How to run agents on runbook scripts for site reliability engineers
Steps for running agents on runbook scripts, with a tool-neutral checklist for site reliability engineers and the options to compare before you choose.
SupaKeeper is published by LunarWerx, the makers of AgentHydra, Copilot AI Productivity Suite, Codex Forecast and Model Odds, which are compared here alongside other tools.
To run agents on runbook scripts, decide which scripts an agent may run, limit its access to what each task needs, check every output against what you wrote down, and choose a tool that fits how your runs are organized.
| Option | What it can do | What to check |
|---|---|---|
| AgentHydra | Shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. [1] | Check that the sessions your runbook runs use appear in the list, and that the view fits how you organize work. |
| Terminal runs of runbook scripts | Run each script yourself from the shell, with your own logs. | Check how you will record output and how you will stop a run that goes wrong. |
| Separate agent or automation tool | A separate tool for running agents or automated steps. | Check what it shows, how it lets you stop a run, and where its logs are kept. |
What you need before you start
Before you run agents on runbook scripts, keep the scripts in one place that you can review and roll back, such as a version-controlled repository. For each script, write down the expected output of every step, so you can tell when a run has gone off track.
Know which systems each script touches, which accounts the agents will use, and which steps must have a person approve them first. Decide ahead of time what an agent may change and what it may only read. Make sure you have a clear way to stop a run and to see what it has already done.
Give each agent only the access its job requires. Keep the shell, credentials and permissions the scripts need on the machine you plan to use, and write down who owns each runbook so that someone can make a decision during an incident.
Step by step
The steps below work with any agent and any runbook script. Each one is a habit you can apply on its own.
- Choose one runbook script with a narrow goal, such as restarting one service or clearing one cache.
- Give the agent the script, the target environment and the limits it must keep, written in plain words.
- Run the script first in a test environment and compare each step with its expected output.
- Ask the agent to say what it will do before each change, and approve every change that touches a live system.
- Keep the run's output and any command that failed, so you can review what happened later.
- Stop the run as soon as an output differs from what you wrote down, and check the system before you continue.
- When a failed command has a known fix, note the pair so the next run avoids the same error.
- After the run, update the runbook with anything that surprised you, and remove steps that no longer apply.
Tools that can do it
AgentHydra is made by the team that publishes this site.
AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. [1]
AgentHydra lets you type a reply into a Claude session's own desktop chat, or send one message to several sessions. [1]
AgentHydra pairs each failed shell command with the command that fixed it, and can copy those pairs out as a rules file for your agents. [1]
AgentHydra exposes its features as MCP tools, which are functions an agent can call, including checks that tell an agent which account it is on and how much quota is left. [1]
AgentHydra is no longer available to the public: LunarWerx, its maker, now uses it only in house. [1]
AgentHydra is no longer available to the public. [1]
AgentHydra is no longer available to the public. [1]
AgentHydra is no longer available to the public. [1]
Other approaches include running each script yourself from a terminal with your own logging, or using a separate agent or automation tool. For either one, check what it shows, how it lets you stop a run, and where its logs are kept.
Common mistakes to avoid
Avoid running a new script against live systems before you have tried it in a test environment. A script that works on one system can behave differently on another.
Avoid giving an agent broad access because it is convenient. Access that is wider than the task makes a mistake harder to contain.
Avoid skipping the output check. If you only look at the final result, you may miss a step that went wrong early in the run.
Avoid losing the record of failed commands. Without it, the same error will return in the next run, and no one will know the fix.
Avoid leaving the runbook out of date. When a script changes, update the written steps at the same time, so the next person follows the current version.
Frequently asked questions
How do site reliability engineers run agents on runbook scripts safely?
Run each script in a test environment first, give the agent only the access that one task needs, and require a person to approve any step that changes a live system. Compare each output with what you wrote down, and stop the run when they differ. Keep a record of commands and results so you can review what happened after the run and update the runbook with anything new you learned.
Does AgentHydra show agent sessions for runbook work in one list?
AgentHydra shows Claude Code, Codex and OpenCode sessions on your machine in one searchable list, updated live while they run. [1]
Can AgentHydra turn failed commands into rules for agents?
AgentHydra pairs each failed shell command with the command that fixed it, and can copy those pairs out as a rules file for your agents. [1]
Is AgentHydra free to use for runbook work?
AgentHydra is no longer available to the public. [1]
Which operating systems can AgentHydra run on?
AgentHydra is no longer available to the public. [1]
AgentHydra is no longer available to the public: LunarWerx, its maker, now uses it only in house. This page keeps its comparison for reference. [1]
Sources
- AgentHydra's own description of itself (the product of this site's publisher)
Related comparisons
- How to advise customers on AI model upgrades
- How AI community managers answer release date questions
- How to anticipate rate limit interruptions for API integrators
- How to audit agent commands in each session in 2026
- How to audit agent sessions across projects
- Audit What an Agent Ran in Each Session for DevOps
- How freelance coders avoid hitting limits mid-task
- How to avoid hitting the weekly cap for Claude users
- How to avoid running out mid-project in Cursor
- How to budget AI usage per sprint