Open-Science Puts 24 Built-in Skills in a Reproducible Research Workbench

Research done in a chat window is hard to trust: the agent's answer arrives with no way to check which code produced it. AIPOCH Open-Science, a 5k-star open-source research workbench, puts skills, code execution, and provenance in one local workspace so every result carries its own evidence.

Why This Skill Matters

Open-Science is an open-source, local-first, model-agnostic AI research workbench for scientists and researchers, licensed Apache-2.0 and available on macOS, Windows, and Linux. It supports the research process from literature review and hypothesis development through code execution, data analysis, simulation, and visualization, and keeps immutable artifact versions with captured producer code and execution history.

Two numbers from the README's capability table frame the skill story: the workbench extends research workflows with 24 built-in skills, and 525 skills are available from its Skills Marketplace, with one-click installation and updates. The five agent backends — Claude Code, OpenCode, Codex, CodeBuddy, and Pi — are all first-class execution options, alongside 48 built-in scientific connectors.

The built-in catalog in the repository's resources/skills/ directory includes protein-structure work (alphafold2, esmfold2, openfold3, boltz, chai1), literature and writing support (literature-review, paper-narrative, docx-generation), figure tooling (figure-composer, figure-style), and infrastructure (remote-compute-ssh, env-management, skill-creator).

Real Workflow: Run a Literature Review from the CLI

Open-Science ships a standalone CLI alongside the desktop app. After installing it (npm requires Node.js; Homebrew and the bundled-runtime packages include it), initialize the profile and start the background service:

# Initialize the local CLI profile and start the service in the background
open-science init
open-science start --no-open

Then create a project and submit a research task that pins a specific skill:

open-science project create "Systematic review"
open-science run --project "Systematic review" \
  --prompt-file ./task.md \
  --approval-profile auto \
  --skill literature-review \
  --wait --json

When the run finishes, list and download the generated artifacts:

open-science artifacts list <session-id> --json
open-science artifacts download <artifact-id> --output ./report.md

In the desktop app the same flow is interactive: create a New project, describe the research goal, use @ to reference project files and / to choose a skill, then inspect each result's evidence under Provenance — content identity, producer code, execution history, inputs, environment inventory, producing conversation context, and reviewer evidence, with anything the app could not verify marked unavailable.

Tips

• Skills are created through conversation or from completed work, and you can import skill packages or GitHub sources. Saving a GitHub repository does not install its skills — imports are explicit.
• Marketplace contributions are published after review; local imports do not publish skills. The marketplace lives in its own repository, aipoch/openscience-skill-marketplace.
• Completed research sessions export as portable .science packages with selected conversation branches, file versions, Notebook records, and verification evidence — imports are read-only and do not execute code.
• Some built-in sources are disabled by default and need per-source opt-in; manage tool-level permissions before letting agents reach external data sources.

When Not to Use This

Open-Science is a research workbench, not a general-purpose coding assistant — its skills and connectors are built around scientific workflows, and its README positions the installed app and release notes as the source of truth for changing catalogs. If your task is web app development or bulk file editing, a coding agent without the scientific wrapper is the lighter tool.

See the leaderboard for more skills.