The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted project code and configurations, which inherently involves processing data that could contain malicious instructions.
Ingestion points: Uses Read, Glob, and Grep to analyze mounted files, installed packages, and terminal input within the user project.
Boundary markers: Explicitly instructs the agent to treat these inputs as "untrusted flows" and never trust preview messages or arbitrary paths.
Capability inventory: Features Write and Edit tools used to generate threat models and implement bounded security controls based on the analysis.
Sanitization: Requires component-by-component path validation and normalization to prevent traversal, as well as strict schema validation for preview messages.
[EXTERNAL_DOWNLOADS]: The skill references technical documentation and API guides from well-known technology service providers.
Evidence: Links to official documentation on webcontainers.io and developer.stackblitz.com for configuring headers, API references, and security boundaries in SKILL.md and references/official-docs.md.

stackblitz-security-basics