Configure reviews, agents, and skills
Choose review providers and models, shape specialist agents, and control the context used for a repository.
Configure each repository from Settings → Review and Settings → Agents & Skills. Global defaults are the starting point; repository settings can override them, and a review can still use an explicit per-run choice where the UI offers one.
Choose a provider and model
Kensa detects the agent CLIs installed on your machine and checks whether each one is ready to use. Select an available provider, then choose a model from its picker or a provider-supported custom value. A saved model pin applies only to its matching provider; it is not reused for a different CLI.
A multi-provider review runs the selected providers in parallel and then deduplicates overlapping findings. Each finding retains its agent, provider, and model attribution. Agreement can increase confidence, but it is evidence to inspect—not a guarantee that a finding is correct. Providers and models do not have identical capabilities, model catalogs, authentication methods, or costs.
Shape the review
Use the repository's Review settings to control the review recipe and its context:
- Pick or override the review providers and the primary model.
- Set repository path filters so excluded changed files do not reach review agents.
- Choose which repository guideline-file globs are included as review context.
- Set separate provider/model choices for Learn, the finding consolidator, or the Codebase Index only when that repository needs a different trade-off.
- Adjust reliability options such as turn and retry behavior through the settings controls.
For a one-off local diff or pull request, choose the review depth in the launch flow. See Run your first PR review for the difference between Quick and Deep Review.
Configure specialist agents and skills
Agents & Skills keeps the two editors separate so you can focus on one choice at a time. An agent has a focus, keywords, file patterns, checklist, repository context, and an enabled state. Start from a preset when it fits, then adapt the specialist mix to the repository.
Skills are reusable instructions discovered from repository and global skill locations. Kensa can select them manually or use auto-selection to score which agents and skills fit a pull request. You can pin a skill, boost it, or exclude it for a repository. Skills describe how to work; active Repository Learnings describe repo-specific facts and decisions that reviews should honor.