The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions guide the agent to run a local Python script (scripts/claude_link.py) and use the pbcopy tool on macOS to copy the generated session link.
[DATA_EXPOSURE]: The skill reads local session metadata from the Claude configuration directory (~/.claude/projects). This is restricted to retrieving the necessary identifiers to construct a session link.
[INDIRECT_PROMPT_INJECTION]: The skill processes local log files which could contain untrusted data.
Ingestion points: The script scripts/claude_link.py reads .jsonl files from the projects subdirectory of the Claude configuration path.
Boundary markers: The script utilizes structured JSON parsing to isolate the bridge-session type and its associated bridgeSessionId field.
Capability inventory: The logic is limited to reading file metadata and contents, then outputting a string. No network or file-write capabilities are present in the scripts.
Sanitization: The session_id input is strictly validated as a UUID, preventing path traversal attacks. Additionally, the extracted bridgeSessionId is validated against a specific regular expression pattern before being used in the final URL.

copy-link