The Agent Skills Directory

[COMMAND_EXECUTION]: The create_handoff.py script executes local git commands (such as git log, git status, and git diff) to automatically populate session metadata. These commands are executed as lists of arguments via subprocess.run to prevent shell injection and are confined to the current working directory.
[INDIRECT_PROMPT_INJECTION]: The core functionality involves creating documents meant to be read by future agent sessions, which creates a potential surface for indirect prompt injection if malicious instructions are placed in these files.
Ingestion points: The 'RESUME Workflow' in SKILL.md directs the agent to load and read the contents of Markdown files from the .claude/handoffs/ directory.
Boundary markers: The skill does not implement specific prompt-level delimiters to isolate handoff content, though it does include instructions for the agent to verify the context against the actual project state.
Capability inventory: The skill allows the agent to read and write files within the project directory and execute hardcoded git commands via helper scripts.
Sanitization: The validate_handoff.py script identifies and blocks handoffs containing unredacted secrets or incomplete placeholders, serving as a basic safety filter for the generated content.

session-handoff