From "this is broken" to a fix PR I merge by hand

Description and use

Mushi Mushi detects silent UI bugs, auto‑generates fix PRs, and requires manual merge, speeding up resolution of user‑reported issues.

Problem

Users reported UI bugs that did not trigger any errors in Sentry, leaving developers unaware of the issue.

Outcome

Mushi Mushi enabled detection of silent UI bugs, automated draft fix PR creation, and maintained manual merge control, reducing time to resolution and improving user confidence.

Workflow

• Capture user report with metadata and screenshot
• Filter spam with Claude Haiku 4.5
• Diagnose with Claude Sonnet 5.5 or GPT‑5.4 to classify and summarize
• Group similar reports using pgvector cosine similarity
• Generate auto‑fix PR via LLM find‑and‑replace edits
• Run CI on PR and display status
• Merge PR manually, marking report fixed and notifying reporter

Evidence

Sentry starts from what the code threw. Mushi starts from what the user felt.

Source

The fix worker loads whole files from the repo at a pinned commit and asks the model for find-and-replace edits through a typed schema.

Source

I kept the merge manual anyway.

Source