AI coding

Your coding assistantneeds a second opinion

Generation tools optimize for finishing the prompt. CodeCritic is a dedicated critique pass - in the editor, on paste, or on the PR - so the same model family is not both author and only auditor.

generator ≠ criticeditor + PRFix in ChatMark as done / Dismisssame engine everywhere

Separate the jobs

Let Copilot, Cursor Chat, or another assistant write. Run CodeCritic as an explicit review with severity, Policy findings, and a triage loop.

Stay where you write

Use the VS Code / Cursor extension for selection or file review, then Fix in Chat to hand a finding back to your coding assistant with a ready prompt.

Escalate to the PR

When the change leaves your laptop, the same engine can comment on GitHub so teammates see the second opinion without redoing your private pass.

A practical loop after AI writes code

Keep the critique short and explicit so it does not collapse back into the chat that wrote the patch.

  • Generate with your assistant, then run CodeCritic on the selection or file - not only ask the same chat “is this good?”
  • Triage: fix blockers, mark done what you fixed, dismiss false positives with a one-line reason.
  • Use Fix in Chat when you want the assistant to propose a patch from a finding, then re-run review if the change is non-trivial.
  • Open the PR only after the private pass; let Action or webhook carry the team-visible gate.

Workflow

How teams wire it

  1. 1

    Write with your AI assistant

    Use Copilot, Cursor, or another generator as you normally would.

  2. 2

    Run a CodeCritic pass

    Editor extension, New Review paste/upload, or wait for PR automation - same engine.

  3. 3

    Triage findings

    Block, fix now, mark as done, or dismiss with a reason - avoid silent mass dismiss.

  4. 4

    Hand off to humans when needed

    Security-sensitive and irreversible changes still need named human ownership on the merge.

Overview

Not the install page, not the philosophy page

The VS Code landing is about marketplace install and sidebar commands. AI versus human explains ownership and limits. This page is the workflow: when to request a second opinion after AI generation, and how to triage findings without rubber-stamping.

Limitations still apply - models miss context and invent explanations. Pair this loop with human review on sensitive surfaces.

FAQ

Common questions

The same session is biased toward defending its own output. A dedicated review pass with structured findings is easier to triage and share with teammates.

Add a critique pass to your AI coding loop

Start free, run one review on code your assistant just wrote, and triage before you open the PR.

Integrations