The Agent Skills Directory

[COMMAND_EXECUTION]: Executes standard git commands (git status, git branch, git log) to verify the repository state against the handover documentation. This is a routine operation for a development tool.
[DATA_EXFILTRATION]: Reads local files (defaulting to docs/handover.md or a path specified in the HANDOVER_FILE environment variable) to restore session context. No network operations or external data transmissions are performed.
[PROMPT_INJECTION]: The skill processes a handover file which may contain instructions from a previous session. This presents a potential surface for indirect prompt injection if the file is tampered with by an external actor. However, the skill includes several mitigation strategies:
Ingestion points: Reads docs/handover.md or a path from HANDOVER_FILE environment variable.
Boundary markers: None explicitly defined for the file content, but the skill treats the file as a snapshot rather than authoritative instructions.
Capability inventory: Limited to reading files and checking git status; requires user confirmation before starting any tasks.
Sanitization: The skill instructs the agent to cross-reference file content with the actual git status and wait for user approval before beginning work.

takeover