The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a pattern where state and conversation history from one agent are directly interpolated into the system instructions of a receiving agent, creating an indirect prompt injection surface.
Ingestion points: The HandoffContext (containing key_facts, decisions_made, and conversation_summary) and ConversationContinuity (containing conversation history) in SKILL.md are used to populate future prompts.
Boundary markers: No boundary markers or delimiters are used in _build_handoff_prompt or build_continuity_prompt to separate untrusted context from the agent's core instructions.
Capability inventory: Receiving agents execute arbitrary tasks and potentially perform further handoffs or tool calls based on the injected context.
Sanitization: There is no evidence of sanitization or escaping for the strings before they are interpolated into the prompt templates.

agent-handoff-protocols