This document explains how to provide feedback, report bugs, request features, and get help with Claude Code. It covers the built-in feedback mechanisms (CLI commands), direct GitHub issue filing, security vulnerability reporting, and community support channels.
For information about how reported issues are processed by automated systems, see Issue Triage System and Issue Deduplication System. For details on data collection and analytics, see Event Logging & Analytics.
Claude Code provides multiple channels for users to report issues and provide feedback, each optimized for different use cases:
| Channel | Use Case | Access Method | Response Time |
|---|---|---|---|
/bug command | Bug reports from within sessions | CLI command | Automated GitHub issue creation |
/feedback command | General feedback | CLI command | Asynchronous submission |
| GitHub Issues | Detailed bug reports, feature requests | Web interface | Community/team triage |
| HackerOne | Security vulnerabilities | Web submission form | Security team review |
| Discord | Community support, discussions | Invitation link | Community-driven |
Sources: README.md52-58 SECURITY.md1-13
/bug CommandThe /bug slash command allows users to report issues directly from within a Claude Code session without leaving the terminal. This command captures context about the current session and generates a pre-populated GitHub issue.
Diagram: Bug Reporting Flow via /bug Command
The /bug command follows the standard slash command pattern in Claude Code's command system. When invoked:
Known Limitations:
Sources: README.md54 CHANGELOG.md246 CHANGELOG.md460
Direct GitHub issue filing is recommended for:
Diagram: GitHub Issue Processing Pipeline
When an issue is filed on GitHub, several automated workflows process it:
| Workflow | File | Purpose | Timing |
|---|---|---|---|
| Issue Triage | claude-issue-triage.yml | Assign labels based on content | On issue creation |
| Duplicate Detection | claude-dedupe-issues.yml | Identify similar existing issues | On issue creation |
| Auto-close Duplicates | auto-close-duplicates.ts | Close confirmed duplicates | Daily cron (3 days after marking) |
| Stale Management | stale-issue-manager.yml | Handle inactive issues | 30/60 day thresholds |
These workflows use Claude agents to analyze issue content and apply appropriate labels and actions. For detailed information about these systems, see Issue Triage System and Issue Deduplication System.
Sources: README.md54 .github/workflows/claude-issue-triage.yml .github/workflows/claude-dedupe-issues.yml
Important: Security vulnerabilities must NOT be reported through public GitHub issues or Discord. Use the dedicated security reporting channel.
Diagram: Security Vulnerability Reporting Flow
Anthropic's security program is managed through HackerOne:
What Qualifies as a Security Issue:
Good Faith Reporting: Security researchers should act in good faith and follow responsible disclosure practices. Do not exploit vulnerabilities beyond demonstrating proof of concept.
Sources: SECURITY.md1-13
The official Discord community provides:
Join: https://anthropic.com/discord
Diagram: Choosing the Right Feedback Channel
Sources: README.md56-58
/feedbackThe /feedback command provides a streamlined way to submit general feedback, suggestions, or comments that may not constitute bugs:
Command Usage:
/feedback
Features:
Recent Improvements:
Sources: CHANGELOG.md246 CHANGELOG.md460 CHANGELOG.md314
When submitting feedback or bug reports through CLI commands, certain information is automatically collected to aid in diagnosis:
| Information Type | Purpose | User Control |
|---|---|---|
| Session ID | Identify specific conversation | Automatic |
| Model version | Reproduce environment | Automatic |
| Error messages | Debug failures | Included in logs |
| Command history | Understand issue context | Not shared by default |
| File contents | N/A | Never automatically shared |
Diagram: Data Collection and Privacy Flow
Users can configure the "Help improve Claude" setting to control telemetry collection. Bug reports always include session context regardless of this setting, as it's necessary for diagnosis.
Sensitive Data Protection:
For comprehensive details, see:
Sources: README.md60-72 CHANGELOG.md390
Minimal Effective Report:
Enhanced Report (for complex issues):
~/.claude.json excerpts)/doctor command outputSources: CHANGELOG.md81 (example of fixed bug)
Claude Code logs certain events to Statsig for product analytics:
MCP Tool Names: User-specific MCP server configurations are sanitized to prevent exposing tool names in analytics events.
For detailed information about analytics infrastructure, see Event Logging & Analytics.
Sources: CHANGELOG.md370
| Need | Action | Command/Link |
|---|---|---|
| Report a bug | Use /bug command | /bug |
| General feedback | Use /feedback command | /feedback |
| Detailed issue | File GitHub issue | GitHub Issues |
| Security vulnerability | Submit to HackerOne | HackerOne |
| Get help | Join Discord | Discord |
| Check system health | Run diagnostics | /doctor |
Sources: README.md52-72 SECURITY.md1-13
Refresh this wiki