giil

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill provides an installation command in SKILL.md and references/TROUBLESHOOTING.md that downloads a shell script (install.sh) from the author's GitHub repository and pipes it to bash. This is the primary method provided for setting up the tool.
  • [EXTERNAL_DOWNLOADS]: The skill fetches code and large binary dependencies from external sources.
  • Downloads the giil installer from the author's repository: https://raw.githubusercontent.com/Dicklesworthstone/giil/main/install.sh.
  • Upon its first execution, the tool downloads the Playwright Chromium browser binary (approximately 200MB) to ~/.cache/giil/ms-playwright/.
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute the giil CLI tool. This tool performs network requests to cloud providers and interacts with the file system to save images or output base64 data to stdout.
  • [DYNAMIC_EXECUTION]: The tool creates and runs a Node.js script dynamically.
  • According to references/COMMANDS.md, the tool generates a script named extractor.mjs in the ~/.cache/giil/ directory to manage the extraction logic through Playwright.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes data from untrusted external URLs (iCloud, Dropbox, and Google Photos links), which creates a potential surface for indirect injection if malicious metadata or content is embedded in the shared pages.
  • Ingestion points: External cloud share URLs are provided as arguments to the giil tool in SKILL.md.
  • Boundary markers: There are no specific prompt delimiters defined to isolate the downloaded image data or metadata from the agent's instructions.
  • Capability inventory: The tool has network access to fetch data from cloud services and file system access to write the resulting image files.
  • Sanitization: The skill documents a 'Verification Pipeline' in references/CAPTURE-STRATEGY.md that validates Content-Type headers, checks magic byte signatures (e.g., FF D8 FF for JPEG), and detects HTML error pages to verify that the retrieved content is a valid image binary.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 09:51 AM