wechat-layout

Pass

Audited by Gen Agent Trust Hub on Nov 12, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DYNAMIC_EXECUTION]: The wrapper scripts dynamically load executable code from a path constructed at runtime.
  • Evidence: In scripts/doctor.mjs and scripts/wechat.mjs, the skill uses await import() on a URL constructed from the ARTICLE2MEDIA_ENGINE environment variable or a relative path ../../article2media/engine/scripts/. This allows the skill to execute code located outside its own directory.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted Markdown data which could contain embedded instructions intended to influence the AI's behavior during the layout or summary process.
  • Ingestion points: Processes article.md, article.zh.md, and rewrite.md (via SKILL.md and references/wechat-guide.md).
  • Boundary markers: No specific prompt delimiters or instructions to ignore embedded commands are present in the provided files.
  • Capability inventory: The skill uses node scripts via the Bash tool, which are capable of reading/writing to the filesystem and making network requests.
  • Sanitization: The skill implements a significant mitigation called "verbatim check" (逐块读回校验), which compares the output against the input block-by-block to ensure no unauthorized changes were made to the text content.
  • [EXTERNAL_DOWNLOADS]: The skill automatically fetches remote images to facilitate local embedding.
  • Evidence: According to references/wechat-guide.md, the layout process identifies network images in the source Markdown, downloads them to a local images/ directory, and converts them to base64 strings for the final HTML output.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 12, 2026, 04:49 PM