Frontier

Frontier is an agent skill for reliably finding recent papers on AI and other technical topics.

It searches OpenAlex and arXiv, keeps papers within an inclusive requested date range (the previous 90 days by default), merges duplicates, and returns a concise source-linked shortlist.

What you get

• Five to ten relevant papers, fewer when the evidence is sparse
• Publication date and status, including clear preprint labels
• Author institutions when OpenAlex provides them
• Abstract-grounded findings and evidence-level labels
• Compact provider coverage and limits

The final shortlist may favor a relevant mix of frontier startups, established university labs, research organizations, and enterprises. Institution diversity is an editorial tie-breaker, not a quota and never overrides topic relevance.

Supported agents

• Claude Code: /frontier
• Codex: $frontier
• OpenCode: native skill activation
• Pi: native skill activation

Install

curl -fsSL https://raw.githubusercontent.com/dannylee1020/frontier/main/scripts/install.sh | sh

Frontier requires Python 3.12 or newer.

How it works

1. Forms two or three distinct technical search queries.
2. Searches OpenAlex and arXiv concurrently.
3. Filters by first public date within the requested window.
4. Merges duplicates by DOI, arXiv ID, then conservative title/author/year matching.
5. Returns a deterministic candidate ranking for agent review and final editorial selection.

OpenAlex affiliation metadata is retained when available. arXiv-only papers have no inferred affiliation. Source failures remain visible as incomplete coverage rather than zero results.

No credentials or special provider configuration are required.