All use cases

Work

Ground-test a routine before it runs 168 times a week

An hourly routine runs 168 times a week, so one bad assumption becomes 168 mistakes while you sleep.

Share

What hurt

Yaori (@Pashoke1) wrote a long “flight manual” for putting a bot on autopilot, checked against xAI's docs: one narrow job per bot, a written checklist as a skill, ground tests, and weather rules. Not affiliated with xAI.

What to install or create

Give the bot a job you can describe in one sentence without the word “help.” Save the process as a skill. Before scheduling: ask for the plan as a dry run, then test on something small and reversible, then do three runs including one ugly input (an empty list, a broken source). Add no-data, stale-data, and retry-limit rules, and review run history every Friday.

Bots

  • One-route bot

    Runs a single narrow routine with written rules and a run log.

Plugins

  • None on day one. Follow the first-run steps.

Routines

  • Friday debrief. Every Friday (author's suggestion). Read run history and fix what really happened, not what you assumed.

The prompt

Weather rules (author-quoted, excerpt)

Quoted from @Pashoke1 — as published. Do not invent a second system prompt.

no-data rule. "if there's nothing new, say so and stop." without it, bots invent work to look useful
stale-data rule. "if the data is older than X, don't act on it"
retry limit. 2 tries, then report. "keep going until it's done" is an open budget
idempotent steps. running it twice shouldn't double anything. check before you write
connectors over clicking.

First run

  1. 1Read https://x.com/Pashoke1/status/2107812715632656608 (credit @Pashoke1).
  2. 2Write the routine's job in one sentence, including what “done” or “nothing” looks like.
  3. 3Ask the bot for the plan only (a dry run) and correct misunderstandings in text.
  4. 4Test on something small and reversible, then three times, once with a broken or empty input.
  5. 5Add the weather rules below, schedule it, and read the run history on Friday.

Approval fence

The author notes a test run does real work, not a simulation. Point the first test at something small and reversible, and keep send, spend, and publish on ask-first.

What to tweak

Boring paperwork around the bot is what lets you close the laptop.

Source

Yaori @Pashoke1 · Original post

Not affiliated with xAI. We did not write this prompt. Third-party template terms still apply.

Browse templates

Browse team blueprints

Pro pack

Want the full setup, not just the idea?

Pro packs give you the job text, prompts, schedules and pitfalls for popular use-cases.

See Pro