writing-livekit-scenarios
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill directs the agent to use the
lkCLI (LiveKit's official tool) to generate simulation baselines. This includes running thelk agent simulatecommand, which analyzes the agent's source code and uploads it to LiveKit Cloud for processing. The skill mandates that the agent warn the user about this upload and obtain confirmation before proceeding.\n- [INDIRECT_PROMPT_INJECTION]: This skill establishes a surface for processing instructions provided via external data sources (scenario files), which steer the agent's behavior during simulations.\n - Ingestion points: Scenario files containing natural language instructions, goals, and personas, as detailed in
SKILL.mdandreferences/scenario-craft.md.\n - Boundary markers: The agent is instructed to detect the simulation context via job metadata to isolate simulation logic from production code paths, as described in
references/connecting-the-agent.md.\n - Capability inventory: The agent environment supports tool mocking, state seeding from per-scenario data, and execution of final-state grading callbacks, per
references/connecting-the-agent.md.\n - Sanitization: No specific sanitization or filtering logic is prescribed for the scenario file content before it is processed by the agent's LLM.
Audit Metadata