skills/garrytan/gstack/design-shotgun/Gen Agent Trust Hub

design-shotgun

Warn

Audited by Gen Agent Trust Hub on Nov 16, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill uses the aside tool to generate and execute browser automation scripts at runtime. These scripts are constructed by interpolating a URL into a JavaScript template, which is then executed to capture screenshots of local or remote websites.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from multiple external sources, including user feedback stored in feedback.json (produced by a local HTTP server) and repository documents like DESIGN.md and PRODUCT.md. These inputs are used to influence the AI's design generation logic, presenting a potential surface for instructions embedded in data to affect agent behavior.
  • [COMMAND_EXECUTION]: The skill relies on the execution of several local binaries and scripts within the author's toolkit (e.g., gstack-skill-start, gstack-paths, and the design binary) to manage project state, generate visual variants, and serve comparison boards.
  • [DATA_EXFILTRATION]: Session metadata, telemetry, and user preferences are recorded and passed to local logging utilities such as gstack-question-log and gstack-skill-end during the design exploration workflow.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes a local design binary and a browser automation suite that are external to the primary instruction files, forming a dependency on the pre-installed gstack environment.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 16, 2026, 08:17 PM