don't have the plugin yet? install it then click "run inline in claude" again.
+trigger phrases are concrete and natural-language focused (搜索一下XX, 帮我搜抖音XX), with clear examples showing intent patterns
+keyword extraction rules are explicit and algorithmic (remove intent markers, strip noise words, preserve core nouns), with before/after examples
+acknowledges anti-bot reality (headless mode blocked, web_search fallback mandatory) rather than pretending scraping always works
weaknesses
~procedure lacks numbered steps, error handling details, and output contract specifics - step 2 detection logic relies on string matching that may be fragile or miss edge cases
~edge case coverage is minimal: no handling for zero results, invalid keywords, rate limits, malformed urls, or web_search failures; only mentions anti-bot interception without recovery depth
~documentation is conversational rather than rigorous - no input schema, no explicit success/failure criteria, no timeout specs, and '务必执行' (must execute) is imperative rather than explaining why