↓ Skip to main content
  1. Agents/

Agents queue

Author
Tom Rochette

Work items the owner has requested, taken in order, highest first. The owner edits this file to steer the section; agents append ? for tom: bullets and mark items done, nothing else.

  1. Build the research index (standing) - create research notes per the AGENTS.md format, one category at a time, each category seeded with the tools practitioners actually meet. Seed categories, in rough order:
    • Harnesses: Claude Code, Codex, Gemini CLI/Antigravity, OpenCode, aider, Crush, Amp, Junie.
    • Surfaces: Cursor, VS Code + Copilot, Windsurf, JetBrains/Junie, Zed.
    • Orchestration: parallel-agent dashboards and worktree managers (Conductor, dmux, Emdash, and neighbors).
    • Protocols: MCP, ACP, AGENTS.md, A2A.
    • Context engines: code search and context packing for agents.
    • Skills: skill formats, registries, and marketplaces.
    • Retrieval: RAG patterns and tooling for code and knowledge bases.
    • Memory: agent memory systems, from file-based notes to knowledge graphs and session archives.
    • Executions: scheduled and event-based agent runs (cron-style tasks, triggers, webhooks).
    • Hybrid execution: deterministic components orchestrating LLM steps for intelligence and structured outputs (tool calling, schema/JSON outputs, validation and retry loops).
    • People and publications: the people and websites shaping the domain (Karpathy, swyx/Latent Space, Willison, Yegge, Orosz/Pragmatic Engineer).

Done: