The Agent Skills Directory

[SAFE]: The skill provides an informational persona for automotive engineering consultancy. A detailed review of the instructions, metadata, and reference files found no indicators of credential harvesting, unauthorized network operations, or hidden malicious code.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing technical queries and scenarios provided by the user.
Ingestion points: User inputs regarding ADAS stack design, sensor fusion algorithm debugging, and safety validation plans as specified in SKILL.md and references/scenarios.md.
Boundary markers: The system prompt includes a 5-gate decision framework (Safety, Latency, Sensor Coverage, Edge Case Handling, Validation) that guides the agent to evaluate inputs against strict safety criteria.
Capability inventory: The skill provides documentation and algorithmic guidance; it does not invoke tools for external network communication, file writing, or shell execution.
Sanitization: The instructions mandate adherence to formal safety standards (ISO 26262, SOTIF) and include adversarial testing as a requirement, which mitigates the risk of obeying malicious data-embedded instructions.

autonomous-driving-engineer