brainstorming
Pass
Audited by Gen Agent Trust Hub on Aug 18, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill utilizes a local Node.js server (
scripts/server.cjs) managed by shell scripts (scripts/start-server.sh,scripts/stop-server.sh) to host a visual interface for the user. This is a standard functional component used to render and interact with design artifacts in a browser during the brainstorming phase. - [PROMPT_INJECTION]: The skill implements an indirect prompt injection surface by reading browser-based interaction events into the agent's context. This allows user interactions in the companion app to influence the agent's behavior.
- Ingestion points: User interaction events (clicks, selections) are written to a state file (
$STATE_DIR/events) which the agent is instructed to read and merge with terminal input. - Boundary markers: The instructions do not specify explicit delimiters for the data read from the event log.
- Capability inventory: The agent has capabilities for file system access, shell command execution (via server scripts), and the ability to trigger downstream planning skills.
- Sanitization: Intermediary JSON serialization is used, but the textual content of the events is passed to the model without specific sanitization filters.
Audit Metadata