Skip to main content
Agent Personas are pre-built AI review agents (formerly “AI Rules”) that you turn on per repository. Each persona reviews your pull requests from a focused perspective - maintainability, performance, security, or design - extending CodeAnt’s default bug and security detection with a specialized reviewer.

​
Pre-built Agents

CodeAnt ships four agents. All are off by default, so you choose the perspectives that matter for each repository.
AgentWhat it flags
Maintainability AgentComplexity, duplication, naming, and coupling - suggests cleaner, more readable structure
Performance AgentInefficient loops, N+1 queries, memory leaks, costly allocations, and hot paths - recommends faster, leaner code
Pentesting AgentInjection, broken auth, exposed secrets, SSRF, and weak crypto - explains the attack and the fix
Design Critique AgentUI/UX quality, accessibility, visual hierarchy, spacing, responsiveness, and design-system consistency

​
Enabling an Agent

Navigate to your Dashboard settings:
  1. Go to app.codeant.ai/x/settings/ai-rules
  2. Under CodeAnt-defined Agents, find the agent you want
  3. Toggle the agent on
  4. Configure its options (see below) and click Save
Each agent shows a status of Active, Configuring, or Off, and a counter at the top tracks how many of the four agents are active for the repository.

​
Configuring an Agent

​
Reporting Threshold

Choose how selective the agent is. A lower threshold surfaces more findings (including minor nits); a higher threshold reports only the most serious issues:
ThresholdBehavior
LowExhaustive - flags everything, down to small nits and stylistic tweaks. Maximizes coverage.
MediumBalanced - reports findings that meaningfully matter, skipping trivial nits (default).
HighStrict - only serious, high-impact risks. Ignores minor or stylistic concerns.

​
Language & Framework Focus

By default, an agent reviews all languages and frameworks. Under Advanced options, you can narrow its focus to a specific set (for example, only Python and TypeScript) so the agent concentrates on the stack you care about.

​
File Scope

Restrict which files the agent reviews using glob patterns under Advanced options:
  • Include files - patterns for files to review (for example, src/**/*.py)
  • Exclude files - patterns to skip, prefixed with ! (for example, !src/tests/**)
If both are left empty, the agent reviews all files. Patterns use minimatch syntax - see the pattern examples on the Rules page for common globs.

​
How Personas Affect Reviews

Personas are configured per repository, not per pull request. Once enabled, an agent reviews every pull request in that repository, subject to its file scope. Enabling a persona adds its specialized guidance to the AI reviewer, so its findings appear inline alongside CodeAnt’s standard review. You can enable multiple personas on the same repository - for example, running the Performance and Pentesting agents together - and each contributes findings from its own perspective.

​
Custom Agents

The same settings page also lets you define your own review agents. Under Custom Agents, you can write repository-specific rules and instructions in your own words. See Rules for details on creating custom review rules.