Steward workflow
An operating model where an AI agent co-maintains an open-source project: 14 skills, four roles and autonomy bands
- 76 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent reads issues and pull requests and can write comments and labels in the repo if given access through `gh`
- On higher bands some actions run without a human, and a watchdog is required
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/stewardDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The agent reads issues and pull requests and can write comments and labels in the repo if given access through `gh`
- On higher bands some actions run without a human, and a watchdog is required
This is third-party code. Review the repository files before installing.
What it does
Steward is not an application but a set of playbooks, skills and processes you drop into any software project. The agent helps triage issues, review and merge pull requests, run quality gates and answer the community, while a human stays at every irreversible edge. The model rests on one idea: reading is cheap and safe, and a public write in the project voice needs a human decision unless it is reversible, mechanical and watched. The architecture has four roles, Watcher, Reviewer, Builder and Steward, and three human-involvement bands A, B and C. The lifecycle is described by area: issues, pull requests, quality gates, community and contributor recognition. Gates are layered: automated review, adversarial review, the test suite as a growing body of regressions and visual verification. The `skills/` folder holds 14 portable `SKILL.md` procedures: `issue-triage`, `pr-triage`, `pr-deep-review`, `release-pipeline`, `label-sync`, `contributor-trust`, `action-watchdog` and others. A setup interview gives the agent a question list about your repos, chat, CI and review tools and writes the project config. The docs are published on GitHub Pages.
Who it is for. Maintainers of open-source projects and teams that build an autonomous or semi-autonomous repo maintenance system around an AI agent.
Good fit when
- The queue of issues and pull requests grows and reviewing every diff eats evenings
- The agent needs a clear line between what it does alone and what goes to a human
- A release and its announcement follow the same steps and you want them as a skill
- Issues and pull requests from new contributors need trust rules
- The agent automated actions need a watchdog process
Not a fit when
- You need a ready application or bot: Steward stays a model and a set of procedures
- The project has no repo and no tracker: all skills assume issues and pull requests
- You are not ready to keep a human on irreversible steps such as merging code
Example request
Read the Steward overview, run the setup interview with me and triage an open issue with the issue-triage skill.Limitations
It needs an agent that reads `SKILL.md`-style procedures or a rules file, a repo with a tracker, an authenticated `gh` and a local steward-os checkout. The skills link to the checkout `setup/` and `docs/` by relative paths, so keep the checkout in place. Ruby is needed for `bin/config-lint`, which runs from the checkout. The quickstart switches on no autonomy, which is added later along the adoption ladder. The config is read from trunk or from outside the repo, never from a pull request checkout.
How to disable. Remove the copied skills from `.claude/skills/` and the steward-os checkout link from the agent rules file. The checkout itself can be deleted, and agent jobs and schedules are switched off separately.
FAQ
How do I add the Steward skills to Claude Code?
Run `mkdir -p .claude/skills` and `cp -R /path/to/steward-os/skills/* .claude/skills/` from the project root. The agent picks them up on the next launch.
Does it work outside Claude Code?
Yes. The skills are runtime-agnostic: put them where the agent loads skills or reference them from a rules file such as `AGENTS.md`.
What do bands A, B and C mean?
They are a human-involvement spectrum: the higher the band, the fewer actions the agent takes without consent. Details are in the architecture overview and the autonomy ladder.
What does the setup interview do?
The agent asks about your repos, chat, CI and review tools and writes the project config from the `setup/config.template.yaml` template.
Can it run unattended?
Only with an independent watchdog. Irreversible acts such as merging code or posting publicly as the project stay with a human.
Discussion
No questions or reviews yet.
Related
WorkflowsNine JS Mastery skills take a change through phases: from idea and spec to verification, tests, release texts and a docs sync
Agentic OS adds a five-step workflow and checks in git hooks and CI so that a step without evidence does not count as done
A .gaai/ folder that splits agent work into discovery and delivery: a story with criteria, then isolated implementation and QA.
Orchestrator skill: from an idea the agent writes requirements and a spec, splits work into tasks, builds the project, checks it
19 skills for UI, code and copy: the agent decides from project evidence and verifies results rather than making a template.
Builder Skills
Skills for agents
A set of small skills from Builder.io for agent work habits: visual plans and diff recaps, docs-first checks, cross-agent review and clean status updates
README in short
Covers the model idea, its parts, a quickstart with copying skills, the setup interview, lifecycle areas and the docs on adoption levels.