The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill dynamically locates and executes an external shell script (clone-conversation.sh) from the user's home directory. Because the script's contents are not part of the skill, it represents an unverifiable dependency that could perform arbitrary actions on the system.
[DATA_EXFILTRATION] (HIGH): The skill reads ~/.claude/history.jsonl, which contains private conversation history, session IDs, and project paths. Accessing these files exposes internal application state and user activity to the AI agent.
[INDIRECT_PROMPT_INJECTION] (MEDIUM): The skill processes untrusted data from the history log and passes it as arguments to a shell command without sanitization. If a project name or session ID contains shell metacharacters, it could trigger unexpected command execution.
Ingestion points: ~/.claude/history.jsonl (File access via tail and jq)
Boundary markers: None. Data is interpolated directly into a shell command string.
Capability inventory: Filesystem read (tail), search (find), and execution of local shell scripts.
Sanitization: None. The skill relies on jq to extract strings but does not escape them for shell safety before execution.

clone