The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses bash scripts to perform file system operations such as directory creation (mkdir -p), path resolution, and metadata extraction using standard utilities like git, awk, and ls. It includes a validation function to ensure the storage directory resides within the project root.
[COMMAND_EXECUTION]: Utilizes 'Dynamic Context Injection' (!command syntax) in the skill definition to automatically gather git branch information, session IDs, and repository status at load time. These commands are restricted to benign metadata collection.
[DATA_EXFILTRATION]: Captures session state, task lists, and conversation history to create persistent handoff documents. This is the core functionality of the skill and is mitigated by a mandatory 'redaction pass' instruction that requires the agent to scrub secrets, API keys, and PII before writing data to disk or the user interface.
[COMMAND_EXECUTION]: Launches background sessions via the claude --bg command when requested. The implementation uses a shell heredoc with a quoted sentinel (<<'HANDOFF_RESUME_PROMPT_END') to safely handle the resume prompt, preventing command injection from the session-generated content.
[SAFE]: The skill provides clear instructions for the user to verify actions, such as a post-write enforcement checklist and the use of visible boundary markers (U+2500 rails) for copy-paste content, ensuring transparency in how data is handled.

handoff