Skills
Navigate
skills/builderio/agent-native/design-editor-repro/Gen Agent Trust Hub

design-editor-repro

Pass

Audited by Gen Agent Trust Hub on Oct 28, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes a local automation helper (osmouse) via execFileSync to simulate operating system-level mouse and keyboard input for testing canvas interactions.
  • [EXTERNAL_DOWNLOADS]: The scripts/clip.mjs utility downloads clip metadata, transcripts, and image frames from external URLs provided by the user to facilitate issue reproduction.
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests data such as transcripts from clips and design files (HTML) which are then processed by the agent.
  • Ingestion points: scripts/clip.mjs (fetches transcript and context) and scripts/copy-design.mjs (imports design content).
  • Boundary markers: None detected; content is read directly into the agent's operational context.
  • Capability inventory: Subprocess execution for mouse automation, Playwright browser control, local file system writes, and database read access.
  • Sanitization: External content is processed and written to local files without explicit sanitization filters.
  • [CREDENTIALS_UNSAFE]: The harness/harness-env.mjs file contains hardcoded default credentials intended for use with a local test database environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 28, 2026, 01:36 PM