菜单
Search the web via the Kagi Search API. Use when the user asks to search the web, look something up online, find current information, or needs external references/URLs.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Extract embedded text and rendered page images from PDFs. Use when the user asks to read, inspect, parse, convert, or extract contents from PDF files without OCR.
Answer questions about the Datadog org structure using Workday data in Snowflake. Use when the user asks about org charts, team structure, who reports to someone, headcount, titles, or any people/org question.
Upload files with the chonk command and return shareable chonk URLs. Use when the user asks to upload files to chonk or share files via chonk.
Interact with Atlassian Jira and Confluence via MCP. Use when the user asks about Jira issues, Confluence pages, sprints, boards, or any Atlassian task.
Format commit messages using the Conventional Commits v1.0.0 specification. Use when writing commit descriptions, generating commit messages, or when the user asks about commit message format.
Fetch a URL and convert it to clean markdown using trafilatura, writing the result to a file to avoid polluting the context window with HTML. Use when the user asks to fetch, read, scrape, summarize, or extract content from a web page.
| name | web-search |
| description | Search the web via the Kagi Search API. Use when the user asks to search the web, look something up online, find current information, or needs external references/URLs. |
Use the bundled search.sh script to query the Kagi Search API. The script reads the API token from the KAGI_TOKEN environment variable.
When this skill is invoked, run a search immediately with the user's query — don't explain the API or ask clarifying questions first. The user wants results.
./search.sh "<query>" # default: 10 results
./search.sh "<query>" 20 # custom result limit
./search.sh --json "<query>" # raw JSON response
The script prints one result per block: title, URL, optional published date, and snippet. Related searches (Kagi object type t=1) are appended at the end when present.
./search.sh "steve jobs biography"
./search.sh "jujutsu vcs tutorial" 5
./search.sh --json "datadog dash 2024 announcements" | jq '.data[] | select(.t==0) | .url'
Search results only include titles, URLs, and snippets. When the user asks for details that aren't in the snippet (summarize, quote, extract, etc.), use the web-fetch skill on the relevant result URL(s) to get the full page as markdown without flooding the context with HTML.
KAGI_TOKEN in the environment (Kagi Search API token, from https://kagi.com/settings/api).