Run the daily operating rhythm with agents that sweep every system each morning, surface what needs a decision, and keep nothing waiting silently. Use when running a business day to day and things fall through the cracks between desks.
Installs into .claude/skills of the current project.
Are you the author of Agent Ops Command?
Add the live security badge to your README. It updates with every re-scan.
[](https://www.skillsdirectory.com/skills/amey-thakur-agent-ops-command)
---
name: agent-ops-command
description: Run the daily operating rhythm with agents that sweep every system each morning, surface what needs a decision, and keep nothing waiting silently. Use when running a business day to day and things fall through the cracks between desks.
---
# Agent operations command
Most operational failure is not a crisis but an omission: an invoice
unsent, a customer unanswered, a deploy half done. A command desk exists
to sweep everything once a day and turn the sweep into a short list a
human can clear, which is the opposite of a dashboard nobody opens.
## Team
- **Sweeper**: checks every system for items past their expected state.
- **Prioritiser** (`prioritize-tasks`): orders findings by consequence
and deadline rather than by age.
- **Briefer** (`status-updates`): writes the one-page morning brief.
Shape: a scheduled daily sweep producing a single ranked action list.
## Method
1. **Define the sweep list once, explicitly.** Which systems, which
condition counts as stuck, and what the expected state is. A sweep
without defined expectations finds nothing.
2. **Look for absence, not just errors.** The invoice not sent and the
reply not written produce no alert anywhere, and they are most of
what goes wrong (see agent-company-blueprint).
3. **Rank by consequence and deadline.** A customer waiting two days
outranks a tidy-up task created this morning, regardless of which was
raised first.
4. **Cap the brief.** Ten items maximum, with everything else counted
rather than listed, because an exhaustive brief is one nobody
finishes reading.
5. **Assign every item an owner and a date.** An item on a list with no
owner is on the list again tomorrow.
6. **Close the loop the next morning.** Yesterday's items appear with
their status first, which is what prevents a rolling backlog
disguised as a daily brief.
7. **Escalate anything aging past a threshold.** Repeated appearance
without progress is itself the finding.
## Run it
In Claude Code, run the sweeper on a schedule across each desk's output
files and any exported system state, then the prioritiser and briefer in
sequence into a dated brief file. The human clears the list. Port to
CrewAI as a scheduled sequential crew, or LangGraph with a node per
system feeding one ranking node.
## Signals it works
- The brief opens with yesterday's items and their outcomes.
- Items have owners and dates, and aging items escalate.
- Things that quietly stalled are caught within a day rather than a week.
## Boundaries
This desk finds and ranks; it does not act on the items, and anything
touching money, customers, or public channels stops at a human. It
reports on systems it can read, so a system outside the sweep list is
invisible to it. Judgement about what actually matters this week remains
the operator's.