The Agent Skills Directory

[COMMAND_EXECUTION]: The skill includes local Python scripts (evaluate_matrix.py, session_scaffold.py, validate_register.py) for data processing. These scripts rely exclusively on the Python standard library and do not perform shell command execution or spawn subprocesses.
[DATA_EXFILTRATION]: No network operations or data exfiltration patterns were found. The instructions and scripts explicitly state that they make no network or LLM calls, ensuring that research data remains within the local environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied data in CSV and JSON formats. It mitigates potential injection risks by utilizing structured data parsing, deterministic logic in Python, and rigorous input validation (such as file size limits and type checking) implemented in the shared _common.py module.
[EXTERNAL_DOWNLOADS]: The skill references several well-known scientific and governmental organizations (e.g., NIH, WHO, RAND) for methodological evidence. These references are documented neutrally and do not involve the download or execution of remote code.
[OBFUSCATION]: No obfuscated strings, Base64-encoded commands, zero-width characters, or homoglyph attacks were detected in any of the analyzed files.
[PRIVILEGE_ESCALATION]: Scripts use restrictive file permissions (0o600) for output and employ symlink protection (O_NOFOLLOW) to prevent unauthorized file access or escalation attacks.

scientific-brainstorming