Concepts
When a user feels something break, Mushi captures it, explains it in plain English, and can hand your editor a fix — without another dashboard to babysit.The loop in plain terms:
- Capture — users report what felt broken (shake, tap, or widget) with screenshot and context.
- Plain read — severity, category, and cause in English your team already uses.
- Connect — similar reports collapse to one row; hot spots emerge from the graph.
- Fix — optional agent drafts a PR; you merge, edit, or ignore.
- Verify — automated replay checks the fix before you ship.
- Remember — past fixes become rules your editor sees on the next PR.
Primary paths
- Architecture — components and data flow
- How classification works — Stage 1 / Stage 2
- How fixes get drafted — sandbox → draft PR
Everything else
Knowledge graph · Inventory & gates · Orchestrator hooks · Quality scoring · Anti-gaming & reputation · Trigger modes · Rewards · Closed-loop evolution · Credentials · Runtime config
Last updated on