The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the robotrock npm package, which is the official SDK for the service. It also integrates with several well-known and reputable packages including ai (Vercel AI SDK), workflow (Vercel Workflow), @trigger.dev/sdk, and @opentelemetry/api to provide durable wait functionality.
[COMMAND_EXECUTION]: The documentation includes instructions for users to initialize projects using CLI tools such as npx eve init. This is a standard setup pattern for the Eve agent framework and does not involve arbitrary or hidden command execution.
[DATA_EXFILTRATION]: The skill handles sensitive communication using standard environment variables for API keys and webhook secrets. It provides a verifyRobotRockWebhook helper to ensure the integrity and authenticity of inbound data from the RobotRock service. Communication with the RobotRock API is the primary and expected function of the skill.
[PROMPT_INJECTION]: The instructions contain guidelines for agent behavior, such as enforcing human-in-the-loop gates for high-value operations (e.g., refunds over $10k). The static analysis flag regarding instruction concealment in references/agent-capabilities.md is a false positive; the text describes UI rendering preferences (using OpenUI Lang cards for primary results while keeping logs in an activity trail) rather than an attempt to hide actions from the user.

robotrock