The Agent Skills Directory

[General] (SAFE): The skill is entirely declarative and instructional. It contains markdown documentation intended to guide an AI agent in providing security advice.
[Indirect Prompt Injection] (SAFE): The skill directs the agent to read internal project files (e.g., ARCHITECTURE.md) to gain context.
Ingestion points: SKILL.md (lines 17-20) identifies project context files.
Boundary markers: None explicitly defined, but the skill lacks executable capabilities.
Capability inventory: No file writes, subprocess execution, or network requests are present in any script or instruction.
Sanitization: Not applicable as the skill does not process external untrusted data via tools.
[Data Exposure & Exfiltration] (SAFE): No credentials or sensitive data are hardcoded. References to environment variables (e.g., ALLOWED_ORIGINS, AWS Secrets Manager) follow security best practices.
[External Downloads] (SAFE): The skill contains links to reputable documentation (OWASP, NestJS, Next.js, React) and does not perform any automated downloads or script executions.

security-expert