The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes source code from external GitHub repositories or local directories, creating a surface for indirect prompt injection where malicious instructions could be embedded in code comments or strings to influence the agent.
Ingestion points: Untrusted data enters the agent context through the analyze_github_repo tool (external repositories) and analyze_directory tool (local files).
Boundary markers: There are no explicit instructions for the agent to use delimiters or to ignore potential instructions found within the analyzed source code.
Capability inventory: The skill includes tools to read and parse source code files (analyze_file, analyze_function, analyze_directory, analyze_github_repo).
Sanitization: No sanitization or filtering of the analyzed code is described to mitigate the risk of embedded instructions.
[EXTERNAL_DOWNLOADS]: The analyze_github_repo tool fetches public repositories from GitHub using the git command for static analysis purposes.
[COMMAND_EXECUTION]: The skill uses the git system command to clone repositories, which is documented as being restricted to HTTPS URLs on GitHub.

time-complexity