2.2 KiB
2026-02-28 Session Notes
Research Automation Pipeline Merge Planning
Nicholai developed a comprehensive merge plan to unify two divergent research automation codebases into a single dual-editor system. The chain/ pipeline (Claude Code native) and isaac-research-chain/ (Cursor native) both automate the same workflow: taking a biological mechanism and ingredient list as input and producing a ranked, cited evidence report via PubMed. The merge targets isaac-research-chain as the destination repository since it has existing git history.
Quality Analysis & Strategic Decisions
Prompt quality comparison revealed isaac's versions are significantly more detailed (2-3x larger), with better examples and field descriptions across all five prompts. All prompts will be retained from isaac's codebase; chain's terser versions are subsets of isaac's content.
Batching strategy analysis: Sequential batching (6 batches of 3 ingredients) produces more thorough per-ingredient analysis but takes 12-15 minutes for step 3, while parallel batching (3 agents of ~6 ingredients) completes in 5-7 minutes with slightly less per-ingredient focus. The merged system will default to sequential for quality with an optional --fast flag to enable parallel mode.
Orchestration will be centered on isaac's 253-line claude.md (renamed to CLAUDE.md for Claude Code convention) expanded to support batch mode selection and dual input sources. The system will maintain isaac's resume capability via status.json checkpoints.
File Organization & Architecture
The merge involves adapting chain's Claude Code architecture (.claude/ agents, skills, citation-standards rules) into isaac's isaac-research-chain structure. Workspace files will migrate from tmp/ to _workspace/. Client documentation will consolidate into a new docs/ subfolder. The system will support two input modes: input.md for direct single-mechanism input and samples.md (from chain) for multi-mechanism catalog selection (e.g., /research 1A).
Implementation involves 11 sequential steps: reorganizing documentation, copying static files, adapting all 5 research agents with path remapping, expanding the orchestrator, merging configuration files, and verification testing across both editors.