The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes a suite of local Python scripts for manuscript scaffolding and data validation. These scripts are dependency-free, use only the Python standard library, and implement strict security bounds including file size limits (5MB), record caps (10,000), and nesting limits to prevent resource exhaustion attacks.
[DATA_EXFILTRATION]: The skill instructions and documentation consistently emphasize a strict confidentiality boundary, explicitly forbidding the transfer of unpublished manuscripts or sensitive data to external services. The bundled tools are designed to be network-free and do not access environment variables or .env files.
[DYNAMIC_EXECUTION]: No evidence of dynamic code execution (eval/exec) or unsafe deserialization (pickle/yaml) was found. The Python scripts use deterministic parsing for JSON and CSV with added protections against duplicate keys and non-finite numbers.
[PROMPT_INJECTION]: The skill instructions reinforce safety guidelines and do not contain patterns intended to bypass or override agent constraints. The focus is on scientific integrity, evidence binding, and human accountability.
[PRIVILEGE_ESCALATION]: The workspace generator (scaffold_manuscript.py) uses restrictive directory permissions (mode 0o700) and explicitly refuses to overwrite existing files, preventing unintended file system modifications.

scientific-writing