Bring Your Own Agent (BYOA)
How Critique's review artifact integrates with your existing AI coding agents — OpenAI Codex, Claude Code, and GitHub Copilot.
Critique is designed to work alongside the AI coding tools your team already uses. If you have an OpenAI Codex subscription, a Claude Code plan, or GitHub Copilot Enterprise, you don't need to replace them — Critique becomes the review and decision layer that tells your existing execution stack exactly what to fix.
BYOA is roadmap / handoff-first
BYOA is external-agent execution — it routes the fix blueprint to your own agent subscription (Codex, Claude Code, Copilot) for execution. This is distinct from Remedy, which is the built-in managed execution path running in E2B sandboxes and is live in beta today. In the current state, BYOA handoff is manual: use the Fix Prompt button on any review run page to generate a structured, grounded prompt you can paste directly into your agent. Automated API handoff is in active development.
BYOA vs. Remedy
| Remedy | BYOA | |
|---|---|---|
| What it is | Built-in managed execution path | External-agent execution handoff |
| Status | Live beta | Roadmap / manual handoff |
| Execution | E2B sandboxes, OpenCode, guarded PR pushback | Your agent subscription (Codex, Claude Code, Copilot) |
| Subscription | Critique credits | Your existing agent plan |
How BYOA works
Critique generates a structured fix blueprint from its review findings — a precise, grounded artifact that describes what is wrong, why, and what an acceptable fix looks like. That blueprint can be handed off to any capable coding agent for execution. Your agent subscription handles the actual code writing; Critique handles the analysis and verification.
The economic logic: spend aggressively on review depth, route execution to whatever agent plan you already pay for.
Supported execution targets
OpenAI Codex
Hand off the fix blueprint to Codex. Your OpenAI subscription handles the actual code writing — zero Critique credits spent on execution.
Critique produces the context-rich analysis (diff, repository structure, specialist findings, lead verdict). Codex receives a structured prompt and writes the patch. The separation keeps costs transparent and lets you use your existing OpenAI spend.
Claude Code
Route execution to Claude Code and let your Anthropic plan do the heavy lifting. Critique reviews, Claude fixes — two subscriptions, one seamless workflow.
Claude Code's extended thinking and agentic loop pair well with Critique's structured finding format. The lead verdict and per-finding remedy suggestions map directly to the kind of task description Claude Code handles best.
GitHub Copilot
Already on Copilot? Plug it in as the execution layer. Your Copilot subscription writes the patch while Critique stays focused on review quality.
This is particularly useful for teams that have standardised on GitHub Copilot Enterprise — Critique adds the deep-context review layer that Copilot's review features don't provide, while execution stays within your existing Microsoft agreement.
Economic advantage
BYOA teams avoid double-paying for capabilities they already have and keep the two concerns cleanly separated: what should change (Critique review) vs. how to change it (your agent).
At scale, BYOA teams typically run:
- Critique for review signal — model-flexible, risk-proportional, credit-based.
- Their existing agent for execution — billed through their existing subscription.
Current state and roadmap
In the current beta, Critique produces a review artifact you can read in the dashboard and copy as a fix prompt for your coding agent. The Fix Prompt button on any review run page generates a structured, grounded prompt you can paste directly into Codex, Claude Code, or Copilot.
Automated handoff — where Critique pushes the blueprint to your agent API without manual copy-paste — is in active development. The sequencing is deliberate: review quality must be reliable before execution is automated.
See Remedy for the built-in managed execution path that is live in beta today and doesn't require an external agent subscription.