TL;DR10+ community deep-research skills represent 10+ philosophies of 'how to do research.' From hyperresearch's persistent vault to jamoeight v2's Co-Scientist 6-agent, from adversarial verification to benchmark alignment. This article puts them all on one table.
Series: Deep Research 前沿 (14 / 15)
🌏 中文版
The previous article covered the project's own skill. This one covers the community.
From late 2025 to early 2026, the community flooded with 10+ deep-research skills. They represent not one solution, but 10+ philosophies of "how to do research."
Community Landscape
By Core Design
| Category | Representative | Core Idea |
|---|---|---|
| Persistent Knowledge | hyperresearch | SQLite index, knowledge accumulates across tasks |
| Enterprise-grade | hashbulla | NATO Admiralty grading, CRAG grounding |
| SOTA Paper Implementation | jamoeight v2 | Co-Scientist, AlphaEvolve, BrowseConf |
| Adversarial Verification | tolmachevmaxim | Optimist/Pessimist/Fact-Checker three agents |
| Scaled | kaynquang | 13 specialized agents |
| Academic-focused | Silence-view | STORM-inspired, citation chasing |
| Platform-agnostic | ramit-mitra | skills.sh compatible, 30+ platforms |
| Universal Adapter | Bhllcoder1 | 15 runtime adapters |
| Data-driven | Socialpranker | 75 report blocks, 280+ stat sources |
| Engineer-oriented | robertnowell | 6-phase, source quality gates |
| Human-in-loop | Weizhena | Human-in-the-loop, OpenCode/Codex |
Detailed Comparison
hyperresearch (jordan-gibbs)
- Core: 16-step pipeline + persistent vault + MCP server + 16 agents
- Unique: "Patch, don't regenerate"—research knowledge accumulates to SQLite
- Advantage: Cross-task knowledge reuse
- Cost: Complex architecture, high maintenance
jamoeight/claude-code-deep-research-v2
- Core: Co-Scientist 6-agent + AlphaEvolve + BrowseConf + BATS
- Unique: v1→v2 upgraded with novel hypothesis generation and evaluator-driven search
- Data: +10.3pp on deep-research benchmarks
- Advantage: Latest SOTA paper implementation
- Cost: Depends on Claude Code ecosystem
hashbulla/deep-research
- Core: 7-phase + NATO Admiralty 2×6 grading + CRAG grounding loop
- Unique: Enterprise-grade grading system
- Advantage: Structured evaluation framework
- Cost: Heavy framework
tolmachevmaxim/deep-research-skill
- Core: 3 agents (Optimist/Pessimist/Fact-Checker)
- Unique: Adversarial verification—not self-verifying, but three roles debating
- Advantage: Concise and effective, file-based state
- Cost: Depends on model quality
Common Architecture Patterns
Across all skills, these are nearly universal:
| Common Pattern | Description | Why Nearly Universal |
|---|---|---|
| Multi-agent parallel research | 2-13 agents researching simultaneously | Coverage breadth + speed |
| Source credibility scoring | Tier grading / 6 dimensions / NATO Admiralty | Filtering low-quality sources |
| Adversarial verification | Pessimist/Fact-Checker/Red-team | Preventing over-optimism |
| Triangulation | 3+ independent sources | Ensuring factual accuracy |
| File-based intermediate state | Crash recovery | Not losing progress on long tasks |
| Multi-tier depth modes | quick/standard/deep/ultradeep | Adapting to different needs |
| Progressive Disclosure | SKILL.md + references/ | Maintainability |
Comparison with Project Skill
| Dimension | Project Skill | External Skills |
|---|---|---|
| Tool Boundary | Strictly Groundlane | Mostly Tavily/Exa/Brave/Serper |
| Output Goal | .research/ → post skill publishing | Direct report output |
| Academic Quality | Has A/B/C/D grading | Some have (hashbulla, Silence-view) |
| Adversarial Verification | ❌ | ✅ Nearly all have it |
| Persistent Knowledge | ❌ | ✅ hyperresearch has it |
| Benchmark Alignment | ❌ | jamoeight v2 has BrowseConf |
| Cross-platform | ❌ Agent-environment only | ✅ skills.sh / Multi-runtime |
Designs Worth Learning
- hyperresearch's persistent vault—knowledge accumulates across tasks instead of starting from scratch each time
- tolmachevmaxim's three-agent adversarial—simple but effective
- Socialpranker's 75 blocks—breaking reports into manageable chunks
- robertnowell's source quality gates—filtering low-quality sources before research starts
Trends to Be Wary Of
- Skill explosion—10+ skills represent methodology fragmentation
- Platform dependency—most skills are tied to Claude Code or Codex
- Benchmark gaming—some skills optimize for benchmark scores, not real research quality
- Complexity creep—16 agents, 10 phases may be over-engineering
References
- hyperresearch (jordan-gibbs) — 16-step pipeline + persistent vault.
- jamoeight/claude-code-deep-research-v2 — Co-Scientist + AlphaEvolve.
- tolmachevmaxim/deep-research-skill — Three-agent adversarial verification.
- Socialpranker/deepdive — 75 blocks methodology.
- deep-research skill definition — Project skill definition.
- deep-research-survey-overview — Previous article: three-phase landscape classification.
Glossary
Loading...