Sentry Issue Hunter

I need you to help me systematically hunt and fix Sentry issues.

Prerequisites: Check if you have access to Sentry MCP tools by looking for any tools containing "Sentry" in their name (e.g., mcp__Sentry__find_issues, mcp__Sentry__get_issue_details). If you cannot find these tools in your available tools list, tell the user to install the Sentry MCP with the following command:

claude mcp add-json Sentry '{"command": "npx", "args": ["mcp-remote@latest", "https://mcp.sentry.dev/sse"] }'

Sometimes it takes a moment for the Claud code to connect to Sentry, so if the user believes they have already installed the tooling, then ask them to wait, and check with /mcp for connection to Sentry before running this custom command.

Think hard and follow this workflow:

Step 1: Find Issues

Use the Sentry MCP tool mcp__Sentry__find_issues to find the top 20 unresolved issues sorted by occurrence count:

• Call it with organizationSlug="compiler-explorer", query="is:unresolved", sortBy="count"
• Present a numbered list of the most impactful issues

Step 2: Issue Selection

Let me choose which issue to investigate, or if $ARGUMENTS is provided, investigate that specific issue ID.

Step 3: Investigation Setup

• Fetch detailed issue information including stack traces using mcp__Sentry__get_issue_details
• Create a new branch from origin/main with format: claude/fix-sentry-ISSUE-ID
• Use TodoWrite to track progress on this investigation

Step 4: Root Cause Analysis

• Analyze the stack trace to identify the problematic code location
• Use search tools (Grep, Glob, Read) to examine the relevant files
• Think hard about the root cause - don't just paper over symptoms
• Use the Sentry MCP to gather multiple occurrences and ensure the root cause explains all (or most) instances

Step 5: Implementation

• Follow CLAUDE.md guidelines for defensive programming and code style
• Ensure errors surface in Sentry rather than failing silently
• Use centralized helper functions rather than spreading type checks

Step 6: Quality Assurance

• Run npm run ts-check to verify TypeScript compliance
• Run npm run test-min to ensure no regressions
• Follow pre-commit workflow (never use --no-verify)

Step 7: Delivery

• Commit with descriptive message referencing the Sentry issue ID
• Push branch and create PR with:
  • Clear title referencing the issue
  • Detailed description explaining root cause and fix
  • Test plan for verification

Step 8: Iteration

• Mark todo as completed
• Ready to hunt the next issue!

Focus on robust solutions that prevent the error category, not just this specific instance.