Create your security check prompt

Ready for your agent. Tailored to your starting point.

Preparing your prompt…

Choosing defaults for this browser. Your agent runs the checks.

This version covers PHP, GitLab, Node.js, LiteLLM and eMudhra. Railway advice is included when your agent detects its CLI.

Where do I paste this?

Copy from Patchwatch. Paste into an agent that can access your computer or project.

Open a local Codex task

Open Codex on the computer you want to check.
Choose your project folder and a local task. For a machine check, use a dedicated folder and keep “This machine” selected here.
Paste the copied prompt into the task’s message box and send it.

A project check needs the actual project folder. A cloud chat checks its own environment unless connected to your machine. A normal web chat without tools will ask you to run commands.

Codex setup guide

Codex illustration: choose your folder and Local, then paste into the message box. — Illustrated paste location · not an assessment result

Watch the website walkthrough · 22 seconds

Recorded page views: defaults → Customize → Copy → paste guide.

Download the GIF

What happens after I send it?

1
Choose your target
The defaults select this machine and suggest an OS from your browser. Use Customize for a project or remote hosts; your agent confirms the execution environment.
2
Paste the prompt into your agent
Use Codex, Claude Code, or another agent with access to the target. For a project, open its repository first. For remote hosts, provide their names and approved access method.
3
Review what applies
Your agent checks installed software and running services, verifies the vendor guidance, and reports findings with evidence. If it cannot run a check, it asks you for the output.
4
Approve and verify fixes
Review the proposed change. Run any sudo or Administrator command yourself and share redacted output. Your agent verifies the running version and service health.

Optional eMudhra diagnostics and credits

eMudhra checks

These optional checks test the client you run them with. Browser, Node.js and container results can differ.

Command recipes and result artwork: eCHUDhra . Research and test infrastructure: Dennis Jackson and community contributors , crediting original reporter Wayne. eCHUDhra is a parody site, unaffiliated with eMudhra.

A failed connection can still show safe-looking artwork. Check the certificate error and a working control connection; one endpoint does not cover all affected chains.

curl · current client

POSIX shell syntax. Tests this curl build and its configured trust store.

curl -fsS https://dv-g3b-aaa-emudhra.we-have-a-problem.com >/dev/null 2>&1 && curl -s https://echudhra.com/chud.txt || curl -s https://echudhra.com/chad.txt

Node.js · current runtime

Tests the Node executable you launch. Other installed runtimes can differ.

node -e "fetch('https://dv-g3b-aaa-emudhra.we-have-a-problem.com').then(()=>'chud').catch(()=>'chad').then(a=>fetch('https://echudhra.com/'+a+'.txt')).then(r=>r.text()).then(console.log)"

Docker · new Alpine container

Ask before running: may pull an image, creates a temporary container and installs curl inside it. Does not test your host or existing application images.

docker run --rm alpine sh -c "apk add -q curl; curl -fsS https://dv-g3b-aaa-emudhra.we-have-a-problem.com >/dev/null 2>&1 && curl -s https://echudhra.com/chud.txt || curl -s https://echudhra.com/chad.txt"

Read the full diagnostic guidance

Got your results? Close the loop.

Keep the same agent conversation open. Approve the fixes you want, then ask it to recheck.

CheckApprove fixesRecheckResolve or name the gap

Done means each selected issue has evidence of a fix or does not apply in the checked scope, with no pending restarts or unexplained gaps. Incident recovery can also require credential rotation and investigation.

Use after your agent has reported or made approved changes.

Paste a report to see what is still open

Paste only the agent’s JSON summary, with secrets removed. This stays in this tab: no upload or saved history. Patchwatch checks report structure and flags missing evidence; it cannot independently validate fixes or certify safety.

Agent’s JSON report