The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to write a document containing conversation details to the OS temporary directory ($TMPDIR, /tmp, or %TEMP%). While data stored in temporary directories is accessible to other users on some systems, the skill proactively mitigates data exposure risks by requiring the agent to 'Redact secrets' (API keys, tokens, passwords) and replace them with placeholders before writing the file.
[PROMPT_INJECTION]: The skill creates a handoff document from conversation history that is intended to seed a new agent session. This creates a surface for indirect prompt injection where malicious instructions from previous inputs could be passed to the successor agent.
Ingestion points: The skill ingests the entire conversation history and any user-supplied arguments to build the document.
Boundary markers: The skill does not explicitly define boundary markers or 'ignore' instructions for the content of the conversation being summarized.
Capability inventory: The skill includes an optional instruction to 'launch the successor,' which involves starting a new background agent process seeded with the generated document.
Sanitization: While it lacks instruction-specific sanitization, it explicitly mandates the redaction of credentials and personal data, reducing the risk of sensitive data exfiltration through injected instructions.

handoff