FrontierScholar 知沿学术

A research wiki of about 1.01M computer-science arXiv papers, built for coding agents. Papers are organised into domains, topics and subtopics; each subtopic has pages on its research problems, method families, datasets and benchmarks, and a frontier page of open gaps and rising approaches. Every claim points back to arXiv ids.

This repository has the fs command-line client and the fs-cli agent skill. Website and API documentation: https://frontier-scholar.com.

Install

pip install frontier-scholar      # Python 3.8+, no dependencies
fs --version                      # CLI and API versions, data snapshot

fs talks to the public endpoint https://frontier-scholar.com and needs no key (see Limits).

Agent skill

skills/fs-cli/ tells a coding agent (Claude Code, Codex, Cursor, Gemini CLI, pi, ...) when and how to use fs:

npx skills add FrontierScholar/FrontierScholar
# or
gh skill install FrontierScholar/FrontierScholar

or copy skills/fs-cli/ into your agent's skills folder. Then ask in plain language, e.g. "Use fs: what are the open problems of adaptive retrieval, and which approaches are rising?"

Commands

fs select "adaptive retrieval" --topk 3
fs search paper "adaptive retrieval" --before 2025-01-01
fs search wiki frontier "adaptive retrieval" --status open
fs read 2601.04618
fs meta dates 2601.04618 2604.15621 --before 2026-06-01

Answers are JSON (wiki pages print as Markdown). They are lookups and keyword (BM25) rankings over one fixed data snapshot: nothing is generated. Wiki pages are summaries written from the papers; the only original text is a paper's abstract. fs COMMAND -h lists every option and output field.

From Python:

from frontier_scholar import Client

r = Client().call("search/paper", {"query": "adaptive retrieval", "topk": 5})
for p in r["items"]:
    print(p["paper_id"], p["date"], p["title"])

The HTTP API is POST https://frontier-scholar.com/api/v1/fs/<command> with the arguments as JSON; see the documentation.

Limits

Without a key, each IP gets 30 requests a minute and 1,000 a day, and calls are smaller (--topk up to 20, read up to 5 targets, meta up to 50 ids, no --grep); curl https://frontier-scholar.com/api/v1/fs/quota shows what is left. With a key from the FrontierScholar team, set FS_API_KEY (in the environment, never in code); calls then use that account's limits. When a limit is reached, fs waits and retries by itself. Queries should be in English: the index is English text. FS_BASE points fs at another service.

Layout

src/frontier_scholar/   the fs CLI and its client (standard library only)
skills/fs-cli/          the agent skill: SKILL.md, references/, CHANGELOG.md
tests/                  CLI tests

The earlier frontier-scholar SDK and skills are on the legacy-sdk branch.

License

Apache-2.0, see LICENSE.