Operations
Dashboards For Claude Agent Work
Metrics, event groupings, and alert ideas for Claude Code and Agent SDK observability dashboards.
Core metrics
Sessions, tokens, cost, LOC
Claude Code exports counters for sessions, token usage, cost usage, lines of code modified, PRs, and commits.
Useful alerts
Rate, denial, retry, spend
The most useful alerts combine volume metrics with structured events such as API errors, permission changes, and tool decisions.
"claude_code.token.usage"
Dashboard The Work, Not The Hype
A Claude dashboard should answer operational questions: Are sessions completing? Which tools are slow or failing? Are permissions being denied? Are token and cost patterns changing? Are MCP calls attributed to the right user? Are background agents getting stuck?
Avoid vanity charts that count prompts without outcome. A session count only matters when it can be tied to success, failure, cost, or risk.
Core Panels
Start with a small dashboard that can survive production review. Show sessions started, successful and failed outcomes, estimated cost, token usage, API latency, tool execution time, tool decision counts, and top failing tools. Add dimensions for repository, team, model, query source, and user identity where policy allows.
Add a separate security view for permission mode changes, hook blocks, MCP server connections, detailed MCP tool calls where enabled, and unusually broad file or shell activity.
- Session health: starts, stops, stop failures, background restarts.
- Efficiency: tokens, cost estimates, cache behavior, turns per completed task.
- Tooling: tool latency, failures, denials, Bash command categories.
- Security: permission changes, MCP detail, hook blocks, identity attribution.
Alert Ideas That Are Worth Paging
Most agent alerts should go to a workflow channel, not a pager. Page only when agent activity can affect production, customer data, credentials, or high-cost automation. For everything else, create daily or per-PR review queues.
Good alerts are tied to action. An alert that says tokens are high is weak. An alert that says a production-bound workflow exhausted retries, used bypass permissions, and modified deployment files is actionable.
Correlate Back To Evidence
Every dashboard row should lead back to evidence: session ID, transcript path or archive key, trace ID, PR link, commit SHA, run ID, and user identity where available. Without those links, a dashboard becomes a wall of symptoms.
The transcript and dashboard should disagree gracefully. If telemetry says a tool failed but the transcript summary says work finished, send the reviewer to the raw event and final repository state.
Primary sources
Sources behind this page
Anthropic
Claude Code monitoring
Reference for Claude Code metric names, event analysis, identity attributes, MCP audit detail, and retry analysis.
Anthropic
Observability with OpenTelemetry
Explains Agent SDK telemetry flow through the Claude Code CLI and the metrics, logs, and traces exported over OTLP.
Anthropic
Track cost and usage
Explains per-step usage, model usage, session scoping, and the client-side nature of SDK cost estimates.
OpenTelemetry
OpenTelemetry documentation
Defines OpenTelemetry as a vendor-neutral observability framework for traces, metrics, and logs.
Cite this page
Claude Logs. "Dashboards For Claude Agent Work." claudelogs.com, updated 2026-07-06. https://claudelogs.com/dashboards
FAQ
What is the first dashboard to build?
Build a session health dashboard with session count, outcome, cost estimate, tokens, tool failures, permission denials, and links to transcript evidence.
Should dashboards include prompt text?
Usually no for broad dashboards. Keep prompt text in restricted traces or logs only when governance allows it.
Related evidence
Freshness
Updates
A freshness surface tracking Claude Code changes that affect session logs, transcript handling, hooks, telemetry, and audit review.
Cost evidence
Token Usage
How to read Claude Code token fields in transcripts and telemetry, estimate cost, and avoid treating local estimates as invoices.
Bibliography
Sources
Annotated source list for Claude Code session logs, Agent SDK session storage, hooks, telemetry, and general log-management guidance.
Free tool
Log Viewer
Drop a Claude Code JSONL transcript and turn it into a readable local timeline of messages, tool calls, tool results, and token fields.
Free tool
Token Analyzer
Estimate Claude Code transcript token usage and cost by turn with a dated model pricing table and CSV export.
Free tool
Audit Checklist
Run an interactive agent-session audit checklist and export a concise Markdown note for PRs, incidents, and handoffs.