← @bgSponge

GitHub - githubnext/chopin: Let's compose rich plans together

Summary

GitHub Next's Chopin prototype uses a hosted agent named Planner to co-author durable documents like implementation plans and RFCs alongside human teams.

  1. Planner co-authors via repository inspection.

    The agent uses read-only tools to inspect GitHub repositories and pull requests, proposing changes while leaving final decisions to the human team.

    People shape the document and own its decisions.

  2. Decisions remain decoupled from prose.

    Chopin stores attributed questionnaire answers and accepted comments separately so that subsequent document rewrites cannot silently alter established team decisions.

    Decisions retain attributed questionnaire answers and accepted comments separately from the prose they produced

  3. Research requests create durable child documents.

    Using the `/research` command generates an inline progress card and eventually publishes a completed research document with its own chat and decisions.

    Completed research publishes as an ordinary child document beneath its parent

  4. Agent capabilities depend on the harness.

    The default copilot-sdk harness prevents the agent from writing to GitHub, whereas the HARNESS=atomic setting grants full shell and filesystem access.

    Under the default copilot-sdk harness it cannot write to GitHub

  5. Scaling is limited to single processes.

    The architecture does not support horizontal application scaling, as only one Chopin process may write to the PostgreSQL database at a time.

    One Chopin process may write to a database at a time. Horizontal application scaling is not supported.

Read the original ↗Save to your Sponge

Summarized with Sponge. Quotes are the author’s own words.

More from @bg

All readings →