Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.bentolabs.ai/llms.txt

Use this file to discover all available pages before exploring further.

An issue is a tracked problem Bento opened for you: a recurring failure mode it grouped across many runs, with a status, a severity, and its evidence attached. It’s the front door, where most triage starts, and the issues group is usually where you begin. An issue moves through six statuses, from new to fixing, review, watching, done, and ignored. It carries a severity from 1 (info) to 5 (critical) and a category that names the failure mode: tool_error, incomplete_task, wrong_tool_use, retry_storm, goal_divergence, or excessive_cost. Its evidence is the trajectories, findings, and runs behind it, plus the signals that detect it (the linked_signal_ids on the issue). The dashboard shows that evidence in three tabs, trajectories, findings, and signals, and the CLI mirrors them.

​
Commands

bentolabs issues --help
CommandWhat it does
issues listList issues. Filter by status, category, severity.
issues get <issue-id>One issue, with its linked_signal_ids and evidence totals.
issues list-trajectories <issue-id>The analyzed runs attached to the issue.
issues list-findings <issue-id>The observations attached to the issue.
issues list-runs <issue-id>The raw runs that contributed to the issue.
issues list-activity <issue-id>Status changes, comments, and system events.
issues comment-on <issue-id>Add a comment.
issues change-status <issue-id>Move it through the workflow.
issues createOpen an issue by hand.
issues update <issue-id>Edit fields like severity or category.

​
Triage an issue

List the open work, then narrow by category or severity to find what matters: 
bentolabs issues list --status new --output table

bentolabs issues list --category tool_error --severity 5 --output table
Open one issue. The response carries its linked_signal_ids and the counts behind each evidence tab: 
bentolabs issues get <issue-id>
Read the evidence in the same order the dashboard shows it. Trajectories are the analyzed runs that exhibit the problem; findings are the observations on top of them: 
bentolabs issues list-trajectories <issue-id> --output table

bentolabs issues list-findings <issue-id> --kind tool_error
Then follow the detectors. Take each id from linked_signal_ids and read the signal behind it: 
bentolabs signals get <signal-id>

​
Work an issue

Comment as you go: 
bentolabs issues comment-on <issue-id> --data '{"body": "Confirmed, retrying on a 429 with no backoff."}'
Move it through the workflow. to_status is required; body is an optional note: 
bentolabs issues change-status <issue-id> --data '{"to_status": "fixing", "body": "Picking this up"}'
Edit a field, or open an issue by hand. create requires title, summary, and category: 
bentolabs issues update <issue-id> --data '{"severity": 4}'

bentolabs issues create --data '{"title": "Checkout agent loops on payment retry", "summary": "Retries the charge with no backoff until it times out.", "category": "retry_storm", "severity": 4}'

​
See also

Trajectories

The analyzed runs an issue points at.

Findings

The observations extracted from those runs.

Signals

The detectors that tie an issue together.