The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill instructs the agent to perform global installations using npm install -g repomix and brew install repomix. While 'repomix' is a known tool, these commands download and execute code from external registries (NPM/Homebrew) at runtime.
COMMAND_EXECUTION (MEDIUM): The skill frequently uses npx repomix and repomix commands to process files. This involves executing a third-party CLI tool with various arguments, which could lead to arbitrary command execution if input patterns or file paths are maliciously crafted.
DATA_EXFILTRATION / INDIRECT PROMPT INJECTION (MEDIUM): The --remote flag allows the tool to fetch and process content from any owner/repo on GitHub (e.g., npx repomix --remote owner/repo).
Ingestion points: Fetches entire remote repositories into the local environment and packages them for the AI.
Boundary markers: None specified in the instructions for the packaged output.
Capability inventory: The skill can read local files, fetch remote content, and write outputs to the filesystem or clipboard.
Sanitization: While 'repomix' includes 'Secretlint' for sensitive data detection, the instructions explicitly show how to bypass this using --no-security-check, increasing the risk of accidental exposure of credentials found in remote code.
METADATA POISONING (LOW): The skill description mentions security audits and analyzing third-party libraries, which might lead a user to over-trust the tool's ability to safely handle malicious codebases.

repomix