The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/build_article.py uses subprocess.run to orchestrate an internal build pipeline by executing other local scripts (lint_article.py, render_theme.py, check_html.py). This is a core functional element of the skill used to process and validate article content.
[INDIRECT_PROMPT_INJECTION]: The skill has a potential attack surface as it ingests untrusted data from the web (via benchmarking and search tasks) and user-provided Markdown files. This content is processed and then used to generate WeChat drafts or messages.
Ingestion points: Markdown article files, web search results for benchmarking, and user-provided sample texts.
Boundary markers: The skill uses :::html code blocks for custom components and provides explicit 'interaction cards' to gate significant actions.
Capability inventory: The skill can write to the local file system (via rendering scripts), create WeChat drafts (weixin_oa_create_draft), and send service messages (weixin_oa_send_service_message).
Sanitization: The skill includes scripts/check_html.py which strictly validates HTML content, blocking dangerous tags like <script>, <iframe>, and event handlers, and scripts/lint_article.py which checks for 'AI-generated' patterns and potential advertising law violations.
[EXTERNAL_DOWNLOADS]: The skill instructions mention installing the pillow Python library for image processing and a related skill yungroai-skills (from the same vendor) to facilitate pre-flight checks. Documentation neutrally notes these dependencies are required for the intended functionality.

weixin-official-account-operator