The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the local repository, specifically agent entrypoints, system prompts, and tool definitions, to generate a TestSuite YAML and a runner script. Malicious local files could attempt to influence the generated output or the behavior of the agent during analysis. • Ingestion points: Local repository content including agent source code, instructions, and tool definitions. • Boundary markers: The generated YAML includes header comments advising manual review of the generated cases. • Capability inventory: The skill writes files to the local system and provides functionality to execute generated scripts with user consent. • Sanitization: No specific code sanitization or escaping mechanisms are mentioned for the ingested code before inclusion in generated artifacts.
[EXTERNAL_DOWNLOADS]: The skill instructions include installing external dependencies such as the vendor's observability SDK and the 'python-dotenv' library via standard package managers. These downloads originate from official registries or the vendor's own infrastructure.
[DYNAMIC_EXECUTION]: The skill generates a functional Python runner script by populating a template with logic derived from the analyzed agent code, involving script generation and potential runtime execution if the user opts into the testing phase.
[COMMAND_EXECUTION]: The skill may execute the generated Python runner or other harnesses during the optional execution step, but this only occurs after receiving explicit permission from the user.

agento11y-test-starter