Custom instructions are configured at the repository level only. There is
no organisation-wide custom instructions setting — each repository maintains
its own instructions independently.
Custom instructions
Custom instructions are guidelines you write for a specific repository. Garth reads them before reviewing every pull request and applies them as part of its analysis. Instructions can be written as plain text or Markdown. Markdown formatting — such as bullet lists, bold text, and headings — is fully supported and can help structure longer or more detailed instructions clearly. You can use custom instructions to:- Enforce team-specific coding conventions that aren’t covered by standard rules
- Tell Garth to prioritise certain areas — for example, “Pay close attention to authentication and authorisation logic”
- Give context about the codebase — for example, “This service is latency-sensitive; flag any blocking I/O in request handlers”
- Set the tone of comments — for example, “Be concise. Skip informational comments on trivial style issues”
- Exclude patterns that are intentional in this project — for example, “We use string concatenation for SQL in the legacy module — do not flag these”
Setting custom instructions
Open repository settings
In the Garth dashboard, navigate to Agent Setup → PR Review → Repository Setup and select the repository you want to customise.
Open the Custom instructions tab
Select the Custom instructions tab on the repository settings page.
Write your instructions
Enter your instructions in the text field. You can write in plain text or use Markdown formatting — bullet lists, bold text, and headings are all supported. Instructions can be as short as a few lines or as detailed as your team needs.
Writing effective instructions
Be specific, not generic
Vague instructions like “write good code” have no effect. Specific instructions like “flag any database query inside a loop” give Garth clear, actionable direction.
Provide context
Tell Garth about the purpose of the repository, the team’s priorities, or known trade-offs in the codebase. The more context you provide, the more relevant the review.
Iterate over time
Start with a few high-priority instructions and refine them based on the reviews you receive. Add more as you identify patterns in the feedback that need adjusting.
