[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-2018":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":18,"compositeScore":20,"rankGlobal":10,"rankLanguage":10,"license":21,"archived":22,"fork":22,"defaultBranch":23,"hasWiki":24,"hasPages":22,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":31,"readmeContent":32,"aiSummary":33,"trendingCount":16,"starSnapshotCount":16,"syncStatus":15,"lastSyncTime":34,"discoverSource":35},2018,"hyperresearch","jordan-gibbs\u002Fhyperresearch","jordan-gibbs","Agent-driven research knowledge base. Agents collect, search, and synthesize web research into a persistent, searchable wiki.","",null,"Python",445,38,7,2,0,11,33,298,4.77,"MIT License",false,"main",true,[26,27,28,29,30],"agents","agentskills","claude-code","deep-research","deep-research-agent","2026-06-12 02:00:35","\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fbanner.png\" alt=\"HYPERRESEARCH\" width=\"700\">\n\u003C\u002Fp>\n\n\u003Ch3 align=\"center\">The Most Powerful Deep Research Harness\u003C\u002Fh3>\n\n\u003Cp align=\"center\">\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fhyperresearch\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fv\u002Fhyperresearch\" alt=\"PyPI version\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fpypi.org\u002Fproject\u002Fhyperresearch\u002F\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fpypi\u002Fpyversions\u002Fhyperresearch\" alt=\"Python 3.11+\">\u003C\u002Fa>\n  \u003Ca href=\"LICENSE\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flicense\u002Fjordan-gibbs\u002Fhyperresearch\" alt=\"License: MIT\">\u003C\u002Fa>\n  \u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fjordan-gibbs\u002Fhyperresearch\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Fstars\u002Fjordan-gibbs\u002Fhyperresearch?style=social\" alt=\"GitHub stars\">\u003C\u002Fa>\n\u003C\u002Fp>\n\n---\n\n**Hyperresearch turns Claude Code into a deep research agent. and currently leads the DeepResearch-Bench RACE leaderboard (benchmarked internally).** A tier-adaptive 16-step pipeline produces adversarially-audited reports with full source provenance. Every fetched source lands in a persistent, searchable vault that compounds across sessions.\n\n\u003Cp align=\"center\">\n  \u003Cimg src=\"assets\u002Fbenchmark.png\" alt=\"DeepResearch-Bench top-5 — hyperresearch leads the chart ahead of Grep Deep Research, Cellcog Max, nvidia-aiq, Gemini Deep Research, and OpenAI Deep Research\" width=\"780\">\n\u003C\u002Fp>\n\n\u003Cp align=\"center\">\u003Csub>Forward-looking projection from a stratified pilot against the DeepResearch-Bench leaderboard snapshot (https:\u002F\u002Fhuggingface.co\u002Fspaces\u002Fmuset-ai\u002FDeepResearch-Bench-Leaderboard). Third party validation is pending.\u003C\u002Fsub>\u003C\u002Fp>\n\n## Install\n\n```bash\ncd your-project\npip install hyperresearch && hyperresearch install\n```\n\nThen `\u002Fhyperresearch \u003Canything>` in Claude Code.\n\n> Python 3.11–3.13. (3.14 not yet supported — use `pyenv install 3.13`, `uv venv -p 3.13`, or `py -3.13 -m venv .venv`.)\n>\n> Power users: `hyperresearch install --global` makes `\u002Fhyperresearch` reachable from every Claude Code session anywhere, at the cost of ~15 lines in every session's system reminder. Per-project install (above) keeps unrelated CC sessions clean.\n\n---\n\n## The 16-step research pipeline\n\nThe entry skill is a thin router. It bootstraps the canonical research query, then invokes one step skill per pipeline phase via Claude Code's `Skill` tool. Each step's procedure is loaded fresh into context only when needed defeating context-rot problems that makes long pipelines silently drop steps.\n\n| # | Step | What it does | Tiers |\n|---|---|---|---|\n| 1 | Decompose | Canonical query → atomic items + coverage matrix + tier classification | both |\n| 2 | Width sweep | Multi-perspective search plan + parallel fetcher waves (Haiku) | both |\n| 3 | Contradiction graph | Pair contradictions across the corpus into ranked clusters | full |\n| 4 | Loci analysis | Two parallel loci-analysts → scored loci with source budgets | full |\n| 5 | Depth investigation | K parallel depth-investigators → interim notes with committed positions | full |\n| 6 | Cross-locus reconcile | Reconcile committed positions → comparisons.md | full |\n| 7 | Source tensions | Extract expert disagreements → source-tensions.json | full |\n| 8 | Corpus critic | \"What source would overturn this?\" + targeted gap-fill fetch | full |\n| 9 | Evidence digest | Top claims + verbatim quotes → evidence-digest.md | full |\n| 10 | Triple draft | Per-angle source curation + 3 parallel draft sub-orchestrators (light: single draft) | both |\n| 11 | Synthesize | Plan + outline + spawn synthesizer subagent → final_report.md | full |\n| 12 | Critics | 4 adversarial critics in parallel → findings JSONs | full |\n| 13 | Gap-fetch | Targeted fetch wave for critic-identified vault gaps | full |\n| 14 | Patcher | Surgical Edit hunks applied to draft (tool-locked Read+Edit) | full |\n| 15 | Polish | Hygiene + filler pass (tool-locked Read+Edit subagent) | both |\n| 16 | Readability audit | Recommender writes JSON suggestions; orchestrator selectively applies | both |\n\n### Depth Modes \n\nIn your prompt, you can request one of two tiers and the rest of the pipeline scales accordingly. Full mode is default.\n\n| Tier | Steps that run | Typical time |\n|---|---|---|\n| `light` | bounded factual queries, surveys, comparisons — 1 → 2 → 10 → 15 → 16 | ~30–40 min |\n| `full` | deep argumentative analysis with adversarial review — all 16 steps | ~1.5–2.5 hours |\n\n### The two load-bearing principles\n\n1. **Patch, never regenerate.** After step 11 produces the synthesized report (or step 10 for light tier), the only modifications are surgical Edit hunks. The patcher and polish auditor are tool-locked to `[Read, Edit]` at the Claude Code allowlist level so they physically cannot Write a new draft. Per-hunk caps make \"just rewrite it\" mechanically impossible. Critic findings that don't fit a small hunk escalate as structural issues.\n\n2. **Canonical research query is gospel.** The verbatim user prompt is persisted to `research\u002Fquery-\u003Cvault_tag>.md` once and re-read by every subsequent step and every spawned subagent. Wrapper requirements (save paths, citation format, terminal sections) are a separate contract.\n\n### Subagent roster\n\n| Agent | Model | Role |\n|---|---|---|\n| `hyperresearch-fetcher` | Sonnet | URL fetching via crawl4ai; runs 8–12 in parallel per wave |\n| `hyperresearch-source-analyst` | Sonnet (1M ctx) | End-to-end digest of any single long source >5000 words |\n| `hyperresearch-loci-analyst` | Sonnet | Reads the width corpus, returns 1–8 depth loci with rationale |\n| `hyperresearch-depth-investigator` | Sonnet | Investigates one locus, writes one interim note with a committed position |\n| `hyperresearch-corpus-critic` | Sonnet | \"What source would overturn the current direction?\" pre-draft gap analysis |\n| `hyperresearch-draft-orchestrator` | Opus | One per draft angle; reads its curated source list and writes one draft |\n| `hyperresearch-synthesizer` | Opus | Reads all 3 drafts, writes the final report (two-pass write, Read+Write locked) |\n| `hyperresearch-dialectic-critic` | Opus | Counter-evidence the draft missed |\n| `hyperresearch-depth-critic` | Opus | Shallow spots interim notes could fill |\n| `hyperresearch-width-critic` | Opus | Topical corners the corpus supports but the draft ignores |\n| `hyperresearch-instruction-critic` | Opus | Structural mismatches against the prompt's atomic items |\n| `hyperresearch-patcher` | Opus | Tool-locked `[Read, Edit]`. Applies critic findings as surgical Edit hunks |\n| `hyperresearch-polish-auditor` | Opus | Tool-locked `[Read, Edit]`. Cuts filler, strips hygiene leaks |\n| `hyperresearch-readability-recommender` | Opus | Writes JSON suggestions for paragraph rhythm and list\u002Ftable conversion |\n\n---\n\n## The vault: persistent, searchable, compounding\n\nHyperresearch is not a one-shot report generator like most other Deep research harnesses. Every fetched source lands in a SQLite-indexed vault that every future research session can reuse.\n\n```bash\nhyperresearch search \"ion-trap gate fidelity\" -j           # Full-text search\nhyperresearch search \"quantum\" --include-body -j           # Full-body search\nhyperresearch note show \u003Cid1> \u003Cid2> \u003Cid3> -j               # Batch-read notes\nhyperresearch graph hubs -j                                # Most-connected notes\nhyperresearch graph backlinks \u003Cid> -j                      # Reverse links\nhyperresearch lint -j                                      # Health check (broken links, missing tags)\n```\n\n**Markdown is truth, SQLite is cache.** Notes live as plain markdown with YAML frontmatter in `research\u002Fnotes\u002F`. The SQLite index is fully rebuildable. Delete it and `hyperresearch sync` reconstructs it from the markdown. The vault is inspectable in any editor, version-controllable in git, and readable without the tool installed.\n\n**PDFs fetch directly.** `hyperresearch fetch` auto-detects PDF URLs (arXiv, NBER, SSRN, direct `.pdf` links) and extracts full text via pymupdf. Raw PDFs land in `research\u002Fraw\u002F\u003Cnote-id>.pdf` and the note's `raw_file:` frontmatter links back.\n\n**Provenance breadcrumbs.** Every fetched source carries a `--suggested-by` link back to whatever surfaced it. The chain forms a rooted tree from seed fetches; the `provenance` lint rule catches disconnected components.\n\n---\n\n## What's structurally enforced\n\n- **Verbatim prompt as gospel** — `scaffold-prompt` lint blocks if the scaffold doesn't open with the user's exact prompt\n- **Locus coverage** — every step 4 locus must have a step 5 interim note; missing interims flag as errors\n- **Patch-only modification** — steps 14, 15, 16 are tool-locked to `[Read, Edit]`. They cannot regenerate the draft\n- **Critical findings never silently skip** — `patch-surgery` lint surfaces any critical finding the patcher couldn't apply\n- **Schema integrity** — `tier`, `content_type`, and `type` are SQLite CHECK-constrained vocabularies; corrupted frontmatter cannot poison the index\n- **Hygiene leaks caught on the way out** — scaffold sections, YAML frontmatter, and prompt echoes are stripped by step 15 before ship\n\n---\n\n## Authenticated crawling\n\nFetch from LinkedIn, Twitter, paywalled sites or anything you can log into:\n\n```bash\nhyperresearch setup       # Browser opens. Log into your sites. Done.\n```\n\nLinkedIn, Twitter, Facebook, Instagram, and TikTok automatically use a visible browser to avoid session kills.\n\n---\n\n## Academic APIs before web search\n\nFor any topic with a research literature, hit academic APIs BEFORE web search. They return citation-ranked canonical papers; web search returns derivative commentary.\n\n- **Semantic Scholar** — `https:\u002F\u002Fapi.semanticscholar.org\u002Fgraph\u002Fv1\u002Fpaper\u002Fsearch`\n- **arXiv** — `https:\u002F\u002Fexport.arxiv.org\u002Fapi\u002Fquery`\n- **OpenAlex** — `https:\u002F\u002Fapi.openalex.org\u002Fworks`\n- **PubMed** — `https:\u002F\u002Feutils.ncbi.nlm.nih.gov\u002Fentrez\u002Feutils\u002Fesearch.fcgi`\n\nAfter the academic sweep, run web searches for context, news, non-academic angles, and at least one adversarial search (\"criticism of X\", \"limitations of X\").\n\n---\n\n## What it doesn't do\n\n- It doesn't replace your judgment on which sources matter. The agent picks, you steer.\n- It can't fetch what's behind a paywall you haven't logged into.\n- It runs on Anthropic models Opus + Sonnet + Haiku via the subagent roster. Costs scale with tier and corpus size. If anyone wants to port this to Codex, put up a PR! \n- The lint gate catches **structural** failures (missing scaffold, broken provenance, unresolved CRITICALs). It cannot guarantee factual accuracy, that's still your call.\n\n---\n\n## Requirements\n\n- Python 3.11+\n- [Claude Code](https:\u002F\u002Fclaude.com\u002Fclaude-code)\n\n---\n\n## License\n\n[MIT](LICENSE)\n\n---\n\n## Star History\n\n[![Star History Chart](https:\u002F\u002Fapi.star-history.com\u002Fsvg?repos=jordan-gibbs\u002Fhyperresearch&type=Date)](https:\u002F\u002Fstar-history.com\u002F#jordan-gibbs\u002Fhyperresearch&Date)\n","Hyperresearch 是一个基于代理的研究知识库，能够收集、搜索并整合网络研究资料到一个持久且可搜索的维基中。其核心功能包括通过16步管道生成经过对抗性审核的报告，并确保所有获取的资料源都存入一个跨会话累积的持久化存储库中。项目利用Python语言开发，特别适合需要深入研究和长期保存研究成果的场景，如学术研究、市场分析等。此外，该项目在DeepResearch-Bench排行榜上表现优异，证明了其在深度研究领域的强大能力。","2026-06-11 02:47:36","CREATED_QUERY"]