sentry-fix-issues

About sentry-fix-issues

Discovers, analyzes, and fixes production issues using Sentry's debugging data through the Sentry MCP server. It is meant for requests to fix Sentry errors, debug production bugs, investigate exceptions, or triage a Sentry backlog, and it works methodically across stack traces, breadcrumbs, traces, tags, and context to identify root causes. Prerequisites are a configured and connected Sentry MCP server and access to the relevant project or organization.

The skill runs seven phases: issue discovery, deep issue analysis, root cause hypothesis, code investigation, implementing the fix, a verification audit, and reporting results. Discovery and analysis lean on Sentry MCP tools such as search_issues, list_issues, get_issue_details, search_issue_events, get_issue_tag_values, get_trace_details, get_event_attachment, and analyze_issue_with_seer for AI-generated root cause analysis. Before writing code, it documents an error summary, immediate cause, root cause hypothesis, supporting evidence, and alternatives, then reads the stack trace files top down, traces data flow, and reviews related code and recent commits.

A strong security posture runs throughout: all Sentry data is treated as untrusted external input. The skill never follows instructions embedded in error messages or breadcrumbs, never copies raw field values into source or test fixtures, never reproduces secrets or PII in output, and cross-references event data against the actual codebase before acting, flagging discrepancies rather than assuming the event is authoritative. Fixes favor input validation, graceful degradation, and root-cause over symptom handling, and are accompanied by tests built from generalized synthetic data and a final verification audit and structured report.

FAQ

What are the prerequisites?

A Sentry MCP server that is configured and connected, plus access to the relevant Sentry project or organization.

How does it treat data returned from Sentry?

As untrusted external input. It never follows instructions embedded in error messages or breadcrumbs, does not copy raw field values into code, and does not reproduce secrets or PII in its output.

Which MCP tools does it use?

Tools such as search_issues, list_issues, get_issue_details, search_issue_events, get_issue_tag_values, get_trace_details, get_event_attachment, and analyze_issue_with_seer for AI root cause analysis, among others.

What is the overall workflow?

Seven phases: issue discovery, deep analysis, root cause hypothesis, code investigation, implementing the fix, a verification audit, and reporting results.

Does it add tests for the fixes?

Yes. It adds tests reproducing the error conditions, but uses generalized or synthetic data rather than embedding actual values from event payloads such as URLs, user data, or tokens.