Search skills/

Deep Research Skill Claude skill

A single research skill with an 8-phase pipeline, parallel search, source scoring, claim and citation checks, and report output in Markdown, HTML and PDF.

199-biotechnologies/claude-deep-research-skill 1 skill Research
Our verdict Recommended with caveats

Real scripts and tests behind the pipeline, but the superiority claim is unsupported, quality depends on search keys you add, and the license is stated only in the README.

4/5 Depth expertise shipped Note Security Review before installing 7mo Maintenance Updated 7 mo ago 1.2K Adoption stars ? License no clear license

What Deep Research Skill does

This repo is one skill, deep-research, that turns a research question into a cited report. You choose a mode by phrasing: quick (3 phases, a few minutes), standard (6), deep (8) or ultradeep (8 plus, up to about 45 minutes). The pipeline scopes the question, plans, retrieves with 5 to 10 parallel searches and a few sub-agents, triangulates sources, refines an outline, synthesizes, critiques (it can loop back to retrieval) and packages the result. Deep modes add multi-persona red-teaming. Citations are written to a sources.json file on disk so they survive context compaction, and reports over about 18,000 words continue through recursive agent spawning.

The skill folder is the substantial part: seven reference files, ten Python scripts and a tests folder. Scripts score source credibility, track evidence and citations, extract claims, validate report structure (nine checks), verify citations by checking DOIs and URLs over HTTP, and convert Markdown to HTML or PDF through WeasyPrint. Search runs first through the author's search-cli (Homebrew tap, with your Brave, Serper, Exa, Jina or Firecrawl keys), falling back to built-in web search. Output lands in ~/Documents. The README claims it outperforms OpenAI, Gemini and Claude Desktop, but the repo includes no benchmark to back that up.

Good for

  • Long, cited research reports where claims should be traced to sources
  • Users willing to supply a search API key for better coverage
  • Anyone wanting a pipeline with scripted citation and structure checks

Not for

  • Quick lookups that a single web search answers
  • Users who need verified performance numbers; the repo gives none

What's inside depth 4/5

Seven reference files, ten scripts, templates, schemas and a tests folder in one skill, though there are no published evals or CI.

Skills
1
Reference docs
7
Scripts
14
Tests / evals
7
Templates
2
Avg SKILL.md
4.2 KB

Counted from the repository file tree. How we score depth.

Dependencies and costs

  • Python 3 (WeasyPrint for PDF output)
  • Optional search-cli via Homebrew with at least one search provider key (Brave, Serper, Exa, Jina or Firecrawl)
  • Optional Exa MCP server

Security review

Review before installing

A research-report workflow made of Markdown phase guides plus local Python helpers for citations, evidence storage and report validation. The only network code checks bibliography DOIs and URLs, and research state is saved to a folder under ~/.claude. Its default search path is a separate Homebrew-installed CLI that needs your own search-provider API keys, and that tool's code was not reviewed here.

Worth knowing

  • Makes network requestsSends HTTP requests to doi.org and to every URL found in a report's bibliography (HEAD request) to check that citations resolve; no keys are sent.scripts/verify_citations.py:129
  • Relies on a CLI or package we could not reviewTells the agent to run the third-party search CLI (search-cli, installed from the author's Homebrew tap) as the primary search tool, driven through Bash.reference/methodology.md:74

What we checked

The repository (32 files) on Oct 29, 2026. Not a full review: we read 2 of 14 code files; it relies on code outside the repository: search-cli (Homebrew formula from 199-biotechnologies/tap; needs Brave/Serper/Exa/Jina/Firecrawl keys); optional pip package weasyprint for PDF output; optional Exa MCP server.

NVIDIA SkillSpector static scan: 12 pattern hits (Dangerous Code Execution 4, Output Handling 2, Excessive Agency 2, System Prompt Leakage 1, Anti-Refusal 1, MCP Least Privilege 1). The 12 SkillSpector hits are a function returning phase prompt text, the phrase 'without warnings' in a report template, a missing allowed-tools declaration, and subprocess calls that exist only in unit tests. No obfuscation, credential reads or hooks found; a pattern scan of all scripts and tests (subprocess, eval, exec, environ, sockets, file deletion) found nothing outside tests.

A review, not a certification. Skills change after we check them — read the files before installing.

License

The README says MIT, but GitHub detected no license file, so treat the license as unconfirmed until a LICENSE file is added.

Can it run unattended?

Yes. The ultradeep mode runs 20 to 45 minutes with disk-persisted evidence, a critique loop and a validation retry limit of three cycles, and long reports continue automatically.

Alternatives to Deep Research Skill

last30days

mvanhorn/last30days-skill

Research skill that searches Reddit, X, YouTube, Hacker News, Polymarket and more for the last 30 days, scores results by engagement, and writes a cited brief.

Recommended Broad access Depth 5/5

Academic Research Skills (ARS)

Imbad0202/academic-research-skills

Five very detailed Claude Code skills for the academic pipeline: deep research, paper writing, peer review, revision and systematic-review screening.

With caveats Check first Depth 5/5

Co-Researcher

poemswe/co-researcher

Academic research plugin with 15 skills; literature reviews check citations and quotes against OpenAlex, arXiv and Europe PMC.

Recommended Check first Depth 4/5

ARIS (Auto-Research-In-Sleep)

wanshuiyin/Auto-claude-code-research-in-sleep

A large markdown skill set for autonomous ML research: idea discovery, experiment runs and cross-model review loops where a second model critiques the first.

With caveats Broad access Depth 4/5

FAQ

What is the Deep Research skill for Claude Code?

It is a single skill that runs a multi-phase research pipeline and produces a citation-backed report with source scoring and automated validation. You trigger it with phrases like deep research on a topic.

How do I install the Deep Research skill?

Clone the repository into ~/.claude/skills/deep-research. Basic use needs no extra dependencies; search-cli with a provider API key is optional for broader search.

Is the Deep Research skill safe?

The scripts we read process files locally and make HTTP requests only to verify citations. The optional search-cli is a separate tool from the same author that handles your API keys, so review it before use.

Deep Research skill vs Anthropic's deep-research skill: which is better?

This repo claims to beat commercial deep-research products but publishes no benchmark. Compare both on a question you know well, and look at whether the citations and claim checks hold up.

Sources: GitHub repository (stars, license, file tree, last push Apr 11, 2026). Review based on the README and SKILL.md files as published.