The Agent Skills Directory

[SAFE]: The skill serves as a legitimate security reference library for AI agents. It provides actionable implementation guidelines and checklists to prevent common vulnerabilities such as SQL injection, Cross-Site Scripting (XSS), and insecure deserialization. The author identity and framework references (Project CodeGuard, CoSAI) are consistent with its security-focused purpose.- [INDIRECT_PROMPT_INJECTION]: The skill operates on untrusted code provided by users or read from external files, which constitutes a significant data ingestion surface. Malicious code could theoretically contain embedded instructions (e.g., within comments or long strings) designed to manipulate the agent during the security review process.
Ingestion points: Code snippets, source files, and configuration manifests across multiple programming languages and environments as defined in SKILL.md and various rule files.
Boundary markers: The skill does not mandate the use of specific delimiters or escaping mechanisms to separate the code-under-review from the agent's instructional context.
Capability inventory: The skill enables the agent to perform extensive code generation, modification, and file system analysis as described in the usage workflow of SKILL.md.
Sanitization: The rules recommend sanitizing inputs within the application code being generated but do not provide a mechanism to sanitize or filter the code that the agent itself processes for review.

codeguard