// USE CASE · TRACING CODING AGENTS
Your coding agents work all day. See what they actually did.
Claude Code has telemetry built in: point it at RunAgain with a few environment variables and every session becomes one trace, with prompts, tool calls, API requests, token usage and accurate cost. No code changes, no wrapper.
THE PROBLEM
Agent-written code, zero visibility
Teams run coding agents across every repo and then answer questions like what did it change, why did it stall, and what does it cost per engineer with a shrug. The most active developer on the team has no timesheet.
// HOW RUNAGAIN DOES IT
Claude Code emits OpenTelemetry log events natively. Set a handful of environment variables with your RunAgain key and every terminal session starts streaming, unchanged.
Each session groups into a single trace: API requests as generation spans with model and tokens, tool results and decisions as tool spans with failures marked, and prompts when you opt in.
Cost comes from Claude Code's own cost figure, so per-session and per-engineer spend is accurate regardless of model mix, and it aggregates like any other trace column.
Coding-agent traces get the same treatment as production agents: error classification, baselines for abnormal sessions, drift monitors on cost or tokens, and Slack alerts.
One coding session as a trace: prompts, tools, cost.
FREQUENTLY ASKED
Do I need to modify Claude Code to trace it?
No. Enable Claude Code's built-in telemetry and point the OTLP logs exporter at RunAgain's logs endpoint with your API key. Sessions appear as traces immediately.
Can I see what tools the agent used in a session?
Yes. Tool results and tool decisions become tool spans inside the session's trace, with failures marked as errors, so a stalled session shows exactly where it stalled.
Does this work for other coding agents?
Any agent that emits OpenTelemetry spans or gen_ai attributes can send to the same OTLP endpoint, and the sandbox proxy can capture outbound calls for agents with no instrumentation at all.
BUILT ON THESE SOLUTIONS
More use cases: llm cost tracking · reduce llm costs · time-travel debugging · regression testing in ci · prompt a/b testing · hallucination detection · conversation tracing · llm-judge calibration · tool-call failure analysis · structured output validation · agent & llm alerting
Be first to run again.
Book 30 minutes and see the loop on your own agents, or write to tamas@runagain.ai.