Ai Code Security Auditor
Audits code and LLM integrations for security vulnerabilities, exploit paths, and agent-specific risks.
$5
Works with the AI Agents you already use
Ai Code Security Auditor
Example session with this skill installed
Audit the new authentication middleware and the OpenAI tool-calling logic I just wrote for security vulnerabilities. Is this safe to deploy?
- Read your context and instructions
- Compiled the ai code security
I have completed a security audit of the new authentication middleware and the LLM tool-calling logic. While the basic JWT validation is sound, there are two high-severity issues related to tool permissions and error logging.
Security Audit — Auth & LLM Integration
Critical
[src/auth/middleware.ts:24] Sensitive data leak in logs — The middleware logs the entire req.headers object when a token fails validation, which includes the raw Authorization header. Concretely, an attacker or a system admin with log access could harvest valid tokens from failed attempts due to unrelated malformed headers.
Fix:
// Change this:
console.error("Auth failed:", req.headers);
// To this:
console.error("Auth failed for user-agent:", req.headers['user-agent']);
High
[src/agents/openai_tools.ts:88] Excessive tool permissions — The execute_shell tool is exposed to the LLM without a restricted allowlist or path validation. An attacker could use prompt injection to trick the model into executing rm -rf / or exfiltrating .env files.
Fix:
const ALLOWED_COMMANDS = ['ls', 'git status', 'npm test'];
if (!ALLOWED_COMMANDS.includes(input.command)) {
throw new Error("Unauthorized command execution");
}
Medium / Low
[src/auth/middleware.ts:12] Weak JWT secret — The code uses a hardcoded default string for development. Ensure this is pulled from a secure environment variable in production.
Next steps
- Apply the log masking fix to
middleware.tsimmediately. - Implement the command allowlist in the tool-calling logic.
- Verify that your production environment variables are properly configured for the JWT secret.
Connects securely to your tools. The creator never sees your data.
About this skill
The problem
AI coding agents often prioritize functional code over secure code, frequently introducing predictable vulnerabilities like hardcoded secrets, injection points, and unsafe defaults. Standard linters catch syntax errors but miss the logic-based exploit paths that lead to production breaches.
What it does
- Executes a deterministic pattern scan for hardcoded secrets, weak cryptography, and unsafe deserialization.
- Audits authentication flows, authorization checks, and IDOR vulnerabilities against a specialized security checklist.
- Identifies agent-specific risks including prompt injection surfaces and excessive tool permissions in LLM integrations.
- Generates structured reports with concrete exploit paths and copy-pasteable code fixes.
- Classifies findings by severity to help developers prioritize critical patches over best-practice deviations.
Frameworks & tools
Works across all languages and frameworks, with specialized focus on LLM integrations, MCP servers, and common web backends (Node.js, Python, Go).
Why this beats prompting it yourself
Generic security prompts often produce "hallucinated" vulnerabilities or vague advice. This skill follows a rigorous multi-step workflow—scanning first, then verifying exploitability against the specific codebase context—to ensure every reported issue is a real threat, not a false positive.
Use cases
- Reviewing generated code from Claude Code or Cursor before merging a PR.
- Auditing LLM-powered applications for prompt injection and insecure tool-calling.
- Hardening authentication and payment processing logic in new features.
- Rapidly scanning a legacy repo for low-hanging security fruit like debug flags or wildcard CORS.
Known limitations
Requires access to the local filesystem to run the initial pattern scanner. Best results come from auditing scoped diffs rather than massive, multi-million line monorepos at once.
Requires local filesystem access for pattern scanning. Optimized for discrete diffs and feature modules rather than full-scale monorepo architectural reviews.
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 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
2 uses
Times used to date
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 2 months ago