The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands mkdir -p to create the necessary directory structure for handoff cards and cat to verify that the file contents were written correctly to the disk.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by generating content meant to be ingested and followed by subsequent AI sessions (via the /pickup command).
Ingestion points: Target AI sessions are instructed to read and process the markdown files stored in .claude/handoffs/ as authoritative context.
Boundary markers: Although the files follow a structured format (YAML frontmatter), they lack explicit security boundaries or warnings to the receiving agent to treat the content as untrusted user data.
Capability inventory: The skill allows writing files to arbitrary directory paths provided as the 'target project', which could lead to file creation in unauthorized locations if the agent's environment is not properly sandboxed.
Sanitization: The skill does not perform sanitization or filtering on the 'Context' or 'Task' fields, allowing potentially malicious instructions to be passed between sessions.

dropoff