The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is designed to search for and extract hardcoded secrets from codebase files. The CATEGORIES.md file includes specific patterns for identifying AWS access keys (AKIA[A-Z0-9]), private keys (-----BEGIN.*PRIVATE KEY), and generic credentials (password.*=, api_key.*=, secret.*=). The workflow then instructs the agent to present these as evidence in the audit report.
[DATA_EXFILTRATION]: The skill identifies and accesses highly sensitive file paths, including .env files, credential stores, and private key files. During the reporting phase, it is instructed to include code snippets or pattern matches as evidence, which results in the exposure of the discovered sensitive data within the agent's output session.
[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes arbitrary external codebase files, creating a significant attack surface for indirect prompt injection. A malicious codebase could include instructions in comments or string literals designed to subvert the agent's analysis or logic.
Ingestion points: Files are read via Glob and Read tools during project reconnaissance (Phase 1) and evidence gathering (Phase 4).
Capability inventory: The skill utilizes Glob, Grep, Read, and Write tools throughout its 5-phase workflow.
Boundary markers: The instructions lack delimiters or explicit warnings to the agent to ignore any natural language instructions discovered within the files being audited.
Sanitization: There is no evidence of sanitization, filtering, or escaping of the content read from the codebase before it is interpolated into the agent's reasoning process or the final report.

security-audit-owasp-top-10