The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes npx to download and execute the fallow package from the npm registry at runtime. The dependency is not version-pinned, which allows for the potential execution of unverified or malicious code if the package is updated or compromised.
[COMMAND_EXECUTION]: The skill executes multiple shell commands to create temporary directories, run external audit tools, and perform data processing via Python heredoc scripts.
[INDIRECT_PROMPT_INJECTION]: The skill ingests data from the scanned codebase and presents it to the agent, creating a surface for indirect prompt injection if the source code contains malicious identifiers.
Ingestion points: Scanned function names and file paths are extracted from JSON files in the .agents/tmp/ directory.
Boundary markers: Audit findings are displayed to the agent without protective delimiters or instructions to ignore embedded commands.
Capability inventory: The skill possesses shell execution, file system write access, and Python script execution capabilities.
Sanitization: No escaping or validation is performed on codebase identifiers before they are injected into the agent's context.

improve-codebase