The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in scripts/agent.py to invoke built-in Windows diagnostic tools like reg, schtasks, and wmic. These calls are implemented with static command lists and no shell execution, which is a safe way to perform the system auditing described in the skill's documentation.
[INDIRECT_PROMPT_INJECTION]: The script scripts/process.py processes external log data from JSON and CSV files. While this creates a data ingestion surface, the script only performs regex-based classification and writes findings to local files, lacking any capabilities (like network access or shell execution of data) that could be leveraged for an attack.
[DATA_EXFILTRATION]: Analysis of the Python scripts confirms there are no network-capable functions (e.g., requests, urllib, socket) or attempts to transfer local system data to external endpoints.
[PROMPT_INJECTION]: The documentation in SKILL.md and the instructions within the scripts do not contain any patterns intended to bypass AI safety guidelines or override agent behavior.

hunting-for-persistence-mechanisms-in-windows