The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and process data from external sources, which constitutes a potential attack surface.
Ingestion points: The agent is directed in SKILL.md to fetch live documentation from https://a2a-protocol.org/latest/specification/ and https://www.jsonrpc.org/specification, as well as to process results from a web search targeting GitHub.
Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore potential instructions embedded within the external documentation.
Capability inventory: The skill is granted access to high-privilege tools such as Bash, Write, and Edit, which could be misused if the external data contains actionable malicious prompts.
Sanitization: The instructions lack any requirement for the agent to sanitize or validate the integrity of the content retrieved from the remote URLs before acting upon it.

a2a-jsonrpc-transport