Verify that a GitHub issue was implemented correctly — catch missing requirements, edge cases, and drift before code review.

TL;DR: A reusable Cursor/Claude Code command that pulls a GitHub issue, extracts requirements, and checks them one by one against your codebase. Outputs a structured report: done, missing, risky.

Why This Exists

When AI writes code from a spec, it often looks complete but isn't. The model implements 80% of the requirements and moves on. Tests pass, the code runs — but subtle things are missing. This command forces a systematic check before you (or another model) even starts reviewing.

It's designed to be the first step in a cross-model review workflow.

The Command

Create this as a custom command in Cursor (.cursor/commands/check-issue.md) or as a prompt file for Claude Code.

# Check Issue Execution

## Objective

Verify that the requirements from a GitHub Issue have been fully implemented.
Identify gaps and risks.

## Input

Issue number (e.g. `123`)

## Step 1: Fetch the Issue

```bash
gh issue view <ISSUE_NUMBER> --json number,title,body,labels,state,assignees,comments --jq '.'
```

Save the full issue body for analysis.

## Step 2: Extract Requirements

List requirements in 3 categories (where applicable):

1. **Functional** — what should it do
2. **Technical** — how should it be built
3. **UI/UX** — how should it look and behave

For each requirement, define 1–3 verification criteria.

## Step 3: Verify Against Code

For each requirement:

1. **Find the implementation** — use codebase search
2. **Check completeness** — does it match exactly what the issue describes
3. **Flag gaps** — what's missing, unclear, or only partially done

## Step 4: Quality Checklist

Check only what applies to this issue:

- Architecture consistency with `CLAUDE.md`
- Type safety and input validation
- Error handling and edge cases
- Authorization (if user/org data is involved)
- UI/UX and accessibility (if UI is involved)

## Step 5: Run Tooling

Only if relevant:

```bash
npm run lint
npm run typecheck
```

## Step 6: Output Report

Keep it short and concrete:

- ✅ **Implemented** — requirements that are fully done
- ❌ **Missing** — requirements not implemented or incomplete
- ⚠️ **Risks** — things that work but might cause problems
- ✨ **Recommendations** — optional improvements

Usage

In Cursor

Save the command as .cursor/commands/check-issue.md
Open the command palette and run it
Set the model to your preferred reviewer (we use Codex 5.3 for this — see Cross-Model Review)
Pass the issue number when prompted

Check Issue Execution 42

In Claude Code

Paste the command as a prompt, or save it as a custom slash command:

/check-issue 42

What It Catches

Category	Example
Missing requirements	Issue says "add pagination" — implementation only renders first page
Partial implementations	Error handling exists but only covers happy path
Spec drift	Model added a caching layer that wasn't in the spec
Type safety gaps	Input accepts `string` but issue implies it should be validated
Auth oversights	Endpoint returns data without checking user permissions

Tips

For large issues, tell the model to focus on the top 5 requirements rather than checking everything
Combine with lint/typecheck — the automated tools catch formatting and type issues, the command catches logic and completeness issues
Use the output as input — the report from this command is the perfect thing to feed into a second model for cross-model review

Cross-Model Code Review — use this command's output as the first step in a multi-model review pipeline

Issue Verification Command