The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes shell commands using dependency names extracted from project manifest files. Specifically, the Node.js and Python fallback scripts in references/unused-detection.md construct ripgrep (rg) commands through string concatenation. The Node.js implementation uses execSync with a single string, which is vulnerable to command injection if a package name in package.json contains shell metacharacters like ;, &, or |.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch and run auditing tools using npx, uvx, and cargo install. These operations download and execute code from well-known public registries including NPM, PyPI, and Crates.io.
[REMOTE_CODE_EXECUTION]: The skill relies on dynamic code execution using python -c and node -e to perform repository analysis and symbol usage counting. This involves executing logic that is assembled by the skill at runtime based on the state of the repository.
[PROMPT_INJECTION]: The skill reads and processes untrusted data from repository manifest files and source code, creating a surface for indirect prompt injection.
Ingestion points: Project manifests (package.json, pyproject.toml, Cargo.toml, go.mod) and the full source code tree.
Boundary markers: Absent. The skill does not use delimiters or specific instructions to isolate content read from files.
Capability inventory: File system modification (package removal commands), shell command execution, and remote tool downloads.
Sanitization: The skill performs minimal regex escaping in some script fallbacks but lacks comprehensive sanitization of data retrieved from the codebase before it is presented to the agent or used in command construction.

dependency-pruning