The Agent Skills Directory

[PROMPT_INJECTION]: The skill generates summaries of untrusted user input from the current conversation to be processed by a subsequent agent session. This creates a vector for indirect prompt injection where instructions hidden in the primary conversation could be summarized and later interpreted as commands by the next agent.
Ingestion points: Current session history as described in SKILL.md.
Boundary markers: The skill uses a structured Markdown template but lacks explicit delimiters or warnings for the next agent to ignore embedded instructions.
Capability inventory: The skill utilizes file writing capabilities to save the handoff document.
Sanitization: Includes a manual instruction to redact API keys, passwords, and PII, which addresses data exposure but not behavioral injection.
[COMMAND_EXECUTION]: The instructions direct the agent to utilize system commands like mktemp or environment variables like $TMPDIR to identify storage locations. This is a standard and safe practice for managing temporary files but involves interacting with the shell environment.
[SAFE]: The skill explicitly instructs the agent to redact sensitive information such as API keys and tokens before generating the handoff file, following security best practices for session management.

handoff