Reported, not yet confirmed.
A checker bot that looks for the output, not the green run
One Grok Bot user's checker bot caught a sync that reported success for six days without writing anything.
You can stop trusting a green "succeeded" and have a bot check that the work actually showed up. @grovr_au, who says they use Grok Bot every day, runs one bot that "checks whether a routine actually did its job, not just whether it says it ran. that's how it caught a sync that had 'succeeded' for six days without writing anything". This is one user's report, from an account with 49 followers and no stated affiliation. We have not run it.
What it replaces#
Before: a routine reports success, and you only notice the missing output when you go looking. After: a separate bot goes looking for you. The author says that failure "has bit me enough times to warrant it's own bot."
How it works#
- One job per bot. "each bot gets one job, plus a short list of things it never does and which bot handles those instead."
- Methods live in skills. "the actual how-to lives in shared skills."
- A weekly skills trim. "one bot's whole job is checking those skills every week for anything stale or repeated, and trimming it down." The stated reason: "less context overload, fewer tokens".
- A checker that looks where the output lands. Asked how it verifies, the author replied: "it checks where the output is expected to see if it landed. if that's in chat it checks there, if it's in slack it checks there etc."
Starter prompt (ours, untested; built from the author's description, since none was published):
You are my routine checker. For each of my routines, read its saved
prompt and write down where its result should appear (this chat, a
Slack channel, a file, a doc). Every weekday at 9 AM, look in each of
those places for output from the last run. List any routine that says
it succeeded but left nothing new, with the last date it really did.
Do not edit, pause or rerun anything. Just report.
Good to know#
- "Succeeded" only means the run finished. A run can finish and still write nothing.
- The checker needs read access to every place your routines write to, or it will report false alarms.
- Keep it report-only. The fix belongs to you or the bot that owns the job.
- This is different from a cost audit like Dr Eggbot, which flags unused or expensive routines.
Same idea in Grok Bot#
This is already a Grok Bot setup. Put each routine's "where should the result appear" line in a skill the checker reads, and give the checker its own weekday routine.
