The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes a bundled Python utility (scripts/check_article.py) to validate the generated article's length, formatting, and screenshot density. Analysis of the script confirms it is a benign text-processing tool that uses regular expressions for validation and does not perform network operations or access sensitive system files.
[EXTERNAL_DOWNLOADS]: The skill uses web access capabilities to research topics, verify technical claims, and fetch content from official documentation or source code repositories. This is a primary intended function of the research skill.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) as it processes untrusted data from user-supplied links and materials.
Ingestion points: User-provided links (including Playwright-based reading of WeChat articles), attachments, and local files like sucai.md (specified in SKILL.md).
Boundary markers: The skill implements a 'Material Mapping Table' and specific structured placeholder formats for screenshots (references/screenshot-guide.md) to delimit different types of content.
Capability inventory: The agent is granted capabilities for local file writing to the docs/ directory, web access for research, and execution of the internal Python validation script.
Sanitization: The skill relies on instructional logic rather than programmatic sanitization, directing the agent to prioritize official source code and documentation over material provided in the input links.

codex-article