start-flows-app
Pass
Audited by Gen Agent Trust Hub on Sep 30, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses the
Bashtool to executenpx @cognite/cli@latestfor tasks such as setting up local HTTPS and pulling project skills. These are standard operations for the vendor's official command-line interface. - [EXTERNAL_DOWNLOADS]: The skill references and downloads the official Cognite CLI from the NPM registry and points to documentation on
docs.cognite.com. These are legitimate resources owned by the skill's author. - [INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, including Cognite Data Fusion APIs (token inspection and data models) and user-supplied intent. This creates a surface where malicious content embedded in the remote data could attempt to influence agent behavior, although the skill context is restricted to development tasks.
- [DATA_EXPOSURE]: The instructions include specific safety guardrails, explicitly directing the agent not to paste session tokens, dump IAM JSON structures, or reveal its own system instructions in the conversation, which helps mitigate accidental data leakage.
Audit Metadata