The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions include installing a2a-sdk for Python and @a2a-js/sdk for Node.js. These are functional dependencies for the protocol described and align with the vendor's specialized domain.
[PROMPT_INJECTION]: The skill implementation creates a surface for indirect prompt injection as it processes and acts upon external message content without demonstrating sanitization or boundary markers.
Ingestion points: Untrusted external data is ingested via context.get_user_message() in Python and context.getUserMessage() in Node.js within the SKILL.md code examples.
Boundary markers: The provided examples do not use delimiters or specific safety instructions to isolate external messages from the agent's internal reasoning.
Capability inventory: The logic includes capabilities for web searching and multi-agent orchestration, which are triggered based on the content of these external messages.
Sanitization: No content validation, escaping, or sanitization logic is present in the server execution handlers provided in the documentation.

a2a-protocol