science-skills
This repository contains the prompt, skill, agent, and MCP server assets for a Claude Science style scientific computing assistant. The contents are mostly runtime-facing configuration and…
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/jimliu--science-skills)HTML
<a href="/listings/jimliu--science-skills"><img src="/api/listings/jimliu--science-skills/badge.svg" alt="findsafeskills safety rating" /></a>Install
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
README.md · README · 131 lines
View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)
This file has changed since it was scanned, so this is not exactly the text that was scanned, and no line is marked. The findings are listed with the safety scan above.
- # Science Skills
- This repository contains the prompt, skill, agent, and MCP server assets for a
- Claude Science style scientific computing assistant. The contents are mostly
- runtime-facing configuration and documentation: skills teach the assistant how
- to perform domain workflows, agents define specialized profiles, and MCP
- servers expose scientific data and chemistry tools.
- ## Source
- The `skills/` and `mcp-servers/` assets match **Claude Science 0.1.71**
- (public release, macOS arm64, built October 26, 2026). They are copied directly
- from the app's embedded asset archive, including helper code, dependencies,
- licenses, and widget assets.
- `VERSION` records this asset snapshot version. `SOURCE.json` records the app's
- build metadata, source checksums, and file-tree checksums for both directories.
- Upstream package manifests and skill metadata are preserved verbatim; their
- dependency versions are independent of the app version. The base prompt and
- agent profiles are retained from the earlier repository snapshot and are not
- covered by this version marker.
- The base prompt and bundled skills in this repository are sourced from the
- Claude Science product. The public product page is available for context:
- https://claude.com/product/claude-science.
- ## Repository Layout
- | Path | Purpose |
- | --- | --- |
- | `SYSTEM_PROMPT.md` | Base system prompt and operating rules for the scientific computing agent. |
- | `skills/` | Skill packages. Each skill is centered on a `SKILL.md` file, with optional helper code, references, requirements, and provider metadata. |
- | `agents/` | Agent profile metadata and prompts for specialized roles such as onboarding, transcript review, bookmarking, and the general science agent. |
- | `mcp-servers/` | Bundled MCP servers and widgets used by the agent for external scientific data and chemistry workflows. |
- ## Skill Catalog
- The `skills/` directory includes workflows for several broad areas:
- - Biomolecular modeling and design: `alphafold2`, `boltz`, `chai1`,
- `openfold3`, `esmfold2`, `fair-esm2`, `diffdock`, `proteinmpnn`,
- `ligandmpnn`, `solublempnn`, `evo2`, and `borzoi`.
- - Scientific analysis and reporting: `literature-review`, `pdf-explore`,
- `figure-style`, `figure-composer`, `paper-narrative`, and
- `indication-dossier`.
- - Single-cell and omics workflows: `scgpt` and `scvi-tools`.
- - Runtime and environment operations: `remote-compute-modal`,
- `remote-compute-ssh`, `compute-env-setup`, `managed-model-endpoints`, and
- `using-model-endpoint`.
- - Product and customization workflows: `customize`, `skill-creator`,
- `product-self-knowledge`, and `self-awareness`.
- Most skills start with YAML front matter:
- ```yaml
- ---
- name: example-skill
- description: When and why this skill should be loaded.
- license: Apache-2.0
- ---
- ```
- The body of `SKILL.md` should give the operational procedure, expected inputs,
- failure modes, validation steps, and any tool-specific constraints needed for
- reliable execution.
- ## Agents
- Agent profiles live under `agents/<name>/metadata.yaml`.
- - `operon` is the general-purpose scientific computing agent.
- - `onboarding` supports first-run user onboarding.
- - `reviewer` audits another agent transcript for unsupported claims,
- fabrication, or plan deviation.
- - `bookmarker` selects transcript spans worth preserving as navigation
- breadcrumbs.
- These files are configuration assets, not standalone applications. Keep prompt
- changes tightly scoped and preserve any eval or benchmark notes near the prompt
- they justify.
- ## MCP Servers
- The `mcp-servers/` directory contains bundled MCP integrations:
- - `bio-tools` vendors multiple Python packages for biological and biomedical
- data retrieval. `run_server.py` launches a named stdio server from
- `mcp-servers/bio-tools/lib/`.
- - `ketcher-chemistry` contains a chemistry MCP server and widget assets for
- structure editing and chemistry workflows.
- The MCP code is runtime infrastructure. When changing it, prefer narrow edits
- and verify both the server entry point and the shape of returned tool data.
- ## Working With This Repo
- Useful inspection commands:
- ```bash
- # List all skills
- find skills -maxdepth 2 -name SKILL.md | sort
- # List agent profiles
- find agents -maxdepth 2 -name metadata.yaml | sort
- # Show launchable bio-tools MCP servers
- find mcp-servers/bio-tools/lib -maxdepth 2 -name server.py \
- | sed 's#^mcp-servers/bio-tools/lib/##; s#/server.py$##' \
- | sort
- ```
- Before editing a skill or prompt:
- 1. Read the relevant `SKILL.md` or `metadata.yaml` completely.
- 2. Preserve existing safety, artifact, provenance, and validation rules unless
- the change explicitly updates them.
- 3. Keep examples executable and aligned with the current runtime surface.
- 4. Add or update references next to the skill when a workflow depends on
- external documentation or non-obvious operational knowledge.
- ## Maintenance Notes
- - Use ASCII in new files unless the surrounding file already requires a wider
- character set.
- - Avoid broad rewrites of prompts or generated MCP bundles without a concrete
- regression target.
- - Treat bundled third-party services, model weights, and APIs as dependencies:
- document licenses, terms, privacy implications, and data movement in the
- relevant skill metadata when applicable.
- - If a skill ships helper code such as `kernel.py`, keep the markdown
- instructions and helper APIs in sync.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.