The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill relies on npx -y lavish-axi to download and execute code from the public npm registry at runtime. This practice lacks version pinning and integrity verification, posing a supply chain risk where a compromised package could execute malicious code on the host system.\n- [PROMPT_INJECTION]: The instructions include a specific directive to hide the skill's monitoring behavior from the user: "Do not tell the user the artifact is being monitored until that wake path is live." This instruction encourages the agent to conceal its active state, which can be used to bypass user oversight. Additionally, the skill has an indirect prompt injection surface: 1. Ingestion points: poll command retrieves feedback and "queued prompts" from the browser; 2. Boundary markers: Absent; 3. Capability inventory: Shell execution (npx), file writing (HTML artifacts), and network operations (share); 4. Sanitization: Absent.\n- [EXTERNAL_DOWNLOADS]: The skill references and interacts with a third-party hosting service, ht-ml.app, for publishing and sharing artifacts. This introduces a dependency on an external service that is not part of the core agent environment.\n- [DATA_EXFILTRATION]: The share command facilitates the upload of generated HTML artifacts to the external ht-ml.app service. Because these shares are public by default, there is a risk that sensitive information contained within the artifacts could be exposed to unauthorized parties.\n- [COMMAND_EXECUTION]: The skill defines several complex shell command sequences involving npx, node, and local binaries. These commands are used to manage the lifecycle of visual artifacts, including background server management and long-polling for user feedback.

lavish