web-query

name	web_query
version	1.0.0
description	Unified web discovery and reading entry point. Use when the user asks to find references, official docs, latest links, or read a public page from either a query or URL. If a page needs login state, prefer reusing cookie/session access before escalating to full browser handoff.
invocation	blue web_query input="OpenAI Responses API docs"
examples	["blue web_query input=\"OpenAI Responses API docs\"","blue web_query input=\"ZimaOS Blue release notes\" max_results=8"]
capability_tags	["search","web","docs"]
interaction_mode	stateless
card_support	none

Web Search Skill

Setup

No external dependencies required. Uses built-in web search capability.

---

Task Routing

User Intent	Action
Need relevant links/sources quickly and do not yet have the right URL	blue web_query input=...
Need official docs/reference pages	blue web_query with precise query terms
Already have a concrete public URL and only need page content	blue web_query input="https://..."
Already have a concrete URL that needs login state but only need readable content	Prefer authenticated read/fetch with Cookie, Authorization, or browser_target_id before using full browser automation
Need page interaction/login/JS rendering	Search first, then switch to browser

---

Command Usage

blue web_query input="ZimaOS Blue release notes"
blue web_query input="OpenAI Responses API function calling" max_results=8
blue web_query input="container sandbox security best practices" max_results=10
blue web_query input="https://platform.openai.com/docs/api-reference/responses"

Parameters:

• input (required; search query or public URL)
• max_results (optional, default 10, max 20)
• max_chars (optional, only for page reads)
• depth (optional, quick, standard, deep)

---

Error Handling

Error	Resolution
input is required	Provide a non-empty search query or URL
Page requires login or interaction	Reuse existing Cookie, Authorization, or browser_target_id first; if the site still needs a live session, switch to browser

---

Notes

• web_query is the canonical public web skill.
• Legacy web_search, web_fetch, and web_read names are compatibility aliases and should not be used as the primary route in new prompts or harness cases.
• If web_query reports login wall, challenge, or browser-required warnings, prefer an authenticated read/fetch path with existing Cookie, Authorization, or browser_target_id before escalating to browser.
• When the login state already exists in Chrome and the site needs a live authenticated browser, prefer browser with relay/local Chrome (Chrome replay) instead of starting from a fresh anonymous tab.
• Blue also uses an internal recent multi-site retrieval profile inside the web layer for Research requests such as what are people saying in the last 30 days / 最近30天大家怎么说.
• That recent profile is native to web_query internals, not a separate public mode or command.
• The current native source mix includes public web, Reddit, Hacker News, GitHub, Polymarket, X, TikTok, Instagram, and Bluesky. Known YouTube URLs can be added as enrichment when they are already present in the request.
• For X, TikTok, Instagram, and Bluesky, Blue currently relies on public-page discovery plus browser-assisted reading rather than a separate public platform command.
• If the user only wants a few links or one readable page, keep using ordinary web_query input=... calls rather than trying to surface the internal recent profile directly.