The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/fetch_zhihu_batch.py uses subprocess.run to re-execute itself with the --retry-failed flag. This is a controlled mechanism for automatic error recovery and does not involve executing arbitrary or untrusted command strings.
[EXTERNAL_DOWNLOADS]: The skill downloads article content and images from Zhihu domains (zhihu.com, zhuanlan.zhihu.com). These network operations are intrinsic to the skill's stated purpose of web scraping.
[SAFE]: User session data (cookies) is managed through a local workspace directory (~/.openclaw/workspace/zhihu_cookies.json). The skill does not hardcode secrets or attempt to exfiltrate credentials to external servers.
[SAFE]: Indirect Prompt Injection Surface: The skill ingests untrusted HTML content from Zhihu and converts it to Markdown.
Ingestion points: scripts/fetch_zhihu_batch.py and scripts/fetch_zhihu_api.py (web scraping).
Boundary markers: Absent.
Capability inventory: Bash (via controlled subprocess), Write (file management).
Sanitization: Implements an html_to_markdown function using regex and HTML unescaping to clean the input data. The risk is considered low as the data is primarily meant for local archival in Obsidian.

zhihu-fetcher