Menu
Search the web or summarize a URL via Kagi (no API credits used; session-token auth).
Instalar com Codex ou Claude Copie este prompt, cole no Codex, Claude ou outro assistente e deixe que ele revise a página da skill e instale para você.
Baseado na classificação ocupacional SOC
| name | kagi |
| description | Search the web or summarize a URL via Kagi (no API credits used; session-token auth). |
The kagi CLI has two verbs: search (discover) and summarize (enrich a
known URL). kagi-search and kagi-summarize are also available as
standalone binaries (shortcuts for the corresponding verbs).
Use for discovery: queries the web and returns Kagi's Quick Answer summary plus optional result links.
# Quick Answer only (default)
kagi search "what is the capital of France"
# Include result links (default: 3)
kagi search -l "search query"
# More links
kagi search -l -n 10 "search query"
# Plain text (strip markdown markup)
kagi search --text "search query"
# JSON for scripts (note: .results is empty unless you pass -l)
kagi search -j -l "search query" | jq -r '.quick_answer.references[].url'
# Shortcut form:
kagi-search "search query"
Reading vs scripting. The default (no -j) output is a synthesized,
cited answer — Kagi's Quick Answer plus a References list of source URLs. Use
bare kagi search whenever you're reading the result: it's already clean
markdown and uses fewer tokens than JSON. Don't reach for -j just because
you're an agent — JSON only pays off when a script parses fields. For that
case use -j (add -l -n 10 to populate .results with raw candidate
links); the JSON shape is {results, quick_answer: {markdown, references: [{title, url, contribution}]}}.
Use when you already have a URL and want a structured summary — works well as a fetch replacement for pages where you don't need the HTML, just the gist.
# Markdown summary (default)
kagi summarize https://example.com/article
# Bullet-list takeaways
kagi summarize --takeaway https://example.com/article
# Plain text (strips markdown + HTML)
kagi summarize --text https://example.com/article
# Raw JSON
kagi summarize -j https://example.com/article
# Target language
kagi summarize --language ES https://example.com/article
# Alias form:
kagi-summarize https://example.com/article
Both verbs share the same output flags: -j/--json, --text, --markdown
(default).
Verified web fetch/search — use INSTEAD OF WebFetch (and curl) to read a public web page, search the web, extract clean article text, or crawl/map a site. Returns trustworthy main content or an honest abstention (never a login page, bot wall, or nav shell passed off as the article); the exit code encodes the outcome. `search` blends the user's Kagi/Perplexity subscriptions and can verify each result. NOT for logged-in/authenticated pages (use agent-browser) or anonymous bot-walled sites (use cloak-browser).
General browser automation — drive Chrome/Chromium over CDP (navigate, click, type, eval, screenshot, snapshot) via the agent-browser CLI. Use for normal or authenticated/logged-in browsing. For anonymous bot-walled / Cloudflare targets (i.e. "stealth" browsing) use the cloak-browser skill instead; for logged-in sites, inject real cookies (see below).
Stealth browsing — an agentic, anti-detect browser. Drive through anti-bot / bot-detection walls (Cloudflare Turnstile, fingerprinting, sannysoft-class checks) via the cloak-browser CLI (agent-browser + CloakBrowser's source-patched stealth Chromium). Use when the user asks for "stealth" browsing, or when a fetch or plain agent-browser is blocked, challenged, or bot-walled on an anonymous (not logged-in) site. Not for authenticated sessions.
Generate speech audio from text with Google's Gemini TTS. Use when the user asks to synthesize speech, create a voiceover or voicemail greeting, or narrate text.
Query and update Atlassian Jira, Confluence, and Compass via Atlassian's official Remote MCP. Use when the user asks about Jira issues, JQL searches, sprints, Confluence pages, or any Atlassian product operation.
List and search recent Claude Code sessions for resumption. Use when the user asks about past sessions, wants to find a session by topic or keyword, or needs help resuming a previous conversation.