The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches content from external proxy services r.jina.ai and defuddle.md using curl as its primary method for URL-to-Markdown conversion.
[COMMAND_EXECUTION]: The skill invokes several system commands and scripts to process various URL types, including curl, python3, pdftotext, and marker_single.
[REMOTE_CODE_EXECUTION]: In scripts/fetch.sh, the skill uses npx --yes agent-fetch, which automatically downloads and executes the agent-fetch utility from the npm registry as a last-resort fallback.
[PROMPT_INJECTION]: The skill has an indirect prompt injection surface because it processes untrusted content from arbitrary URLs and presents it to the agent.
Ingestion points: External web content retrieved via fetch.sh, fetch_feishu.py, and fetch_weixin.py (file: SKILL.md).
Boundary markers: Absent. The instructions direct the agent to display the full Markdown content directly without explicit delimiters or warnings to ignore embedded instructions.
Capability inventory: Subprocess execution for data processing, network access for fetching URLs, and file system writes to the ~/Downloads directory (file: SKILL.md).
Sanitization: The skill performs basic validation for line count and common error pages but does not sanitize or escape the retrieved Markdown content before display.

qiaomu-markdown-proxy