What is Steward
What is Steward?
Steward is question-based review for pull requests and pushes to main. Your team writes yes/no questions about what matters in your code, like “Does every new interactive element work by keyboard?”. Steward scores each question on every change, checks the low scores against the code, and posts findings, with file and line evidence, in the PR thread.
It flags first. Findings are comments, and only the questions you mark as hard limits can block.
Your questions, not a general opinion
A general reviewer reads the diff
A general AI reviewer applies one broad lens to each change: it summarizes, comments and suggests. That is useful, and it is the same lens on every codebase.
Steward asks the questions you wrote
You state what matters in your repo as yes/no questions, grouped by the context they share, such as Accessibility or Stripe. Every change is scored against them, so the review reflects your own standards, including ones no off-the-shelf tool could know.
How a Steward review works
One change at a time: score the questions, turn low scores into candidate concerns, check them against the code, and post what survives.
| Dimension | Steward | A general AI reviewer |
|---|---|---|
| Reviews against | Yes/no questions your team defines, grouped by context | A general sense of what is wrong in the diff |
| Triggers | Every pull request and every push to main | Varies by product |
| Handling noise | A low score is only a candidate concern; a check against the code drops false alarms before posting | Varies by product |
| Output | Findings with file and line evidence in the PR thread | Comments, summaries and suggestions |
| After posting | Concerns stay open until they are fixed or dismissed | Varies by product |
| Blocking | Flag-first; only hard-limit questions can block | Depends on how it is configured |
Findings in the PR thread.
Concerns land on the pull request itself, with the reasoning behind each. This example is a real PR review in an earlier comment format.
Steward FAQ
- What is Steward?
- Steward is question-based review for pull requests and pushes to main. Your team writes yes/no questions about what matters in your code, grouped by context such as Accessibility or Stripe. Steward scores each question on every change, checks low scores against the code, and posts findings with file and line evidence in the PR thread.
- How is Steward different from a general AI code reviewer?
- A general reviewer applies one broad lens to the diff. Steward reviews against the questions your team defines, so the review reflects your own standards. Concerns stay open in the PR thread until they are fixed or dismissed.
- Does Steward block merges?
- It is flag-first: findings are comments. Only questions you mark as hard limits can block.
- How does Steward avoid false alarms?
- A low score only makes a question a candidate concern. A second step checks each candidate against the code and drops false alarms before anything is posted.
- Does Steward review pushes to main as well as pull requests?
- Yes. Steward scores every pull request and every push to main against your questions.
- Can my coding agent install it?
- Yes. https://www.steward.foo/llms.txt points agents to /install, which is the only install path. The agent connects to the MCP server at https://mcp.steward.foo, and once the GitHub App is installed a human confirms a short code with one click.
Install Steward.
One GitHub App install and one human click to confirm a short code. Scoring starts in shadow mode, so nothing is posted to your PRs while it calibrates.