Oh-My-Research (OMR) is an Agent Skill that turns papers, web sources, and other materials into evidence-bound, publication-quality surveys and reports.
You bring a topic and sources. The skill collects, analyzes with clear evidence boundaries, optionally deepens with structured thinking (first principles, Socratic, triangulation, …), then writes a long-form deliverable chapter by chapter — and exports a polished Word, PDF, or Markdown document in your preferred language.
Most “research agents” optimize for speed, chat summaries, or code demos. OMR optimizes for something else:
High-quality deep research reports grounded in collected materials and evidence.
That means:
- Report-first — the primary outcome is a survey, report, manuscript, or brief readers can trust
- Evidence-bound — claims stay within what sources prove, suggest, or only allow you to infer
- Think before you write — structured paradigms deepen analysis when confidence is low or gaps are high
- Publication-safe — the finished document uses normal citations and prose; workflow jargon stays private
- Not a coding lab — no experiment/prototype evaluation path; research stays research
One skill. One conversation. It detects what you want (and what the workspace already has), then runs the right step.
collect → analyze ⟲ think → synth (survey or report)
Analyze and Think form a side-cycle: deepen with structured paradigms when confidence is low or gaps are high, then return to analysis until the judgment is ready for synthesis.
| Phase | What happens |
|---|---|
| Collect | Papers, URLs, repos, datasets land in materials + indexes |
| Analyze | Brief, evidence map, and judgment — with explicit evidence grades |
| Think | Side-cycle on analyze: first principles, red team, steelman, … then back |
| Synth | Long survey or report written incrementally, then exported as DOCX, PDF, or Markdown |
Optional paths: capture ideas, take a stance (decide), or reconcile when new evidence contradicts prior claims.
Default workflow pattern: Evidence-Deep (best for literature surveys and structured research reports). Other patterns (rapid brief, idea-first, stance-first, loop) are available when you need them.
OMR covers the research lifecycle. Pair it with FlowJet for the document and diagram skills that reports actually need:
- Docs writing — structure and polish long-form Markdown
- draw.io — figures and architecture diagrams
- anydoc — convert PDFs and office files to Markdown (the same conversion path COLLECT uses)
- Other collaborative skills shipped with FlowJet (planning, MCP, and related tooling)
Install both; use OMR for collect → analyze → think → synth, and FlowJet when you need figures, conversion, or extra writing help around the same workspace.
git clone git@github.com:caesar0301/flowjet-agent.git
pip install flowjet-agent # or: uv tool install flowjet-agentClaude Code Marketplace
/plugin marketplace add caesar0301/oh-my-research
/plugin install oh-my-research@oh-my-research
/reload-plugins
npx skills
npx skills add caesar0301/oh-my-researchOnce the skill is loaded, talk naturally — or name an operation:
| You might say | What runs |
|---|---|
“Start research on …”, init "…" |
Bootstrap the workspace |
Paste a URL / DOI / arXiv / GitHub link, or collect … |
Add materials |
“Analyze what we have”, analyze |
Build evidence + judgment |
“First principles on the judgment”, think … |
Deepen the latest artifact |
“Write the survey / report”, synth |
Produce the long report |
| “New paper contradicts our claim” | Propose reconcile |
You can also run the full graph-guided flow with workflow.
Reports are written chapter by chapter (with resume if interrupted), then assembled. Language defaults from your timezone/locale (e.g. Asia/Shanghai → Chinese, Asia/Tokyo → Japanese), or set --language explicitly. Prefer --format docx, pdf, or md for the deliverable you will share.
- A research workspace (
AGENTS.md,.omr/state, materials, indexes, plans) - Working artifacts: brief, evidence map, judgment (and optional plan / decision / ideas)
- A self-contained survey or report under
docs/{survey,report,manuscript,brief}/ - Professional Word, PDF, or Markdown export with cover, TOC, and consistent formatting
- Quality checkpoints before you publish (analyze gate + document lenses + publish gate)
Internal IDs, gate names, and OMR workflow terms stay in private working files. The published document reads like a normal research report.
| Op | Purpose |
|---|---|
init |
Create the research workspace |
collect |
Ingest sources into materials + indexes |
analyze |
Deep analysis → brief, evidence map, judgment |
think [method] |
Structured elicitation on a research artifact |
synth |
Incremental long report → Word/PDF/Markdown |
idea / decide |
Optional speculation or stance |
reconcile |
Repair state when evidence conflicts |
workflow |
Graph-driven multi-step run (default Evidence-Deep) |
qa |
Run quality-gate checklists |
version |
Tag / backup research artifacts |
Details: skills/oh-my-research/SKILL.md · full reference: references/REFERENCE.md
- Trust auto-detection — override with a named op when you know the next step
- Evidence grades matter — do not treat “suggests” as “proves”
- Deepen when unsure — use THINK before locking analysis and writing
- Long reports are incremental — outline → chapters → continuity → assemble; resume if needed
- Share the document, not the workflow — the exported document should stand alone for its reader
For contributors working in this repo:
make help
make format # python + json + yaml
make check # CI: format checks + lintPackage version lives in skills/oh-my-research/SKILL.md (metadata.version); history in CHANGELOG.md. CI: .github/workflows/ci.yml.
Apache License 2.0 — see LICENSE.