Agentic workflows, automation systems, and tools for research productivity.
Agent Skill / Coding Workflow
Light RIP is a lightweight Review-Implement-Plan skill package for Claude Code and Codex coding workflows. It defines a compact tiered process for coding tasks: tiny tasks stay in the main session, medium tasks add reviewer subagent review, and large tasks split planning, implementation, and review across subagents while keeping final fixes and verification coordinated by the main session.
Agent Skill / Notebook Runtime
agent-drylab is a Markdown notebook runtime for exploratory single-cell and public-bio analysis. It pairs an AI research agent with persistent tmux-backed Python, R, and Bash REPL sessions, dependency-aware cell graphs, and a local token-gated web view. Each cell is a first-class object with a run-state sidecar and a declared artifact manifest, so partial output never leaks across reruns, dependencies cascade explicitly, and long notebook runs stay auditable and resumable.