Reusable research workflows for Codex, Claude Code, and other Agent Skills hosts.
Install one focused capability—or compose a complete path from research question to paper artifact.
OpenPaper Skills is the standalone distribution of research skills developed in
AI-Human Research OS.
Each skill packages instructions, references, scripts, and templates behind one
SKILL.md, so an existing agent project can gain a research capability without
adopting an entire workspace.
Tip
Need the complete long-horizon environment—project templates, research memory, human context, paper wiki, and operating rules? Start with AI-Human Research OS.
gh skill is available in GitHub CLI v2.90.0 or later and is currently in
public preview. Preview a skill before installing it:
gh skill preview pengqianhan/open-paper-skills research-bibleThen install one skill for Codex at user scope:
gh skill install pengqianhan/open-paper-skills research-bible --agent codex --scope userOr install the full maintained catalog:
gh skill install pengqianhan/open-paper-skills --all --agent codex --scope userReplace codex with another supported host such as claude-code when needed.
See GitHub's gh skill guide
for install scopes, updates, and version pinning.
You do not need the whole stack. A few useful starting combinations:
| If you want to… | Start with… | Add when needed… |
|---|---|---|
| Turn an idea into an executable research plan | research-bible |
task-file-builder for clean session handoffs |
| Keep papers, concepts, and project links connected | paper-wiki-manager |
human-cognition-cache for human context and knowledge state |
| Produce publication-ready diagrams and plots | drawio-paper |
codex-paper-figure-skill for editable figure concepts |
| Explain a paper or difficult concept interactively | explain-anything-html |
discover-academic-skills to locate adjacent capabilities |
| Build a complete durable research workspace | AI-Human Research OS | Use this repository only for selectively installed skills |
Everything under skills/ is part of the installable OpenPaper
catalog and is discovered by GitHub CLI through the
skills/<skill-name>/SKILL.md convention.
| Skill | Description |
|---|
| codex-paper-figure-skill | Create editable, journal-style academic figures and diagrams from paper text or figure descriptions. |
| discover-academic-skills | Find and evaluate research-oriented agent skills with academic-relevance and quality gates. |
| drawio-paper | Generate publication-quality academic diagrams and statistical plots. |
| explain-anything-html | Produce rich HTML explanations of papers, articles, documentation, and complex concepts. |
| explain-paper-html | Build an evidence-traceable interactive HTML lesson from an academic paper, verified in a browser before hand-off. |
| filetree-simple | Generate and lint a compact top-level FILETREE.md navigation map from directory-owned entrypoints. |
| human-cognition-cache | Maintain a project-local, git-trackable cache of human context and knowledge state. |
| karpathy-coding-rules | Apply a focused set of coding rules and working conventions. |
| map-then-territory | Draw a human-verifiable route map from start to destination, then drive agents through the territory edge by edge. |
| okf-repo-organizer | Organize repositories and knowledge corpora into Open Knowledge Format bundles. |
| paper-search | Search six literature sources in one command with one result shape, honouring the human's per-source on/off switches. |
| paper-wiki-manager | Maintain a structured paper wiki, including paper notes, concepts, topics, and visualizations. |
| project-dispatch | Act as the root agent: brief, dispatch, track, and review project work run through os-harness. |
| research-bible | Turn ML/AI research principles into research plans, experiment loops, logs, and debugging routines. |
| research-project-manager | Create, validate, update, sync, and archive Research Projects under projects-folder/ against one template contract. |
| sell-research-honestly | Audit evidence and turn research into persuasive, audience-specific value communication without overclaiming. |
| session-handoff | Maintain or resume a repository-root HANDOFF.md that transfers a cross-session task arc to a cold session. |
| skill-organizer | Register a newly added hub skill into its collection's index.md and README.md. |
| task-file-builder | Draft context-rich task.md briefs for fresh agent sessions. |
| uv-env | Create and manage Python environments and dependencies with uv. |
| writing-great-prompt | Turn an intent into a lean, copy-ready prompt contract with destination, evidence, authority, and completion bar. |
Catalog maintenance notes live in skills/README.md.
| Directory | What it contains | Installed by --all? |
|---|---|---|
skills/ |
OpenPaper-maintained, publishable skills | Yes |
collected-skills/ |
Third-party skills retained for provenance, evaluation, or adaptation | No |
Collected skills keep their upstream attribution and license terms. Before
using one, inspect its SKILL.md, scripts, network access, source, and license.
The repository-level MIT license does not replace third-party terms. See the
collected catalog and intake rules.
OpenPaper Skills has one development source and one installable distribution:
AI-Human Research OS / research-skills-hub
│
│ daily, manual, or repository-dispatch sync
▼
pengqianhan/open-paper-skills
├── skills/ installable catalog
└── collected-skills/ provenance archive
The sync-upstream-skills
workflow mirrors child directories containing SKILL.md. Repository-specific
README files are preserved, .syncignore exclusions are honored,
and .upstream-revision records the exact source commit
after synchronization.
Maintainer setup: trigger an immediate cross-repository sync
Add the following workflow to
AI-Human-Research-OS/.github/workflows/notify-openpaper-sync.yml:
name: Notify OpenPaper skill sync
on:
push:
branches: [main]
paths:
- "research-skills-hub/open-paper-skills/**"
- "research-skills-hub/collected-skills/**"
jobs:
notify:
runs-on: ubuntu-latest
steps:
- name: Trigger OpenPaper sync
env:
GH_TOKEN: ${{ secrets.OPENPAPER_DISPATCH_TOKEN }}
run: |
gh api repos/pengqianhan/open-paper-skills/dispatches \
--method POST \
-f event_type=research-skills-updated \
-f "client_payload[source_sha]=${GITHUB_SHA}"Create OPENPAPER_DISPATCH_TOKEN as a fine-grained token scoped only to
pengqianhan/open-paper-skills with Contents: read and write, then save it as an
Actions secret in AI-Human Research OS. In this repository, enable
Settings → Actions → General → Workflow permissions → Read and write.
Validate metadata and repository settings before publishing:
gh skill publish --dry-runTag releases, then pin installations when reproducibility matters:
gh skill install pengqianhan/open-paper-skills research-bible --pin v0.1.0Record the full research-skills-hub commit SHA in each sync commit or release
note so an installed release can be reconciled with its development source.
open-paper-skills/
├── skills/ # maintained skills published through gh skill
├── collected-skills/ # third-party provenance and evaluation archive
├── scripts/ # catalog synchronization tooling
├── assets/readme/ # editable README visual system
└── .github/workflows/ # automated upstream synchronization
OpenPaper-owned code and documentation are licensed under the
MIT License. Every entry in collected-skills/ retains its upstream
license and attribution.