Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

1 Commit
 
 
 
 

Repository files navigation

agentcheck

Find out what is going to break your agent setup, before it does.

You installed Claude Code or OpenClaw or Codex, it worked, and now you are running agents against real work. This tells you which of the twenty-one things that reliably go wrong are already wrong on your machine.

curl -fsSL https://raw.githubusercontent.com/belkenbot/agentcheck/main/agentcheck -o agentcheck
chmod +x agentcheck
./agentcheck

No install. One file. Delete it when you are done.

What it will not do

Read the script before you run it, it is one file and it is commented. But so you know what you are reading:

  • No network calls. Not one. It cannot phone home because there is nothing to phone home with.
  • Read-only. It never writes, moves, or deletes anything.
  • Nothing leaves your machine. No telemetry, no upload, no account, no email gate.
  • It never prints a secret. When it finds an exposed credential it tells you the file and the permissions, never the value.

What it checks

Memory — whether your agent will still know anything tomorrow, and whether your instruction file has grown past the size where the tail gets silently truncated at load. That one is nasty. There is no error. The rules at the bottom of your file just stop reaching the model.

Cost — whether every trivial task is going to a frontier model. Summarising, tagging and classifying are the bulk of agent traffic and none of it needs the expensive model.

Blast radius — whether anything is written down that tells your agent where to stop, whether you have a blanket wildcard in your permission allowlist, whether unattended execution is on globally, and whether your scheduled jobs are landing inside live sessions.

Secrets — credentials in world-readable config, and .env files sitting in a git repo that does not ignore them.

Durability — no backup of the config you spent months tuning, hand-patched vendor code that the next update will silently revert, a Node version in the known dead zone, transcript sprawl, and disk headroom.

Observability — whether anything would tell you that an agent stopped working, and whether your scheduled jobs are throwing their errors into /dev/null.

Where these came from

A four-machine fleet running since March: ninety-plus agents, a shared Postgres memory layer, cron orchestration, browser automation, running real production work every day.

Every check in this file is a failure that actually happened there. Not a best practice from a blog post. A specific thing that broke, on a specific date, with a specific cost. A few examples of what is behind the checks:

  • A scheduled job configured against a live session key injected a foreign model into that conversation, forced a compaction, and wiped working memory. The symptom showed up days later as "the agent forgot everything" and took weeks to trace.
  • A hand-patched dependency was reverted by a routine update. The tool still ran, so nothing looked broken, and the capability it provided quietly disappeared.
  • One hundred and eighteen connection-refused lines in a log were written off as cosmetic noise. They were the actual blocker. Nothing shipped for weeks because the real error was hiding in noise nobody had cleaned up.
  • A credential was hardcoded into a scheduled job's prompt. After a rotation the job read the current value and put it in an alert message.

The full writeup of what broke, what it cost, and how each one got fixed: https://belken.ai/field-manual

Exit codes

Always 0. This is a report, not a gate. If you want it in CI, parse the output.

Licence

Do what you want with it. Fork it, rename it, ship it inside your own thing. If you add a check that caught something real, open a PR and say what it caught.

Built by Belken Ventures.

About

A read-only health check for AI agent setups. Memory, cost, blast radius, secrets, durability, observability. No network calls, nothing leaves your machine.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages