The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes several ingestion points for untrusted data from external servers, which could contain malicious instructions designed to influence the agent at runtime.
Ingestion points: The Resource class provides methods like fetchJson(), fetchText(), fetchArrayBuffer(), fetchBlob(), and fetchImage() to retrieve data from remote URLs.
Boundary markers: No specific delimiters or instructions for the agent to ignore embedded content within the fetched data are present in the provided utility documentation.
Capability inventory: The skill enables significant capabilities including network writes (Resource.post, Resource.put), Web Worker execution (TaskProcessor), and browser-level actions like Fullscreen.requestFullscreen.
Sanitization: The documentation does not specify sanitization or validation logic for the content retrieved via the Resource class.
[COMMAND_EXECUTION]: The skill provides documentation for the Resource class which is used to execute HTTP requests (GET, POST, PUT), effectively allowing the agent to perform network-based command execution.
Evidence: Examples in SKILL.md show the use of Resource.fetchJson, resource.post(), and resource.put() to interact with external APIs.
[DATA_EXFILTRATION]: The networking utilities provided can be used to transmit data to external servers, creating a potential path for exfiltration if the agent is instructed to send sensitive information.
Evidence: The documentation for Resource and DefaultProxy demonstrates how to construct requests with custom headers (like Authorization) and query parameters, and how to route requests through a proxy.

cesiumjs-core-utilities