TilloTech Docs Harbourmaster
Harbourmaster reviews GitHub pull requests with a trusted runner around an OpenCode multi-agent review pipeline. These pages describe how to operate the tool and how its review decisions are produced.
How-to guides
- How to request a Harbourmaster review: request a review and re-review the latest pull request head.
- Add Harbourmaster reviews to a repository: add the advisory GitHub Actions workflow.
- Diagnose a Harbourmaster review failure: recover from a skipped job, a failed check, or a missing review.
- Configure model routing: override model tiers through Action inputs or environment variables.
- Build a pilot dashboard: convert telemetry logfmt into a Markdown, HTML, or JSON report.
Explanation
- The Harbourmaster review pipeline: understand the runner, scoped OpenCode stages, and publication verification boundary.
- The shared-action trust boundary: understand why the workflow uses immutable references, same-repository pull requests, and exact-head checkouts.
Reference
- GitHub Action: action inputs, authentication, permissions, defaults, and runtime behavior.
- Command-line interface: GitHub command arguments, environment variables, output, and exit codes.
- Configuration: TypeScript defaults, model routes, OpenCode configuration, AWS settings, and precedence.
- Review behavior: diff filtering, risk tiers, reviewer selection, agent scope, publication verification, degraded runs, and failure classes.
- Telemetry: Effect logfmt events and field definitions.