figma-comment-fix-loop
Warn
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and acts upon untrusted stakeholder comments retrieved from the Figma platform.
- Ingestion points: The skill fetches raw comment data from the Figma REST API (
GET /v1/files/<fileKey>/comments) as described in Phase 0. - Boundary markers: There are no instructions providing delimiters or specific warnings to ignore instructions that might be embedded within the text of the stakeholder comments.
- Capability inventory: The skill utilizes powerful capabilities including
figma_executefor interacting with the Figma Plugin API, shell access for executingcurl,osascript, andscreencapture, and the ability to generate and run local server scripts. - Sanitization: No sanitization, filtering, or validation of the comment content is mentioned before the agent uses it to propose and execute visual fixes.
- [DYNAMIC_EXECUTION]: The skill explicitly instructs the agent to dynamically generate and execute a local HTTP server to facilitate image exports. In Phase 3, the agent is directed to "write a small one" (a server) that binds to ports in the 9223-9232 range, handles POST requests containing binary data, and configures specific CORS headers and IPv6 bindings.
- [COMMAND_EXECUTION]: The skill relies on shell command execution for its primary workflows. This includes using
curlto interact with the Figma API while handling sensitive environment variables, and utilizing macOS-specific utilities likescreencaptureto record the user's desktop canvas. - [PRIVILEGE_ESCALATION]: The skill uses
osascript(AppleScript) to programmatically activate the Figma application and manipulate its windows via theSystem Eventsprocess. The instructions note that this requires the user to grant "Accessibility" permissions to the agent's environment, which provides the skill with high-level control over the graphical user interface of other applications.
Audit Metadata