The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads and executes an installation script from Tailscale (tailscale.com) to establish network connectivity between servers.
[REMOTE_CODE_EXECUTION]: Instructions include a verification step that pipes the response from a local network endpoint directly into the Python interpreter for JSON formatting using the json.tool module.
[COMMAND_EXECUTION]: The skill configures the agent to use the exec tool to run a local Node.js script (a2a-send.mjs) for sending messages and files to other agents.
[DATA_EXFILTRATION]: The communication script (a2a-send.mjs) possesses the capability to read local files from the filesystem via the --file-path argument and transmit their contents to a remote URL provided in the command line.
[PROMPT_INJECTION]: The skill establishes an indirect prompt injection surface by allowing the agent to ingest and process data from external peer agents.
Ingestion points: The agent reads and potentially acts upon the stdout responses from a2a-send.mjs as documented in SKILL.md and the tools-md-template.md reference.
Boundary markers: No delimiters or "ignore instructions" markers are defined in the provided TOOLS.md template for peer responses.
Capability inventory: The agent utilizes the exec tool to interact with the communication script.
Sanitization: There is no evidence of sanitization or validation performed on the messages received from external agents before they are processed by the local agent.

a2a-setup