sophon-research

name	sophon-research
description	Research AI evals, benchmarks, models, tools, leaderboards, capabilities, organizations, people, and papers using Sophon's public read-only catalog and local CLI. Use when comparing AI benchmarks, discovering papers, mapping eval/model/tool relationships, or collecting cited AI research context.
license	MIT
compatibility	Requires internet access and the sophon-research CLI.

Sophon Research

Use this skill when the user asks for AI research context from Sophon's public catalog: evals, benchmarks, models, tools, leaderboards, capabilities, organizations, people, and papers.

Prefer the local sophon-research CLI. If it is not on PATH and this skill is installed as a Codex plugin, resolve the plugin root from this file at <plugin-root>/skills/sophon-research/SKILL.md and use <plugin-root>/bin/sophon-research (../../bin/sophon-research relative to this file). Use ./bin/sophon-research only after verifying the current checkout is Sophon Research itself, with .codex-plugin/plugin.json name sophon-research and skills/sophon-research/SKILL.md present.

Sophon is read-only, unauthenticated, and should be queried through its API/text endpoints rather than scraped pages.

Quick Start

Check the CLI and API:

sophon-research doctor
sophon-research api

Search and fetch structured entities:

sophon-research search "swe-bench" --limit 5
sophon-research search "swe-bench" --type evals --json
sophon-research search "swe-bench" --type eval --per 5 --sort recent --json
sophon-research get evals swe-bench --json

Fetch paper context only when Sophon exposes it under the source license:

sophon-research paper <paper-slug> --text
sophon-research paper <paper-slug> --text --output /tmp/paper.txt
sophon-research paper <paper-slug> --pdf --output /tmp

Use --json when another tool or agent will consume the result. Human output is compact and may be truncated; JSON output is stable but still bounded for large paper text unless --output is used.

When To Use

• Discover evals, benchmarks, datasets, leaderboards, and top scores.
• Compare models, tools, and agents across Sophon entities.
• Find papers related to an eval, model, capability, person, or organization.
• Map relationships across entities before checking primary sources.
• Gather source-linked context for an AI research summary.

When Not To Use

• Do not use Sophon during benchmark-solving contexts where internet access is disallowed.
• Do not scrape Sophon pages when CLI/API endpoints are available.
• Do not treat Sophon as the final authority for high-stakes technical claims. Verify important claims against primary papers, official docs, or benchmark repositories.
• Do not commit downloaded papers, generated research outputs, caches, or local files created during research.

Workflow

1. Start broad with search, then narrow by type when the user asks about a specific class of entity. Use --sort recent when recency is more important than relevance.
2. Fetch the best candidate entities with get <type> <slug> --json.
3. Follow relationships in the entity JSON: scores, capabilities, publishers, papers, tools, organizations, and related models.
4. For papers, preview text with paper <slug> --text; use --output for full text or any PDF.
5. Summaries must include citations or source links from Sophon results and note when a claim was inferred from relationships rather than directly stated.

References

• Read references/api.md when you need endpoint details, CLI flags, JSON mode, or download safety behavior.
• Read references/research-workflow.md when producing comparisons, relationship maps, cited summaries, or paper-oriented research notes.