How do I find sessions that hit errors in Claude Code?
Rius shows each Claude Code session as one trace, so you find sessions that hit errors by filtering the trace list to the status failed. A failed tool call is marked as an error on its span, and the Users page counts failed sessions per developer. You can also ask Claude Code itself to show your last failed session. For the wider picture, see Coding agents.
How it works
One Claude Code session is one trace, with the same session id throughout. The trace holds the session, each turn, each model call, each tool call and each subagent.
- Status comes from the spans. A trace is
failedwhen at least one span carries an OpenTelemetryERRORstatus. A Bash, Read or Edit call that fails is marked as an error on its own span, so the session surfaces without anyone flagging it. - Recovered sessions are labeled. If a tool call failed but the session itself finished clean, the trace is badged recovered. The errors are still counted. Only the verdict changes.
- Every enabled folder records the clues. Even in structure-only mode, the plugin sends tool names, error types, subagent types, and the kind of command a shell call ran (
test,build,lint,package,gitorother) with its exit code. That is enough to tell a failing test run from a failing file read. - Content mode adds the details. A folder enabled with
/rius:enable-content-herealso sends the error details of a failed tool call, with recognized secrets removed. A folder enabled with/rius:enable-hereleaves those details out.
The trace list shows a session while it runs. A session that is still open reads as in progress until you exit Claude Code, which closes it.
Set it up
-
Install the plugin, sign in and enable the folder. The Claude Code plugin guide covers this in full.
claude plugin marketplace add glassflow/rius-coding-agents && claude plugin install rius@rius-coding-agentsThen run
/rius:loginand/rius:enable-here. Run/rius:statusand check that the first line readsRius tracing: on. -
Open Traces in the console and pick a time range. The presets are 15m, 1h, 1d, 7d (the default), 30d and a custom range. Add a filter on the status with the value
failed. The strip above the list then shows the count, error rate, tokens and cost for just those sessions. In a shared workspace, click a cell in the Sent by column to narrow the list to one developer. -
Open a failed session. The waterfall lists every turn, model call, tool call and subagent in order. Failed spans are highlighted along the path from the root to the error. In structure-only mode, the span gives you the tool name and the error type. With content on, it also gives the error details.
-
To jump between failures, press
]efor the next failed span and[efor the previous one. To look across sessions, open the Spans tab, filter the kind to Tool and the status to failed, and read the failing steps from every session in one list. -
Open the Users page. Each developer shows their Failed traces for the selected window, and opening a developer lists their recent and failed sessions. With plugin 0.6.2 or later, sessions appear under the email the developer signed in with.
-
To ask in plain language, run
/mcp, pickriusand sign in once. Then ask Claude:Show me my last failed session and which tool call went wrong.The trace tools report
root_statusnext tostatus. A trace withstatusofErrorandroot_statusofOkis a recovered session. Aroot_statusofErrormarks a genuine failure.
What you get
- A list of failed sessions you can filter by time range, developer and status, and share as a link or save as a search.
- A path from a failed session to the exact failing step, with tool name, error type and the exit code of a shell command.
- Error details of a failed tool call when the folder sends content.
- A per-developer count of failed sessions on the Users page, and an error figure for the
claude-codeagent on the Agents page. - A separation between sessions that failed and sessions that recovered, so a retry does not look like a broken run.
When something else fits
- The Users and Agents counts include a session only after it ends. A session that crashed or was killed never counts there, so look for it in the trace list, where it stays in progress.
- If you want to compare spend and failures across a whole team, Track Claude Code usage and cost across a team covers the shared workspace setup.