Pattern Proposal: Telemetry-Observable Agent Steps

Summary

Instruments pipeline execution and agent interactions with OpenTelemetry to track latency, token usage, and step errors.

Candidate Pattern

• ID: telemetry-observable-agent-steps
• Classification: intelligence
• Maturity: established
• Type: implemented

Observed Signals

No specific signals listed.

Evidence

Supporting Evidence

• pyproject.toml - Includes opentelemetry-api and opentelemetry-sdk dependencies.
• src/cas_runner/governance/models.py (TraceableMetric) - Encapsulates metric records carrying trace context.

Scope Assessment

• Evidence Strength: high
• Confidence: high
• Applies to: ideas-to-life-cas

Rationale

• Intent: Provide comprehensive visibility into the latency, cost, and error boundaries of agentic runs.
• Context: Integrated via OpenTelemetry SDK dependencies configured in the runner runtime.
• Force: Multi-step agent pipelines suffer from opacity when failures occur inside nested LLM calls
• Force: Debugging cost spikes and slow steps requires distributed trace context

Traceability

Artifact source: proposed-patterns