add-research-link

name

add-research-link

description

Add a new AI-for-research project or resource URL to this repository's README.md and README_zh.md, including a clickable auto-updated Star count for GitHub repositories. Use when the user provides a URL or asks to add a link, resource, or project to the awesome-AI-for-research list and expects source-grounded bilingual descriptions, correct categorization, synchronized README entries, and GitHub Star metadata.

Add Research Link

Workflow

1. Confirm you are in the awesome-AI-for-research repository root and that both README.md and README_zh.md exist.
2. Treat a bare URL from the user as a request to add that link.
3. Fetch the linked source before writing:
   • For GitHub repositories, use the GitHub API or page metadata and prefer the repository description plus README summary.
   • For project pages, papers, blogs, benchmarks, or workshops, use the page title, abstract/intro, and meta description.
   • If a project page links to one clearly official GitHub repository, use that canonical repository URL for the README entry. Ignore dependencies, examples, website-template sources, and unrelated outbound repositories. Keep the project-page URL when ownership is ambiguous or no official repository is linked.
   • If the URL is unreachable, search the exact title or repo path before falling back to a cautious description.
4. Derive:
   • name_en: concise official English display name.
   • name_zh: normally the same official product/project name; translate only generic descriptive titles.
   • desc_en: one short English sentence, no hype, ideally under 28 words.
   • desc_zh: faithful Chinese version of desc_en.
   • section_en and section_zh: matching existing sections from the map below.
5. Insert the entry with scripts/insert_readme_entry.py. For a GitHub repository URL, the script must add the clickable <!--stars:OWNER/REPO--> marker automatically.
6. Refresh the new Star count with python scripts/update_stars.py. If network access is unavailable, keep ⭐ updating; the scheduled GitHub Action will resolve it after merge.
7. Run validation:
   • git diff --check
   • Confirm each README gained exactly one numbered entry and that the new English and Chinese entries describe the same resource.
   • Ensure README.md has no Chinese description text unless it is part of an official name.

Category Map

Use one of these exact section pairs:

English section	Chinese section	Use for
API-based	基于 API	General AI scientist systems that primarily run through hosted/API model workflows.
LLMs combined with genetic and search algorithms	遗传算法和搜索算法结合 LLM	AlphaEvolve, evolutionary search, population search, program evolution, optimizer/search systems.
Research systems based on Claude Code or Codex	基于 Claude Code 或 Codex 的科研系统	Research systems built directly around Claude Code, Codex, coding-agent CLIs, or repo-local agent workflows.
Skills	Skills	Skill libraries, tool libraries, scientific databases, agent tools, and reusable capability packs.
Heuristic Learning using Claude Code or Codex as optimizer	Heuristic Learning using Claude Code or Codex as optimizer	Heuristic learning, gradient-free code optimization, executable policies/world models, and coding-agent optimizer examples.
Autoresearch related	Autoresearch related	Autonomous research loops, overnight experiment runners, self-organizing research agents, and multi-agent research execution.
AI4LLM	AI4LLM	Systems specifically improving LLM training, model development, or ML engineering with research agents.
RSI (recursive self improvement) / Harness Engineering	RSI (recursive self improvement) / Harness Engineering	Agent harness design, evaluation harnesses, runtime substrates, recursive self-improvement, and workflow scaffolds.
Infrastructure for agent-friendly research	智能体友好型科研基础设施	Corpora, APIs, preprint platforms, artifacts, structured scientific memory, and agent-native research infrastructure.
Benchmarks for research agents	科研智能体评测基准	Benchmarks, arenas, leaderboards, evaluation suites, and task collections for research agents.
Paper collections	论文合集	Workshops, accepted-paper lists, paper collections, and reading lists.
xCode series	xCode 系列	Coding-agent products or CLI agents such as OpenCode, Kimi Code, MiMo Code.
Related projects and resources	相关项目与资源	Awesome lists, curated resource collections, and adjacent indexes.

If classification is genuinely uncertain, choose the closest existing section, keep the description factual, and mention the ambiguity in the handoff. Do not invent a missing catch-all section.

Insert Command

Run from the repository root:

python .agents\skills\add-research-link\scripts\insert_readme_entry.py `
  --section-en "Benchmarks for research agents" `
  --section-zh "科研智能体评测基准" `
  --name-en "ExampleBench" `
  --name-zh "ExampleBench" `
  --url "https://example.com/examplebench" `
  --desc-en "A benchmark for evaluating research agents on example scientific tasks." `
  --desc-zh "用于评估科研智能体处理示例科学任务能力的基准。"

Use --dry-run first when the section choice or generated text is uncertain.

GitHub Star Counts

• Add a Star marker only when the resource URL identifies a GitHub repository.

• Use this exact linked form after the project link:

  [<!--stars:OWNER/REPO-->⭐&nbsp;updating<!--/stars-->](https://github.com/OWNER/REPO)
  

• Keep the marker in both READMEs with the same case-sensitive OWNER/REPO value.

• Never add Star metadata to project pages, papers, blogs, or non-GitHub resources.

• Never hand-edit a resolved count. scripts/update_stars.py and .github/workflows/update-stars.yml own the displayed number.

• Treat ⭐ updating as a temporary fallback, not a fabricated count.

Rules

• Modify only README.md and README_zh.md for ordinary link additions; the shared Star updater and workflow are repository infrastructure, not per-entry files.
• Do not create a new category unless the user explicitly asks for taxonomy changes.
• Preserve existing order and append to the end of the chosen section.
• Do not duplicate an existing URL. If the URL is already present, report the existing location instead of adding another entry.
• Keep English and Chinese entries aligned: same category, same project, matching meaning.
• Keep descriptions brief and source-grounded. Avoid unsupported claims such as "best", "first", or "SOTA" unless the linked source explicitly says so.