Model and Agent Routing

Pick the right model and agent mode by task class to control cost and quality.

CopilotCursorKilo
AuthorNeexoCore
Updated
costmodelsagents

Overview

Match tooling to task risk — do not use the heaviest model for every edit.

LOW tasks

  • Autocomplete or inline edit
  • Single-file bug fixes with clear scope
  • Cheapest model; no agent mode required

MEDIUM tasks

  • Chat or agent with explicit file list
  • One planning pass, then implementation
  • Mid-tier model; enable skills for planning or review

HIGH tasks

  • Dedicated planning skill before code changes
  • Consensus or multi-pass review for auth, schema, tenant isolation
  • Strongest model only for analysis and review — not for bulk scanning

IDE notes

  • Copilot: use agent mode with plugins; diagnostics to confirm instructions load
  • Cursor: subagent delegation for review; rules for always-on guardrails
  • Kilo: sticky models per agent; shared AGENTS.md for baseline context

Raw content

Copy into your project — e.g. .instructions.md, .agent.md, or SKILL.md

## Overview

Match tooling to task risk — do not use the heaviest model for every edit.

## LOW tasks

- Autocomplete or inline edit
- Single-file bug fixes with clear scope
- Cheapest model; no agent mode required

## MEDIUM tasks

- Chat or agent with explicit file list
- One planning pass, then implementation
- Mid-tier model; enable skills for planning or review

## HIGH tasks

- Dedicated planning skill before code changes
- Consensus or multi-pass review for auth, schema, tenant isolation
- Strongest model only for analysis and review — not for bulk scanning

## IDE notes

- **Copilot:** use agent mode with plugins; diagnostics to confirm instructions load
- **Cursor:** subagent delegation for review; rules for always-on guardrails
- **Kilo:** sticky models per agent; shared `AGENTS.md` for baseline context

Next steps

Matched by shared tags and category. Browse all cookbook