The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes and summarizes the current conversation history, which may contain untrusted data from external sources previously retrieved by the agent (e.g., code snippets, web content). This creates a surface where malicious instructions embedded in that history could potentially influence the resulting summary or the agent's behavior.
Ingestion points: The skill is instructed to summarize the "current conversation content" in SKILL.md.
Boundary markers: The skill does not provide specific instructions for using delimiters or boundary markers to isolate untrusted external content within the generated summary.
Capability inventory: The skill uses local filesystem write capabilities to save the handoff document.
Sanitization: The skill includes a specific rule to "remove API key, passwords, personal identity information" during document generation.
[COMMAND_EXECUTION]: The skill requires the agent to interact with the environment to determine system paths and write files.
The instructions direct the agent to resolve the platform's temporary directory using Python APIs (e.g., tempfile.gettempdir()) or shell environment variables (e.g., $TMPDIR or /tmp).
The agent is instructed to save the markdown document to the determined absolute path and inform the user.

handoff