AgentLint

AgentLint

AgentLint is an automated auditing system that inspects agent configurations, plugin manifests, tool definitions, and system prompts for issues, violations, and best-practice gaps.

Overview

AgentLint runs 33+ checks across 4 categories:

Category	Checks	Description
Agent Config	7	Name length, description presence, system prompt validation, tool count, maxTurns range, provider validity, dangerous tool audit
Tool Definitions	5	Name length, description quality (action verbs, ambiguous phrasing), parameter count and descriptions, capability declarations
Plugin Manifests	3	Version presence, capability scoping, WASM runtime warnings
System Prompts	4	Instruction clarity, conflicting directives, token efficiency, unsafe patterns

CLI

cortex agentlint check             # Run checks on default agent config
cortex agentlint config            # Lint current agent configuration from config file

Output shows pass/warning/error counts with per-issue suggestions and severity badges.

REST API

Method	Path	Description
GET	/api/agentlint/check	Run checks and return results
POST	/api/agentlint/check	Run checks with custom config

Web UI

The AgentLint page (accessible from the sidebar) provides:

• Summary cards — total checks, passed, warnings, errors
• Color-coded issue cards — severity badges and actionable suggestions
• Run Checks button — on-demand re-audit

Example Output

AgentLint Results:
  Total checks: 33
  ✅ Passed: 28
  ⚠️ Warnings: 3
  ❌ Errors: 2

Issues:
  ⚠️ tool_desc_quality: agent-tool has generic description (avoid "useful", "helpful")
  ⚠️ system_prompt_length: system prompt exceeds 4000 tokens
  ❌ missing_tool: declared tool 'web_search_new' not found in registry

Module

src/agent/agentlint.ts — contains all check functions, severity scoring, and the runAgentLintChecks() orchestrator.

See Also

• Configuration — Agent and provider configuration
• Adding a Tool — Tool definition best practices
• Plugin System — Plugin manifest standards