The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and decision flowchart guide users to execute evaluation scripts locally using the bun runtime environment (e.g., bun path/to/eval.ts). This execution is required to run the production logic and capture performance metrics.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted input via the TEST_CASES configuration in the provided implementation patterns.
Ingestion points: Untrusted data enters the agent context through the TEST_CASES array in the eval.ts file.
Boundary markers: There are no explicit boundary markers or "ignore" instructions within the code snippets, as the skill relies on the production logic being tested to handle inputs.
Capability inventory: The skill includes the capability to write to the local file system (fs.writeFile, fs.writeFileSync) and relies on terminal execution for operation.
Sanitization: No input sanitization or validation is implemented within the provided templates, as the primary purpose is to observe raw model behavior.
[EXTERNAL_DOWNLOADS]: The skill metadata includes a source link to https://il.ly/skills/eval, which is a resource provided by the author 'illyism'. This link serves as a reference for the skill's origin.

eval