Skip to main content

Gate vs stewardship

Codesteward is one platform with two review modes. Same specialists, policy, graph, and learning — different triggers and scopes.

Gate (PR merge check)​

  • Triggered by PR/MR webhooks, UI Gate, CLI, or GitHub Action
  • Plans units from the changed paths (diff packing)
  • Publishes inline comments, a check run, and optional SARIF to GitHub Code Scanning
  • Ends with a verdict (pass / fail based on severity floors and policy)

Best for: every merge into protected branches.

Stewardship (branch / path audit)​

  • Triggered from UI Steward, CLI stew steward, or scheduled jobs you wire
  • Scopes by path, package, or whole tree — not only a PR diff
  • Produces durable findings you track over time (status, reactions, reopen)

Best for: long-lived service branches, migrations, security sweeps, legacy packages.

Shared foundations​

  • Policy: STEWARD.md + .codesteward/rules from the base / default branch
  • Findings: one schema, fingerprinting, lifecycle
  • Learning: 👍/👎 and dismissals feed the next run
  • Graph: rebuild/query scoped to the org and session

See the UI guide for how both modes appear in the product, and How a review works for the agent stages.