Steward on Steward

Steward runs on Steward. This page shows how: the config it uses, the pull requests it has shipped, and a plain split of what’s live today versus what’s still aspirational. Nothing here is a demo. If a capability isn’t running yet, this page says so, which is the same discipline the output-loop asks for everywhere else.

The config

This repo’s own setup/config.yaml is a filled instance of the template: one public repo, trunk main, continuous deploy (merge to main → GitHub Pages), the Jekyll build as the gate, no chat community, conservative autonomy bands, and no scheduled jobs enabled — because nothing runs unattended here yet.

What’s live today

  • Supervised sessions. Contributions and edits go through an agent working with a human, who approves every public write. This is Level 1, and it’s how the docs you’re reading were built.
  • Human-gated PR review. Every change lands via pull request with a human in the loop. See #1 (the output-loop playbook) and #3 (the minimalism ladder), both merged after human-gated PR lifecycle review.
  • The editorial loop. Discovery over delivery: work is proposed, reviewed, and trimmed in the open, exactly as the output-loop playbook prescribes.

What’s aspirational (not yet running)

  • Scheduled autonomy. No cron jobs, no unattended runs. Every scheduled_jobs entry is enabled: false.
  • Chat monitoring / mentions sweep. No community platform is wired; community.chat.platform is empty.
  • Autonomous public writes. No credentials file is configured; the agent never posts as the project unsupervised.

That gap is deliberate. Steward is meant to be adopted one rung at a time, and this repo sits near the bottom of its own adoption ladder, which is a reasonable place for a careful project to start.


Back to top

This site documents Steward — an operating model for AI-assisted project maintenance. MIT licensed.