The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/compress.py invokes the claude CLI using subprocess.run. The implementation follows security best practices by avoiding shell=True and passing user file content through stdin rather than command-line arguments, which mitigates shell injection vulnerabilities.
[DATA_EXFILTRATION]: To perform its primary function, the skill sends the content of user-specified files to the Anthropic API. It implements a significant safety measure in scripts/compress.py via the is_sensitive_path function, which uses a heuristic denylist to block common sensitive files (e.g., .env, SSH keys, cloud provider credentials, and certificates) from being processed or transmitted.
[EXTERNAL_DOWNLOADS]: The skill uses the anthropic Python SDK and the tiktoken library as optional dependencies to manage API communications and token counting.
[SAFE_PRACTICE]: The skill incorporates a 500KB file size limit to prevent excessive resource consumption and potential denial-of-service conditions.
[SAFE_PRACTICE]: Backup files are stored in platform-specific local application data directories (e.g., %LOCALAPPDATA% on Windows) rather than the project directory. This prevents the agent from re-ingesting backup content into its active context, reducing noise and accidental information exposure.

caveman-compress