The Agent Skills Directory

Command Execution: The skill provides instructions for setting up the developer environment using yarn install and cargo build, and for executing tests via yarn test:e2e. These are standard, necessary actions for the intended use case of testing the Relay framework locally.
Indirect Prompt Injection Surface: The testing harness processes markdown files to extract and execute code blocks, which represents a potential surface for indirect prompt injection if the fixtures are sourced from untrusted locations.
Ingestion points: Markdown fixture files located in packages/relay-e2e-test/fixtures/.
Boundary markers: The harness identifies code using markdown fences and title metadata, but the agent lacks explicit instructions to disregard surrounding natural language.
Capability inventory: The agent has access to Bash and file system tools, and the workflow involves executing compiled output.
Sanitization: The skill does not describe specific sanitization for markdown fixture content prior to execution.
Dynamic Execution: The skill extracts code from markdown fixtures at runtime to compile and run them using React and Relay. This is the core functionality of the e2e test harness and is appropriate for the skill's stated purpose of providing a flexible testing environment.

relay-e2e-test