Current behavior
The SDK captures spans via OpenTelemetry and exports them to Sentry. There is no built-in way to emit span data to stdout or the local logging system at runtime.
Gap
When building agents, most telemetry ends up in spans rather than logs. Developers currently need Spotlight (which requires a proxy/sidecar) to inspect spans locally. This is especially painful in agent-in-a-loop scenarios where the agent needs to self-debug — it has no way to see its own span data without external tooling.
Proposed behavior
Add an opt-in mode that streams each span to stdout as it finishes, one line per span. Output should include enough context to reconstruct parentage but does not need to render a tree — spans complete at different times so tree rendering isn't practical in a streaming model.
Minimal per-line fields:
- span name
- trace ID
- span ID
- parent span ID
- duration
- status
- key attributes (e.g.
op, description)
Activation options:
Sentry.init({ logSpans: true })SENTRY_LOG_SPANS=1 environment variable (primary ergonomic win — drop it in .env or .bashrc and every process emits span data with zero code changes)
Implementation notes:
- Must write to
process.stdout directly, not console.log, to avoid recursion with console instrumentation
- Registers as an additional
SpanExporter on the existing OTel TracerProvider
- Structured JSON by default for machine consumption; optional pretty format TBD
Context
This is an exploratory idea. The primary value proposition is agents self-debugging: when you're building an agent, most of the interesting telemetry is already in spans, not logs. Giving the agent (or developer) direct stdout visibility into its own trace data — without needing Spotlight or a proxy — fills a real gap.
Raised by David Cramer.
Action taken on behalf of David Cramer.
Current behavior
The SDK captures spans via OpenTelemetry and exports them to Sentry. There is no built-in way to emit span data to stdout or the local logging system at runtime.
Gap
When building agents, most telemetry ends up in spans rather than logs. Developers currently need Spotlight (which requires a proxy/sidecar) to inspect spans locally. This is especially painful in agent-in-a-loop scenarios where the agent needs to self-debug — it has no way to see its own span data without external tooling.
Proposed behavior
Add an opt-in mode that streams each span to stdout as it finishes, one line per span. Output should include enough context to reconstruct parentage but does not need to render a tree — spans complete at different times so tree rendering isn't practical in a streaming model.
Minimal per-line fields:
op,description)Activation options:
Sentry.init({ logSpans: true })SENTRY_LOG_SPANS=1environment variable (primary ergonomic win — drop it in.envor.bashrcand every process emits span data with zero code changes)Implementation notes:
process.stdoutdirectly, notconsole.log, to avoid recursion with console instrumentationSpanExporteron the existing OTelTracerProviderContext
This is an exploratory idea. The primary value proposition is agents self-debugging: when you're building an agent, most of the interesting telemetry is already in spans, not logs. Giving the agent (or developer) direct stdout visibility into its own trace data — without needing Spotlight or a proxy — fills a real gap.
Raised by David Cramer.
Action taken on behalf of David Cramer.