The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the repomix and npx CLI tools via the Python subprocess.run function in scripts/repomix_batch.py. The command is built as a list of arguments, which effectively mitigates shell injection vulnerabilities.
[EXTERNAL_DOWNLOADS]: The skill uses npx to download and run the repomix package and can fetch repository data from remote URLs (e.g., GitHub). These operations target well-known and reputable services.
[DATA_EXFILTRATION]: The skill reads local files to package them and specifically searches for .env files in parent directories to load configuration variables. While this is a common pattern for skill configuration, it involves accessing potentially sensitive file paths.
[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface because its primary purpose is to aggregate untrusted third-party code for AI consumption.
Ingestion points: Local file system paths and remote repository URLs processed by scripts/repomix_batch.py.
Boundary markers: The tool uses XML or Markdown tags to delimit files, although it does not inject explicit instructions for the AI to ignore instructions within the content.
Capability inventory: All external interactions are handled through subprocess.run calls to the repomix utility.
Sanitization: The skill integrates Secretlint to detect and warn about hardcoded secrets, but it does not filter or sanitize the natural language content of the files.

repomix