design-consultation
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSDYNAMIC_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the web via search results and browser snapshots for design research. It mitigates this risk with explicit instructions to treat all external content strictly as data, never as instructions, and includes boundary markers in its processing logic.
- [EXTERNAL_DOWNLOADS]: The skill references well-known and trusted services, including
aside.comfor browser automation and Google Fonts/Fontshare for typography verification. It does not perform unattended installations; it directs the user to official sources if tools are missing. - [DYNAMIC_EXECUTION]: The skill generates JavaScript snippets to drive the
asidebrowser automation and executes TypeScript-based utility scripts usingbun. These are well-scoped to the skill's primary function of visual research and token extraction. - [COMMAND_EXECUTION]: The skill utilizes a suite of local binaries provided by the vendor (
garrytan) under thegstackplatform. These tools manage session lifecycle, telemetry, and project-specific state without requesting elevated privileges. - [DATA_EXFILTRATION]: The skill communicates with external agents (e.g., Codex) for design reviews and uses a browser for research. It implements mandatory sanitization of queries (stripping IPs, secrets, and private hostnames) and uses an egress control wrapper (
gstack-egress-run) to manage network exposure.
Audit Metadata