Add Claude Code skills for orchestrating the AI Scientist v2 pipeline#93
Open
strelkon wants to merge 7 commits intoSakanaAI:mainfrom
Open
Add Claude Code skills for orchestrating the AI Scientist v2 pipeline#93strelkon wants to merge 7 commits intoSakanaAI:mainfrom
strelkon wants to merge 7 commits intoSakanaAI:mainfrom
Conversation
Convert the AI Scientist v2 research pipeline into 5 Claude Code skills (slash commands) that can be used within Claude Code or Claude Co-Work: - /ai-scientist: Full end-to-end pipeline - /ai-scientist-ideate: Research idea generation - /ai-scientist-experiment: BFTS experiment execution - /ai-scientist-writeup: Paper generation from results - /ai-scientist-review: AI peer review of papers Also adds CLAUDE.md with project context and skill usage documentation. https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R
Rewrite skills so that ideation, writeup, and review stages use Claude Code directly (web search for citations, native PDF reading for review, direct LaTeX generation for writeup) instead of calling external LLM APIs. Only the experiment stage (BFTS tree search) still requires API keys since it performs parallel LLM-driven code generation that can't run in a single Claude Code session. https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R
…ills Replace WebSearch-only approach with Semantic Scholar API calls via the existing Python tool. Falls back to WebSearch if S2_API_KEY is not set or the API is unavailable. https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R
Rewrite the experiment skill so Claude Code itself acts as the code generator, executor, and analyzer — replacing the external BFTS engine that required Bedrock/OpenAI API keys. Implements the same 4-stage workflow (initial implementation → hyperparameter tuning → creative research → ablation studies) with compatible output format. All 5 pipeline stages now run without external LLM API keys. Only S2_API_KEY (Semantic Scholar) is recommended for literature search. https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R
Create a distributable plugin at plugin/ with:
- Plugin manifest (.claude-plugin/plugin.json) with user config for
AI_SCIENTIST_ROOT and S2_API_KEY
- 5 namespaced skills: ideate, experiment, writeup, review, pipeline
(invoked as /ai-scientist:ideate, /ai-scientist:experiment, etc.)
- Marketplace manifest for installation from the GitHub repo
- Plugin README with setup and usage instructions
Skills use ${AI_SCIENTIST_ROOT} instead of hardcoded paths, making the
plugin portable across any machine with the AI-Scientist-v2 repo.
https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R
- Add required `type` and `title` fields to userConfig entries
- Remove unsupported `env` field
- Use `directory` type for PROJECT_DIR
- Replace ${AI_SCIENTIST_ROOT} with ${user_config.PROJECT_DIR} in all
plugin skills (correct plugin substitution syntax)
- Pass S2_API_KEY from user config to Python subprocess env
https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R
With disable-model-invocation: true, skills were invisible to Claude — it had no knowledge they existed unless the user typed the exact slash command. Setting it to false allows Claude to see skill descriptions in its context and invoke them when relevant, which is essential for Claude Co-Work where users describe tasks in natural language. https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Convert the AI Scientist v2 research pipeline into 5 Claude Code skills (slash commands) that can be used within Claude Code or Claude Co-Work:
Also adds CLAUDE.md with project context and skill usage documentation.
https://claude.ai/code/session_01NDGeLAQMLY7zFAg2yvYE9R