The Agent Skills Directory

[Command Execution] (MEDIUM): The skill defines a PostToolUse hook that executes $HOME/.claude/hooks/ralph-activity-log.sh whenever the Bash tool is used. This involves executing an external script located in the user's home directory which is not part of the skill's own package, presenting a risk of executing unverified code.
[Data Exposure & Exfiltration] (LOW): The skill accesses ~/.ralph-state/activity.log. While intended for state tracking, it establishes a pattern of reading from hidden directories in the user's home folder which could contain sensitive command history.
[Indirect Prompt Injection] (LOW): The skill ingests untrusted data from ~/.ralph-state/activity.log and fix_plan.md to make decisions about agent exit conditions.
Ingestion points: Read via grep and file system checks.
Boundary markers: None; the skill searches for raw strings like 'done' or 'complete'.
Capability inventory: Can execute Bash commands and recommend agent termination (<promise>DONE</promise>).
Sanitization: No sanitization is performed on the content of the logs or the fix plan before it influences the exit logic.

ralph-graceful-exit