The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to analyze external codebases and projects specified via the $ARGUMENTS parameter. This creates an attack surface where instructions hidden within the analyzed files—such as code comments, markdown documentation, or string literals—could attempt to influence the agent's review logic or report output.
Ingestion points: Reads files and directory structures from the project path provided in $ARGUMENTS (SKILL.md).
Boundary markers: The skill does not define specific delimiters or instructions to treat analyzed content as untrusted data.
Capability inventory: The instructions call for the agent to discover, analyze, and report on codebase structure and implementation, involving file system exploration and content analysis.
Sanitization: There are no instructions for sanitizing or escaping the content read from the external project path.
[SAFE]: The content of the skill consists of engineering best practices, design patterns (e.g., Tool Design, Agent Loops, Memory Management), and structured review templates. No evidence of malicious behaviors such as credential harvesting, unauthorized persistence, or malicious command execution was found within the provided instructions.
[EXTERNAL_DOWNLOADS]: The skill contains references to documentation and source code repositories for well-known libraries and services, including Anthropic's SDK, LangChain, and OpenAI's documentation. These references are used for architectural guidance and do not represent a remote code execution risk.

agent-review