Skip to main content

deep-search

Deep multi-round web research with parallel fetching. Triggers: deep search, research, 深度搜索, 调研, investigate, deep research.

설치로 이동

소스 정보

저장소
ZhangHanDong/octos
최근 소스 활동
2026년 4월 1일 18:36
감지된 SKILL.md 언어
영어
스타
1
포크
1

설치 방법

기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.

소스 파일 검토

설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.

파일 탐색기
4 개 파일

SKILL.md 표시 중

SKILL.md
소스 지침 · 읽기 전용 미리보기
name
deep-search
description
Deep multi-round web research with parallel fetching. Triggers: deep search, research, 深度搜索, 调研, investigate, deep research.
# Deep Search ## Overview The `deep_search` tool performs deep multi-round web research. It iteratively searches from multiple angles, fetches pages in parallel, chases the most-referenced external sources, and produces a structured research report. ## Usage Call `deep_search` with a query and optional depth. The tool will: 1. **Round 1**: Search the web using Perplexity Sonar (preferred) or other available engines 2. **Rounds 2+**: Generate follow-up queries from different angles (time-qualified, subtopic-focused, controversy/analysis) 3. **Parallel fetch**: Crawl all discovered URLs concurrently (8 connections) 4. **Reference chasing**: Extract outbound links from crawled pages, fetch the most-cited external sources 5. **Report**: Build a structured report with overview, source previews, and search query log 6. **Save**: Everything saved under `./research/<query-slug>/` ### Parameters - **query** (required, string): The research topic or question to investigate. - **depth** (optional, integer, default: 2): Research depth: - `1` = Quick: single search round + crawl (~1 min, up to 10 pages) - `2` = Standard: 3 search rounds + reference chasing (~3 min, up to 30 pages) - `3` = Thorough: 5 search rounds + aggressive link chasing (~5 min, up to 50 pages) - **max_results** (optional, integer, default: 8): Number of search results per round (1-10). - **search_engine** (optional, string): Preferred search engine. Options: `perplexity`, `duckduckgo`, `brave`, `you`. Defaults to auto-detection (prefers Perplexity). ### Example ```json { "query": "AI regulations worldwide 2026", "depth": 2 } ``` ### Output Returns a structured research report including: - Overview (initial search answer) - Source details with inline previews (first 2000 chars of each page) - List of all search queries used - Summary with page count and save location Use `read_file` on specific source files for full content when you need detailed synthesis. ### Saved Files Results are saved to `./research/<query-slug>/`: - `_report.md` -- structured research report - `_search_results.md` -- combined raw search results from all rounds - `01_<domain>.md` -- full page content from first source - `02_<domain>.md` -- full page content from second source - etc. ### Environment Variables - `PERPLEXITY_API_KEY` -- enables Perplexity Sonar (recommended, best for deep research) - `BRAVE_API_KEY` -- enables Brave Search - `YDC_API_KEY` -- enables You.com search Without API keys, DuckDuckGo HTML search is used as fallback.
GitHub에서 보기