The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from local source code files to identify bugs and suggest fixes. This creates a vulnerability surface where malicious instructions embedded in code comments or logic could influence the agent's actions. Evidence: The process involves reading modified files completely (SKILL.md) and fixing issues immediately without explicit boundary markers or content sanitization to distinguish between data and instructions. Capabilities include executing shell commands and writing to the filesystem.
[COMMAND_EXECUTION]: The skill executes shell commands using git to identify modified files in the repository. Evidence: git diff --name-only HEAD~1 and git diff --name-only --cached are used to determine the scope of the review.

fresh-eyes