The Agent Skills Directory

[SAFE]: No malicious logic detected. The skill provides comprehensive security research guidelines and reference materials sourced from the OWASP Cheat Sheet Series.\n- [COMMAND_EXECUTION]: The skill requires access to powerful tools like Bash and Task to enable the agent to conduct code research, search for patterns, and verify potential vulnerabilities. This capability is appropriate for the skill's primary function as a security auditor.\n- [SAFE]: Automated scanner hits for malicious patterns (e.g., eval with user input, reverse shell patterns, and obfuscated execution chains) in SKILL.md, languages/python.md, and references/supply-chain.md are false positives. These code snippets are explicitly documented as examples of high-risk vulnerabilities for the agent to identify during a review.\n- [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it is designed to analyze untrusted code. This risk is inherent to its primary diagnostic function and is mitigated by structured instructions and detailed reference materials that guide the agent in maintaining an objective analysis.\n
Ingestion points: Reads files and executes search commands on untrusted project codebases using Read, Grep, and Glob tools.\n
Boundary markers: Explicit delimiters for untrusted code are not defined in the instructions.\n
Capability inventory: Powerful system tools like Bash and Task are available to the agent.\n
Sanitization: Ingested code content is not automatically sanitized before processing.\n- [CREDENTIALS_UNSAFE]: The reference files (e.g., references/api-security.md, references/misconfiguration.md) contain numerous example credentials, API keys, and hardcoded passwords. These are presented as vulnerable patterns for identification and do not represent actual secrets used or exposed by the skill.

security-review