The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the tdx command-line tool, which is the official CLI provided by Treasure Data. The commands documented (e.g., tdx agent test) are standard operations for agent lifecycle management and testing within the vendor's ecosystem.
[INDIRECT_PROMPT_INJECTION]: The skill defines a testing workflow that ingests data from test.yml files to simulate agent conversations. This presents a theoretical indirect prompt injection surface if an attacker can influence the contents of the test definitions.
Ingestion points: The user_input and criteria fields in test.yml are loaded and processed by the testing tool.
Boundary markers: The documentation does not describe specific delimiters or instructions used to separate test inputs from system instructions during execution.
Capability inventory: The skill facilitates the execution of agent conversations and evaluation of responses using the tdx CLI.
Sanitization: No explicit sanitization or validation mechanisms for the YAML content are described in the instructions.

foundry-agent-test