How do I find sessions that hit errors in Cursor (beta)?

Rius shows each Cursor (beta) chat as one trace, and a chat that hit an error has the status failed in the trace list. Filter the list to the cursor agent and the failed status, then open the trace to see the failing step highlighted in the waterfall. The plugin records the span tree, tool names and error types, and it marks a shell command that exits non-zero as an error, which is what makes these chats findable. The Coding agents guides cover setup.

How it works

• One chat, one trace. The plugin for Cursor (beta) turns each agent chat into a trace under an agent named cursor. Spans arrive while you work: when the chat starts, after each answer and every 20 tool calls.
• Status comes from the spans. A trace is failed when at least one span carries an OpenTelemetry ERROR status. Your agent does not report the status. Rius derives it.
• What marks an error. A tool call is a span of its own, and a shell command that exits non-zero is an error on that span. Tool names, error types and subagents are recorded even in the default structure-only mode.
• Recovered runs. If a child span errored but the chat itself finished clean, the trace is still failed but carries a recovered badge. The errors are still counted.
• When a chat closes. A chat’s trace closes when Cursor ends the session. If it never says so, your next chat closes it once it has been idle for 12 hours. The trace list shows a chat while it runs. The Users and Agents figures count it once its trace is closed.
• No tokens or cost. Traces from Cursor (beta) carry no token counts and no cost, because Cursor does not give plugins usage. You find errors by status, tool, time and developer, not by spend.

Set it up

1. Install the plugin, sign in and turn tracing on for the folder you work in. Tracing is off until you enable a folder. The full steps are on the Cursor (beta) guide.

   /rius-login
   /rius-enable-here
   /rius-status

2. Work as usual, then open Traces in the console. The time range defaults to 7d. Click a cursor cell in the agent column to filter to these chats, or click Add filter to build the filter yourself.

3. Set the status to failed. Filters combine with AND, and they live in the URL, so you can copy the link for a teammate. To keep the search, use Save current search… from the Saved menu.

4. Narrow by person. From plugin 0.6.2, each span carries user.id, set to the email you signed in with, so the User column and the Users page group chats by developer. The Sent by column shows whose key sent each trace. Chats sent with a key stored by use-key carry no user and land in the anonymous row. The team usage guide shows how the Users page lists failed traces per developer for Claude Code, and the same columns apply here.

5. Open a failed trace. The waterfall highlights the path from the root to the error, and the error span shows its tool name and error type. Press ]e to jump to the next failed span, or [e to go back.

6. To find the failing tool across many chats, open the Spans tab, choose Tool in the All kinds menu and filter by error status. Exploring traces covers filters, saved searches and the keyboard shortcuts in full.

7. Or ask in plain language. After a second sign-in to the Rius MCP server (click Connect next to rius in Cursor’s MCP settings, or run cursor-agent mcp login rius), ask “Show me my last failed Cursor chat and which tool call went wrong.”

What you get

• A list of failed chats, filtered by agent, developer, time range or sender, that you can save and share as a link.
• The exact step that failed: the error span, its tool name and its error type, shown inside the chat’s full span tree with timing and model names.
• A way to tell a real failure from a recovered one, so a retried tool call does not look like a lost chat.
• With /rius-enable-content-here, the prompts, answers, and tool inputs and outputs on the failed span, with recognised secrets removed. Redaction is best effort. Structure-only mode, the default, sends none of this.
• A single view for the whole team, where each developer’s failed chats are grouped under their email.

When something else fits

• If you need token counts or cost for coding sessions, Claude Code and Codex (beta) report them, and Cursor (beta) does not.
• Cursor’s cloud and background agents and Tab completions are not traced, so errors there will not appear in Rius.
• Tool calls in an interactive cursor-agent session and the agent in the Cursor editor are not tested yet, so check /rius-status and the trace list before you rely on them.

Related