Scored across 32 tools
Several tools overlap heavily: text, text_main, text_clean, find_text, and query_text all retrieve text in similar ways; extract_table/table_to_json is a pure alias; discover/route_discover and click/activate also have unclear boundaries. The descriptions are detailed, but an agent could easily select the wrong tool among the text and extraction families.
All names are readable snake_case, but conventions are mixed: single-word verbs/nouns (body, text, query, click), noun-first compounds (cookies_set, network_stores_clear), and special forms (table_to_json) coexist. The naming is predictable within families but not consistently verb_noun across the whole set.
32 tools exceeds the heavy threshold, and several entries are redundant or near-duplicates, such as table_to_json being an explicit alias for extract_table and the overlapping text and extraction tools. The broad scope explains some of the count, but the set would be stronger with fewer, more distinct tools.
The core lifecycle is well covered: navigate, query DOM, extract structured data, interact via click/type/submit, inspect network captures, and manage cookies. Minor gaps exist, such as no wait-for-selector, iframe handling, or navigation history, but agents can work around these with eval, settle, and the existing extraction tools.
