meets that tier (Growth for most presets; SEO Audit Pipeline requires Max).

Preset catalog today: Content Pipeline, Research & Report, Monitor & Alert, SEO Audit Pipeline — listed on the Workflows tab in the dashboard.

Preset vs custom recipes

• Preset — pick a card in the gallery → review steps in the deploy confirmation modal → Deploy workflow. The API hires each Specialist, creates cron jobs, and wires Run next links between steps.
• Custom — Cloudia or the Workflow Architect returns a structured json:workflow-plan block in chat. The UI strips it from the message and shows Deploy this workflow; you still confirm in the same modal before anything is created. Custom plans can define up to ten steps with your own duty names, schedules, and prompts.

If deploy fails partway through, already-hired Specialists from that attempt are rolled back so you are not left with a half-built pipeline.

Shared workspace state

Good recipes point every duty at the same state file — for example ~/files/content-pipeline/state.json with items moving through statuses like planned, generated, published. Each stage reads the file, processes one eligible item per run (unless you explicitly asked for batch mode), writes artifacts under ~/files/{recipe-name}/, and saves the file before exiting.

That pattern makes failures visible: if a stage did not produce the expected markdown or JSON, the next cron tick has nothing to pick up. Guardrails: chaining tasks safely.

What deploy creates

1. One Specialist per recipe step (counts against your maxEmployees cap).
2. One or more scheduled duties per Specialist (counts against cron limits).
3. Within-role chaining — multiple duties on the same Specialist run in creation order.
4. Cross-role chaining — the last duty of step N triggers the first duty of step N+1.
5. Pipeline graph visible on the Team Task Board and Workflows tab.

Check Specialist and cron headroom before deploy — especially on Free (2 / 2). Limits FAQ · pricing.

After deploy

• Open the Workflows tab to see the graph; use + Add another workflow if you already have pipelines.
• Inspect duty threads on first manual run — prompts assume shared state paths; adjust in Specialist settings if your folder names differ.
• Edit Run next on the board if you add a manual step without redeploying the whole recipe.
• Hiring Content Pipeline or Instagram Manager may surface a nudge to review the Workflows tab — expected product behavior.

Why recipes help

• Less prompt drift — every stage uses the same playbook, not a one-off chat improvisation.
• Visible handoffs — missing files or stuck statuses show up in state.json and duty logs.
• Faster onboarding — Cloudia can propose a recipe before you hire three roles by hand.

Related: Chaining tasks · Scheduled tasks · Troubleshooting