The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes three Python scripts (scripts/caveman_compressor.py, scripts/caveman_lint.py, and scripts/token_savings_estimator.py). Analysis confirms these scripts are limited to text processing using standard libraries (re, json, argparse). They do not perform network operations, file system modifications, or arbitrary command execution.
[PROMPT_INJECTION]: The SKILL.md instructions define a specific communication style ('caveman mode'). While it includes instructions for persistence ('ACTIVE EVERY RESPONSE once triggered'), these are behavioral guidelines for the agent's persona. Importantly, the skill includes 'Auto-Clarity Exceptions' that require the agent to drop the persona for security warnings, irreversible actions, and clarifications, which acts as a built-in safety guardrail.
[DATA_EXPOSURE]: No hardcoded credentials, sensitive file paths, or data exfiltration patterns were identified. The scripts operate on provided text inputs and output to the console.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute external code. All dependencies are handled via the Python standard library, and the references point to well-known development resources for documentation purposes only.
[OBFUSCATION]: No malicious obfuscation techniques such as Base64-encoded commands, zero-width characters, or homoglyph attacks were detected. The scripts use standard character-masking logic to protect code blocks during text processing, which is a legitimate functional requirement.

caveman