The Agent Skills Directory

[DATA_EXFILTRATION]: Accesses sensitive application state and private data in the home directory. The skill specifically targets session histories, memory files, and configurations from multiple AI tools stored in paths such as ~/.claude/, ~/.codex/, and ~/.gemini/.
[COMMAND_EXECUTION]: Runs a local discovery script (scripts/scan_agent_state.py) that lists all dot-directories in the home folder using ls -d ~/.[a-z]*. This allows the agent to identify and inspect metadata for any installed application on the system that uses hidden configuration directories.
[DATA_EXFILTRATION]: Reads and consolidates conversation transcripts (e.g., history.jsonl, .aider.chat.history.md) from previous AI sessions. These files may contain sensitive project details, PII, or credentials inadvertently shared in prior interactions.
[COMMAND_EXECUTION]: Uses the shell to calculate file lengths (wc -l) and perform git log/diff operations across the project directory, which is part of its core logic but contributes to the broad command execution surface.

project-memory