Skip to content

Provide Feedback

Effective bug reports for Cursor, Claude Code, and Codex lead with the diagnostic each vendor asks for: a Cursor Request ID, Claude Code’s /bug transcript or /debug log, and Codex’s codex --version plus config block. Feature requests, documentation fixes, and code contributions each route through their own official channel.

Your feedback and contributions help shape the future of AI-assisted development. Learn how to report issues, suggest features, contribute code, and help improve documentation.

  1. Search existing issues to avoid duplicates

  2. Verify the issue is reproducible

    • Try in a fresh environment
    • Disable extensions/plugins
    • Update to latest version
  3. Capture the diagnostics maintainers actually ask for

    • Cursor: grab a Request ID (Command Palette > “Report AI Action” or follow Getting a Request ID) plus the build from Help > About
    • Claude Code: run /debug to read the session debug log, then /status for version/model/account; attach /doctor output if it flags anything
    • Codex: paste codex --version and the relevant ~/.codex/config.toml block (model, approval_policy, sandbox_mode); type /status in the TUI to confirm the active model and writable roots
## Bug Description
[Clear, concise description of the issue]
## Steps to Reproduce
1. [First step]
2. [Second step]
3. [See error]
## Expected Behavior
[What should happen]
## Actual Behavior
[What actually happens]
## Environment
- Tool: [Cursor / Claude Code / Codex]
- Version: [e.g., Cursor 3.11, claude-code 2.1.207, @openai/codex 0.144.1]
- Model: [e.g., Opus 5, Sonnet 5, GPT-5.6 Sol]
- OS: [e.g., macOS 15.5]
- Diagnostics: [Cursor Request ID / Claude Code /debug log / codex --version + config]
## Additional Context
[Screenshots, logs, related issues]

Claude Code Bugs

  • In-tool (primary): /bug sends the conversation to Anthropic and can open a public GitHub issue
  • GitHub: anthropics/claude-code
  • Discord: #bug-reports channel

Codex Bugs

DO:

  • Explain the problem you’re trying to solve
  • Provide concrete use cases
  • Suggest implementation approaches
  • Link to similar features in other tools
  • Gauge community interest first

DON’T:

  • Demand immediate implementation
  • Duplicate existing requests
  • Bundle multiple features in one request
  • Assume your workflow is universal
## Feature Description
[What you want to add/change]
## Problem Statement
[What problem does this solve?]
## Use Cases
1. [Specific scenario 1]
2. [Specific scenario 2]
## Proposed Solution
[How might this work?]
## Alternatives Considered
[Other ways to solve this]
## Additional Context
[Mockups, examples, references]
  1. Official Channels

  2. Community Voting

    • Upvote existing requests
    • Comment with additional use cases
    • Share in social media for support

Fix Typos & Errors

  • Correct spelling/grammar
  • Fix broken links
  • Update outdated information
  • Clarify confusing sections

Add Examples

  • Real-world use cases
  • Code snippets
  • Screenshots/GIFs
  • Video tutorials

Translate Content

  • Translate to new languages
  • Review existing translations
  • Localize examples
  • Cultural adaptations

Create Tutorials

  • Step-by-step guides
  • Best practices
  • Workflow examples
  • Tips and tricks
  1. Fork the repository, then clone your fork

    Terminal window
    git clone https://github.com/<your-username>/cursor-and-claude-code-developer-toolkit
    cd cursor-and-claude-code-developer-toolkit
  2. Create a branch

    Terminal window
    git checkout -b docs/improve-setup-guide
  3. Make your changes

    • Follow style guide
    • Test all examples
    • Preview locally
  4. Submit pull request

    • Clear description
    • Reference issues
    • Screenshots if UI changes

Writing Style:

  • Clear and concise
  • Active voice
  • Present tense
  • Second person (“you”)

Technical Style:

  • Use code blocks with syntax highlighting
  • Include file paths and line numbers
  • Test all code examples
  • Provide both simple and advanced examples
  1. Read contributing guidelines

    • Check CONTRIBUTING.md
    • Review code style guide
    • Understand project structure
  2. Start small

    • Fix a bug
    • Improve tests
    • Enhance documentation
    • Then tackle features
  3. Communicate first

    • Discuss major changes
    • Get feedback on approach
    • Coordinate with maintainers

Good PR Checklist:

  • Descriptive title
  • Clear problem/solution explanation
  • Tests included
  • Documentation updated
  • Follows code style
  • Passes CI/CD checks
  • Addresses review feedback

PR Template Example:

## Description
[What does this PR do?]
## Type of Change
- [ ] Bug fix
- [ ] New feature
- [ ] Breaking change
- [ ] Documentation
## Testing
[How has this been tested?]
## Checklist
- [ ] Tests pass
- [ ] Documentation updated
- [ ] Code follows style guide

Bug reports get tools fixed; shared workflows get the whole community faster. When a prompt, rule file, or MCP setup saves you real time, write it up.

  • Post the concrete artifact, not the takeaway. “I added a .cursor/rules/ file that pins our Drizzle conventions” beats “Cursor is great for databases” — paste the rule, the CLAUDE.md block, or the ~/.codex/config.toml snippet.
  • Channels: the Cursor forum, Claude Code Discussions, Codex Discussions, and social (#CursorIDE, #ClaudeCode, #OpenAICodex).
  • In-tool quality signals also count: Claude Code’s “How is Claude doing this session?” rating and Codex’s / feedback feed directly back to the teams shipping these tools.

Cursor Beta

Settings > Beta > set update frequency to Early Access.

  • Ships pre-release builds first
  • Pair with Help > About to report the exact build

Claude Code Beta

Terminal window
claude update
  • claude update keeps you on the newest release
  • File regressions with /bug so the transcript is attached

Codex Beta

Terminal window
npm install -g @openai/codex@latest
  • Run codex --version after updating
  • Try pre-release surfaces in the App’s settings

Pre-release builds break in ways stable builds do not — that is the point, and a clean regression report is the most valuable thing you can file:

  1. Pin the version that broke and the last one that worked (cursor --version, claude --version, codex --version). A version delta lets maintainers bisect instantly.
  2. Capture the diagnostic before restarting — Cursor Request ID, Claude Code /debug log, or Codex /status — because a fresh session often loses the failing state.
  3. Run pre-release builds in a sandbox or a throwaway branch, never against --dangerously-skip-permissions on production code, so an experimental agent can’t trash real work.

This guide itself is open for contributions:

  1. Fork the repository

    Terminal window
    git clone https://github.com/<your-username>/cursor-and-claude-code-developer-toolkit
  2. Make improvements

    • Fix errors
    • Add examples
    • Enhance explanations
    • Update for new features
  3. Submit changes

    • Create pull request
    • Explain improvements
    • Link to sources

Current priorities:

  • More real-world examples
  • Video tutorial scripts
  • Translations
  • Advanced use cases
  • Troubleshooting scenarios

Active contributors receive:

  • Contributor badges on forums
  • Credits in release notes
  • Early access to features
  • Direct communication with teams
  • Conference tickets (top contributors)

Your contributions are permanently recorded:

  • Git history
  • Contributors page
  • Release notes
  • Annual reports

The difference between a report that gets fixed and one that gets closed as “cannot reproduce” is almost always the diagnostic payload, not the tone. Lead with the artifact the maintainer needs:

  • Cursor: a Request ID ties your report to the exact model call on Cursor’s side. Without it, “the agent hallucinated a file” is unactionable; with it, support can trace the request. Capture it via Command Palette > “Report AI Action”.
  • Claude Code: run /bug while the bad behavior is still on screen so the transcript is attached automatically, or paste the trailing lines of /debug output. Note whether you were on --dangerously-skip-permissions or a hook fired, since both change behavior.
  • Codex: state the surface (App / CLI / IDE / Cloud), the model from /status, and your approval_policy / sandbox_mode. A workspace-write vs read-only sandbox often explains “it refused to edit my file”.

Cursor:

Claude Code:

Codex:

  • Critical bugs: 24-48 hours
  • Feature requests: Reviewed monthly
  • Documentation: 1-2 weeks
  • General feedback: Best effort

Thank you for helping make AI-assisted development better for everyone. The most useful thing you can do is file one well-diagnosed bug with a Request ID, /bug transcript, or codex --version attached — that is what turns a frustrating session into a fix.

Frequently asked questions

What diagnostics does a Cursor, Claude Code, or Codex bug report need?

Cursor asks for a Request ID (Command Palette, "Report AI Action") plus the build from Help > About. Claude Code asks for /debug to read the session debug log and /status for version, model, and account, with /doctor output if it flags anything. Codex asks for codex --version, the relevant ~/.codex/config.toml block, and /status from the TUI.

Where do I report a bug for each tool?

Cursor is closed-source, so its forum bug reports category is the live tracker; Help → Report Issue attaches your Request ID, and urgent cases go to support@cursor.com. For Claude Code the primary path is /bug in-tool, which sends the conversation to Anthropic and can open a public GitHub issue. Codex bugs open as a new GitHub issue with the report template.

How do I write a feature request that gets traction?

Explain the problem being solved, provide concrete use cases, suggest implementation approaches, link to similar features in other tools, and gauge community interest first. Avoid demanding immediate implementation, duplicating existing requests, bundling multiple features into one request, or assuming your workflow is universal.

How long does feedback take to get a response?

Critical bugs are answered in 24-48 hours. Feature requests are reviewed monthly. Documentation changes take 1-2 weeks. General feedback is handled on a best-effort basis.

How do I contribute to this guide?

Fork the repository, clone your fork, and create a branch. Make changes that follow the style guide, test every example, and preview locally, then submit a pull request with a clear description, referenced issues, and screenshots for UI changes. Current priorities are more real-world examples, video tutorial scripts, translations, advanced use cases, and troubleshooting scenarios.