Bug reports & AI diagnostics
Two paths
Section titled “Two paths”| Path | When | Your action |
|---|---|---|
| Automatic crash recovery | Unhandled JavaScript error or unhandled promise rejection | A toast may offer to submit a bug report after the incident is captured. |
| AI panel — Report this error | The AI panel shows a hard error after a failed prompt or runtime issue | Click Report this error below the message (only for real errors, not soft “no changes” warnings). |
Both flows require you to be signed in so the report can be associated with your session for support. If you are not signed in, the app will tell you to sign in first.
What goes into an AI error report
Section titled “What goes into an AI error report”When you use Report this error, the app sends a sanitized bundle to the server. It is designed to help reproduce issues without storing full file contents or secrets.
Typical fields include:
- Error text shown in the panel and severity.
- Last submitted prompt (and a short list of recent prompts in the conversation).
- Selected model and agent (if any), view state, and conversation slot id.
- Activity stream tail — recent assistant/tool/status lines from the run.
- Thinking snippet (if present).
- Terminal tail — last lines from the in-app terminal log.
- Workspace path and currently open file path (paths only, not full file bodies).
- Runtime address, connected flag, and MCP server names reported by the runtime.
- Pending diff file paths (if a diff was in progress).
- Breadcrumbs — recent app events the reporter recorded (auth, workspace, deploy, network, etc.).
The server stores a text record for support review. This is separate from telemetry usage events; bug reports are explicit, user-initiated (or crash-offered) submissions.
Privacy and security
Section titled “Privacy and security”- Do not paste API keys or secrets into the AI panel expecting them to stay local; if they appear in errors or logs, they could be included in truncated form. Prefer env-backed configuration.
- Paths and messages are length-limited on the client and sanitized again on the server.
Related
Section titled “Related”- AI edits — Model picker, agents, and session behavior.
- Agent runtime integration — Sessions, SSE, structured
@plan, and reliability notes. - Troubleshooting — Common AI and connection issues.