Skill Security Auditor
Scan any Agent Skill ZIP/folder before install: catch hidden instructions, exfiltration, credential theft, dangerous shell, and obfuscation. Get a risk score plus a shareable trust report—optional CI/pre-install hook.
Free
Works with the AI Agents you already use
Skill Security Auditor
Example session with this skill installed
I just downloaded a skill folder called 'auto-coder-pro'. Can you audit it and tell me if it is safe to install? I am worried about it accessing my SSH keys.
- Read your context and instructions
- Compiled the security auditor
I have completed the security audit for the auto-coder-pro skill package. I found several undeclared side effects involving file system access.
Skill Trust Report
- Package: ./auto-coder-pro
- Scanned at: 2023-10-27 14:22 UTC
- Auditor: Skill Security Auditor v1.0.0
- Overall risk: High (62/100)
- Install recommendation: Block
Declared behavior
The skill claims to automate boilerplate code generation and refactoring tasks within the local workspace. It declares no network requirements.
Observed behavior / side effects
Static analysis revealed the skill attempts to read files outside the project root, specifically targeting hidden directories in the user's home folder.
Findings
| Severity | ID | File:line | Summary | Evidence |
|---|---|---|---|---|
| High | CR | scripts/utils.py:42 | Credential path access | os.path.expanduser('~/.ssh/id_rsa') |
| Medium | FS | SKILL.md:12 | Undeclared file read | Instruction to "read config from home dir" |
| Low | PK | N/A | Missing version | No version tag in frontmatter |
Side-effect declaration vs reality
| Claim | Reality | Match? |
|---|---|---|
| Workspace only | Accesses ~/.ssh | No |
| No Network | No outbound found | Yes |
Scripts & network surface
The script scripts/utils.py contains logic to check for the existence of SSH private keys. No active curl or fetch commands were found, but the presence of key-reading logic in a "coding" skill is a critical red flag.
Constraints if installing anyway
- Run exclusively inside a Docker container with no volume mounts to the home directory.
- Manually remove the credential-reading functions from
scripts/utils.py.
Next steps
- Review the flagged code in
scripts/utils.py:42. - Do not grant this skill access to your terminal until the SSH-related instructions are removed.
- I can generate a sanitized version of this skill if you want to proceed safely.
Connects securely to your tools. The creator never sees your data.
About this skill
Pain
Your agent often runs skills with broad permissions. One shady SKILL.md can hide prompt injection, call home, or fish for secrets—and a pretty marketplace blurb is not a security review. Teams stall installs because they cannot prove what a skill will do.
What it does
- Inventory a skill ZIP or folder; compare declared behavior vs what the files actually contain
- Heuristic + agent-assisted checks for prompt injection, exfil URLs, credential theft, dangerous shell, obfuscation, and related patterns
- Risk bands Low → Critical with a clear Allow / Allow-with-constraints / Block recommendation
- Writes a shareable markdown trust report; optional
scan_skill.pyfor a pre-install or CI exit-code gate
When to use
- Before installing any third-party skill ZIP or folder
- When a skill “feels off,” asks for unexpected network/secrets, or ships opaque scripts
- When you want a repeatable trust gate in CI or a team install checklist
Trust & approval
Read-only by default—no call-home. Recommendations are explicit (Block vs Allow). High-impact actions need your approval. Complements Agensi’s own 8-point scan; this is an install-time second opinion for your machine/team, not a formal certification.
Limitations
Static/heuristic review can miss novel attacks. Does not execute untrusted skill scripts with your network or secrets. Not a substitute for Agensi marketplace scanning or your own security policy.
Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Security.
Static/heuristic review can miss novel attacks. Does not execute untrusted skill scripts with your network or secrets. Not a substitute for Agensi marketplace scanning or your own security policy. Not a formal certification.
How to use
- 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
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before buying a skill.
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
Free to use
No account required to browse
Trust & safety
Security scan
Security scan passed 13 days ago
- Free with an Agensi account
Permissions
Local read of a skill ZIP/folder, optional offline run of scripts/scan_skill.py, and a markdown trust report written in the workspace. No network and no environment-variable access: those strings are detection patterns inside untrusted skills, not capabilities this skill uses. Browser is not required.