creating-cloud-persona

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The reference agent examples provided in the skill perform network requests to well-known and whitelisted services to fulfill their intended purposes. These include the Brave Search API for research, the Algolia-hosted Hacker News API for monitoring, the Spotify API for music release checks, and the official npm registry for package version tracking.
  • [DYNAMIC_EXECUTION]: The internal logic of the platform's deployment tools, provided as reference source code, uses dynamic import() to evaluate agent metadata during the authoring process. This mechanism is restricted to analyzing local agent scripts during the deployment phase and is a core component of the platform's extensibility model.
  • [INDIRECT_PROMPT_INJECTION]: The skill provides patterns for agents designed to process untrusted data from communication platforms like Slack and GitHub. The documentation explicitly advises on defensive programming practices, such as the use of boundary markers and system prompt instructions that prioritize grounded data over external content to mitigate the risk of indirect injection.
  • [COMMAND_EXECUTION]: Multiple reference agents utilize the ctx.sandbox.exec capability to run development-oriented shell commands (e.g., git, gh, find, rg). These operations are used for legitimate repository maintenance and file navigation within an isolated sandbox environment provided by the platform.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 08:58 AM
Security Audit — agent-trust-hub — creating-cloud-persona