The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various shell commands such as find, grep, cat, and ls, and uses inline python3 snippets to extract metadata from configuration files (like package.json and tsconfig.json) and the project directory structure.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to transmit analyzed codebase metadata to the Google Gemini API (generativelanguage.googleapis.com) to generate the content for the CLAUDE.md file.
[PROMPT_INJECTION]: The skill contains a surface for indirect prompt injection because it reads untrusted data from the local filesystem and interpolates it into a prompt sent to a remote LLM.
Ingestion points: Reads local project files, including package.json, Makefile, and source files, which could contain malicious content if the project is untrusted.
Boundary markers: Project data is injected into a JSON payload; however, there are no specific delimiters or instructions to the model to ignore potentially malicious content embedded in the codebase data.
Capability inventory: The skill has capabilities for filesystem inspection, shell command execution, network communication via curl, and writing files to the local disk.
Sanitization: There is no evidence of sanitization or escaping of the ingested file content before it is included in the API request.

claude-md-generator