Custom review instructions
Give the AI review your team's conventions with review.customContext, and run your own checks with sandboxTools.
The AI review applies a generic checklist. To make it also enforce your team's rules, write them in review.customContext in the repository's .praxiom.yml.
Add instructions
version: 1
review:
customContext: |
All public methods have XML doc comments.
Handlers return Result<T>; flag any "throw new" used for an expected failure.
This repository targets net10.0; flag references to net48 or netstandard2.0.codingassist adds this text to the AI review prompt, after the diff and its context, as rules to enforce in addition to the generic checklist. Findings that come from these rules are posted like any other finding.
- The text is truncated to 2,000 characters.
- The instructions apply to the AI review only. The built-in static analyzers do not read them.
- The model applies them as instructions, not as exact pattern matches: write each rule as a sentence a reviewer would understand.
Run a deterministic check
When a rule must be checked exactly, run a tool that checks it with sandboxTools. The tool runs on the changed files in the analysis sandbox, and its findings are passed to the AI review with the other static findings.
sandboxTools:
- name: ruff
cmd: ruff check /workspace --output-format sarif --exit-zero
triggerGlobs: ["**/*.py"]
parser: sarifThe keys, the parsers and the limits of the sandbox are in the .praxiom.yml reference.