Extract date from article title text or URL href. Tries these patterns in order: 1. ``X月X日`` in title text → 202

Extract a regex pattern from LLM output that may contain analysis text. Tries: 1. Content inside backtick code fences 2.

Select the *max_articles* newest articles from a blog listing page. Rule-only: tries several date formats found across different blog sites.

Run all three merge stages and return results.

Deduplicate by server:port:type; rename name collisions with _N suffix.

Helper to call the static method.

Extract subscription URLs matching *pattern*.

Send *prompt* through weighted routing. Returns empty when all fail.

Extract subscription links from an article page. Returns (links, pattern_saved).

Parse LLM JSON response. Strips markdown fences if present.

Resolve target string to a list of SiteConfig.

Run all (or a single) sites, respecting concurrency limit. Args: target: Optional site name. When set, only that site runs.

Extract scheme + host from a URL for resolving relative links.

Download a file with retry.

Build a router with mocked _try_provider returning canned responses. *providers* maps provider name to a list of return values (one per call).

Load config from YAML file. Missing llm section yields empty LLMConfig.

Return shared Clash base config dict. Used by both ``_merge_yaml`` and ``_build_provider``.

Group proxy names by region keywords. Returns dict of region_label -> [proxy_name]. Unmatched names go into "🌍 其他".

Generate a YAML comment header.

Extract subscription links from page markdown. Looks for downloadable subscription files (.txt/.yaml URLs), subscription service URLs

Crawl4AI 爬取页面，返回结构化内容.

Ask LLM to produce a regex matching the site's subscription links. Returns a regex string or None. Post-processes the response to ext

输出到 nodes/{site}.{ext}.

Build two default proxy groups: url-test + select.

Build a Clash Meta config using proxy-provider references.

Concatenate all .txt files, dedup by line hash.

Extract proxies from all .yaml files, dedup, rebuild groups.

Print a summary table of all site results.

Warn if task_routing references provider names not in the provider list.

Generate README.md with subscription table from config site list.

httpx 下载文件，加大 read timeout 避免大 yaml 超时.

Decode base64 v2ray 节点，按行 hash 去重.

Run the full site processing pipeline.

Persist config, preserving link_pattern and llm section.

Three-layer verification: syntax, recall, precision.

Generate and write README.md.

Both providers tried when first fails; second succeeds.

When every provider fails, return empty string (never raise).

Router with empty provider list returns empty string.

When the first provider succeeds, return its text immediately.

Statistical: 10,000 picks should approximate the weight ratio.

Ensure extract_links flows through ask().