The Agent Skills Directory

[DATA_EXPOSURE_EXFILTRATION]: The skill implements robust local-only processing. Scripts are designed to work without network access and explicitly forbid reading environment variables or credentials. File operations are protected by path resolution checks and size limits.
[REMOTE_CODE_EXECUTION]: All tools are implemented using only the Python standard library. There are no external dependencies and no dynamic execution of user-supplied code.
[COMMAND_EXECUTION]: The skill uses shell commands to run local Python scripts. Input validation in _common.py prevents common command injection vectors by rejecting symlinks and enforcing strict file path and suffix restrictions.
[INDIRECT_PROMPT_INJECTION]: While the skill processes user-supplied manuscripts and metadata (CSV/JSON), it uses strict schema validation and structural linting to mitigate the risk of malicious data influencing agent behavior. Reports are designed to emit identifiers and counts rather than echoing raw manuscript text.
[PRIVILEGE_ESCALATION]: File operations use atomic writes with restricted permissions (0o600) to ensure that generated reports are only accessible to the current user.

peer-review