routine-authorlisted
Install: claude install-skill crewship-ai/crewship
# Routine Author
A playbook for turning "make a routine that does X" into a valid, saved Crewship
draft — grounded in what this crew actually has, reviewed and published by the user.
## When to Activate
- The user asks you to **build / create / automate a repeatable workflow**:
"make a routine that…", "automate X", "set up a recurring job", "every morning do Y".
- Distinguish from a one-off task: a routine is worth authoring when the work
repeats or needs a trigger (schedule / webhook / event). For a single ad-hoc
job, just do the work — don't author a routine.
## Procedure
> **Author from what you already know — do NOT probe or verify.** Everything you
> need is already in your prompt: `[CONNECTED INTEGRATIONS]`, `[AVAILABLE ROUTINES]`,
> and the DSL reference below. Writing a routine is a **paper exercise** — you
> *describe* steps, you don't *perform* them now. So:
> - **Do NOT** run a command to "check" / "list" integrations, routines, or
> endpoints — you already have them in your prompt.
> - **Do NOT** fetch the source URL to "see what it returns." The routine's own
> `agent_run` step extracts the data at RUN time; put "extract the top 5 …" in
> that step's prompt and move on.
> - **Do NOT** test a webhook, ping a host, or re-fetch the routine after saving
> to "verify." None of that is authoring.
> - First call `get_routine_draft` for the intended slug. This loads the same
> saved draft the user edits, or a revision-zero baseline for a new recipe.