The Agent Skills Directory

Local Command Execution: The skill utilizes repository-local tools to perform validation, such as executing dotnet run for a validator project, running a Python script for quality checks, and executing shell scripts for evaluation runs. These commands are constrained to the local environment and are intended for verifying the schema and quality of generated test files.
Indirect Prompt Injection Surface: The skill processes external skill content as input to generate rubric items and test stimuli. This creates an ingestion surface where instructions within the processed skill could theoretically influence the generated output.
Ingestion points: The 'Skill content' input provided during the scaffolding process (referenced in SKILL.md).
Boundary markers: The skill does not explicitly define delimiters for the input content, but it provides strict instructions to paraphrase and generalize input into natural language stimuli.
Capability inventory: The skill possesses the capability to write files (eval.yaml and fixtures) and execute local validation commands.
Sanitization: The instructions explicitly require the agent to avoid 'overfitting' by rewriting technical skill terminology into natural developer requests and outcome-based rubric items, which serves as a functional layer of sanitization.

create-skill-test