skim

Generate plain-language narratives and technical summaries from arxiv papers.

Three Ways to Use skim

#	Method	Best for	API key needed?	Cached?
1	CLI + OpenAI	Regular use, any OpenAI-compatible API (OpenAI, OpenRouter, Ollama, local models)	Yes	Yes
2	CLI + Claude	Already have a Claude Code subscription, don't want another API key	No	Yes
3	Claude Code Plugin	Already inside Claude Code, want one-command summaries	No	No

---

Option 1: CLI + OpenAI

Install once, use anywhere from your terminal. Works with any OpenAI-compatible API — OpenAI, OpenRouter, Ollama, or any local model server.

uv tool install git+https://github.com/dipta007/skim    # or pipx, or pip, see the Install section below
skim init                                               # select "openai", enter API key
skim -p 2509.16538 -t story                             # generate summary

Option 2: CLI + Claude

Same CLI, but uses your existing Claude Code subscription — no API key needed. Requires the claude CLI to be installed and logged in.

uv tool install git+https://github.com/dipta007/skim    # or pipx, or pip, see the Install section below
skim init                                               # select "claude"
skim -p 2509.16538 -t story                             # generate summary

Option 3: Claude Code Plugin

Already inside Claude Code? Install the plugin and use slash commands — no setup, no API key.

/plugin marketplace add dipta007/skim

/plugin install skim@dipta007-skim

Then, inside your claude-code:

/story 2509.16538
/deep 2509.16538

Claude reads the paper and generates the summary directly.

---

Summary Types

Type	What you get
story	A plain-language, analogy-driven narrative — no jargon, no equations
deep	A structured technical summary with methodology, results, and key contributions

Browser Viewer

Open summaries in the browser with proper LaTeX math rendering, dark/light theme toggle, and a readable serif font:

skim -p 2509.16538 -t deep --open

Cache Management

Summaries are cached locally so repeated lookups are instant. To clear the cache:

skim clean                    # remove all cached summaries
skim clean -p 2509.16538      # remove cache for a specific paper

Configuration

Config lives at ~/.config/skim/config.toml. Re-run skim init to change settings.

Example configs

OpenAI backend:

[api]
backend = "openai"
key = "sk-your-key"
base_url = "https://api.openai.com/v1"
model = "gpt-5.4-nano"

[output]
dir = "~/papers/skim"

Claude backend:

[api]
backend = "claude"
key = ""
base_url = ""
model = "sonnet"

[output]
dir = "~/papers/skim"

Install

With uv (recommended):

uv tool install git+https://github.com/dipta007/skim

With pipx:

pipx install git+https://github.com/dipta007/skim

With pip:

pip install git+https://github.com/dipta007/skim

From source:

git clone https://github.com/dipta007/skim.git
cd skim
make install

Development

git clone https://github.com/dipta007/skim.git
cd skim
make install    # Install dependencies + set up git hooks
make test       # Run tests
make lint       # Check code style
make format     # Auto-format code

make install also configures git hooks that run the formatter/linter on commit and tests on push.

Roadmap

• Support hf papers read cli to fetch papers from Hugging Face and summarize them (less tokens)
• Support local PDF files (not just arxiv IDs)
• skim list — show all previously summarized papers
• skim history — recently read papers
• Export to PDF (from browser viewer / CLI flag when --open is not used)
• Custom prompt types — let users add their own .md prompts beyond story/deep (interactive)
• skim search — semantic search over local summaries or global database
• Global database with public summary gallery

Related Projects

See RELATED.md for a curated list of similar tools, Claude Code skills, plugins, agents, and MCP servers for AI-powered research workflows.

Contributing

See CONTRIBUTING.md for development setup, code style, and PR guidelines.

Star History

License

MIT