The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill includes evaluation tasks requiring the agent to ingest and act upon code and configuration files, which represents a surface for indirect prompt injection attacks.
Ingestion points: Task files such as evals/email-arbitrary-choice/task.md and evals/csv-roundtrip/task.md contain embedded file content that the agent must process to complete the tasks.
Boundary markers: Content is delimited by text headers (e.g., =============== FILE: package.json ===============), which do not offer robust protection against malicious instructions embedded within the files.
Capability inventory: The skill instructions involve shell commands (npm, yarn, pnpm) and file writing, creating potential impact if the agent follows injected instructions.
Sanitization: The skill does not provide mechanisms for validating or sanitizing ingested code or file content before the agent acts on it.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing the fast-check library from the npm registry and includes links to its official GitHub repository and website.
Evidence: Commands like npm install fast-check and references to github.com/dubzzz/fast-check and fast-check.dev represent standard development practices involving a well-known library.

fast-check-testing