Agent Security Review

    1

    A first-pass audit of agent config files, then a checklist: credentials, untrusted content, network, shared memory.

    $90

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Agent Security Review

    Example session with this skill installed

    Do a security review of how coding agents are set up in this repo. Report what you find and do not change any file.

    • Read your context and instructions
    • Compiled the agent security review

    Summary

    The agent ran the file audit, then walked the four areas of the checklist (credentials, untrusted content, network, shared memory) and ranked the fixes. It changed no files and did not print the secret value. Output below is from running the audit script on a small test project.

    Audit output

    $ python3 scripts/audit_agent_files.py .
    HIGH .claude/settings.json:1: permission checks switched off
    HIGH .mcp.json: MCP server tracker uses plain http
    HIGH .mcp.json: MCP server tracker holds a literal value for API_TOKEN
    HIGH .mcp.json:1: secret-like value (assigned secret)
    WARN .gitignore: .env is not listed, so a secret file could be committed
    WARN AGENTS.md:1: pipes a download into a shell
    INFO .mcp.json: MCP server docs is remote; its replies are untrusted text
    

    Exit code 1, because of the HIGH lines.

    Ranked fixes

    AreaQuestionAnswerEvidenceFix
    CredentialsDoes a config hold a literal secret?yes.mcp.json, server trackerUse ${API_TOKEN} and set it in the environment. Rotate the token, because it was in a file.
    Untrusted contentCan the agent run a command after reading outside text without approval?yes.claude/settings.json:1Turn permission checks back on.
    NetworkDoes an MCP server use plain http?yes.mcp.json, server trackerUse https, or a server on this machine.
    Shared memoryWho can edit the instruction files?not verifiedCheck repo write access and review changes to AGENTS.md.

    What was not checked

    Dependencies, whether any hook really blocks anything, and the machine itself. A clean audit would not be proof of safety.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    When a coding agent works in your repo it reads text, runs commands and holds your credentials. This skill reviews what it can reach and what can reach it.

    Step one is a script over the files that steer an agent (CLAUDE.md, AGENTS.md, MCP and settings files, hooks, rules): secret-like values (never printed), permission checks switched off, MCP servers on plain http, unpinned packages, literal tokens in config, a tracked .env file, a .gitignore without .env, a download piped into a shell.

    Step two is a 21-question checklist across four areas. The agent answers each as yes, no or not verified, with the file and line as evidence, and ranks the fixes. It changes nothing.

    Standard library only. Written for Python 3.10 or later, tested on 3.13 only. 15 tests. A file it cannot read or parse is reported, never skipped silently.

    What it does not do It reads a fixed list of files and a few repo settings. It does not run the project, scan dependencies for known vulnerabilities, or test whether a hook blocks anything. A clean audit is not proof of safety. It does not fix anything.

    Before you use this These skills are new and have not been tested widely. They come from a project that works and that I use. You are responsible for what you run and for reviewing what your agent does with them, so check results before you rely on them. The author is not liable for loss or damage, in line with the MIT licence.

    Skips symlinks and files over 500 KB. Does not run project code or check dependencies for vulnerabilities. Requires manual verification of skipped files.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed today

    Requires

    Python 3

    Permissions

    Terminal / Shell
    Read Files

    No network access. It reads project files and runs git ls-files to find tracked env files. It writes nothing and changes no file.

    Listedtoday

    What's inside

    Frequently Asked Questions

    A fixed list of agent files: CLAUDE.md, AGENTS.md, GEMINI.md, MCP and settings files, Cursor, Kiro and Copilot instruction files, and hook files. It looks for secret-like values, permission checks switched off, MCP servers on plain http or running unpinned packages, a tracked .env file, a .gitignore without .env, and a download piped into a shell.

    No. It prints the kind of finding and the file and line, never the value. Obvious placeholders such as your_api_key_here are ignored.

    No. The script and the review only report. Fixing what they find is a separate task that you start.

    No. The script reads a fixed list of files and a few repo settings. It does not scan dependencies, run the project, or test whether a hook blocks anything. The checklist asks the agent to mark anything it could not verify as "not verified".

    The script runs on any project folder with Python, and the instructions are written for agents that read SKILL.md. I have tested it with Claude Code only.

    The SKILL.md instructions, the audit script, a 21-question checklist, an examples file and 15 unit tests. Standard library only.

    New in Security