The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references the installation of components via apm and npx @mizchi/waxa. These tools are provided by the skill author for the purpose of skill evaluation.
[COMMAND_EXECUTION]: Instructions involve running CLI commands such as waxa init, waxa iterate, and waxa audit to manage the evaluation lifecycle of agent skills.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external scenario data from tasks/*.yaml files and evaluate the resulting outputs. It defines ingestion points in the evals/ directory and utilizes a restricted execution environment for the agent being tested to mitigate risks associated with untrusted prompt content.
[DYNAMIC_EXECUTION]: Implements a multi-stage evaluation loop where code snippets (JS expressions) and LLM-based graders are dynamically applied to analyze the performance and safety of other skills.

waxa-eval