Workflow chains
AuraScore 81/100

Multi-Agent Orchestration Observability and Telemetry Specification

Specify distributed tracing, token accounting, causal state tracking, and SLO telemetry for complex multi-agent execution graphs.

Use this template when deploying production-grade multi-agent debugging or engineering workflows that require fine-grained observability. It defines a unified telemetry standard covering distributed trace propagation, latency budgets, and cost governance.

Template

Role: Staff Site Reliability Engineer specializing in distributed agent runtime observability and telemetry pipelines.

Context

  • Orchestration Framework: {{orchestration_engine}}
  • Workflow DAG Structure: {{workflow_dag_topology}}
  • Telemetry Backend: {{telemetry_backend}}
  • Target Latency SLOs: {{slo_latency_targets}}
  • Trace Export Standard: {{telemetry_export_standard}}
  • Total Cost & Token Budget: {{cost_budget_ceiling}}

Task

Produce an Observability and Telemetry Specification Report that establishes distributed tracing standards, metric instrumentation, and anomaly detection rules across the multi-agent execution graph.

Method

  1. Define trace context propagation mechanisms across asynchronous node boundaries in {{orchestration_engine}} using {{telemetry_export_standard}}.
  2. Design span models for LLM calls, tool executions, vector retrievals, and deterministic evaluation gates within {{workflow_dag_topology}}.
  3. Establish structured log attributes capture (causal session IDs, node parentage, model version, prompt/completion token splits, temperature).
  4. Specify latency metrics and warning thresholds aligned with {{slo_latency_targets}} for each critical execution path.
  5. Formulate real-time token tracking and cost-attribution algorithms to enforce constraints against {{cost_budget_ceiling}}.
  6. Architect anomaly detection alerts in {{telemetry_backend}} for infinite reasoning loops, context window saturation, and tool error spikes.
  7. Detail sampling strategies that balance diagnostic depth against telemetry ingestion overhead for high-throughput workflows.

Constraints

  • MUST specify W3C Trace Context or native OpenTelemetry propagation headers for all inter-agent messages.
  • MUST NOT log raw credentials, PII, or unsanitized proprietary source code within telemetry span attributes.
  • Span attributes and metric schemas MUST integrate natively with {{telemetry_backend}}.
  • Token attribution mechanisms MUST calculate real-time cumulative spend relative to {{cost_budget_ceiling}}.

Output format

Deliver an engineering specification report divided into four sections:

  1. Telemetry Architecture & Context Propagation (tracing standard, headers, and span lifecycle)
  2. Instrumentation Schema (exact key-value semantic conventions for nodes, LLM invocations, and tool executions)
  3. SLO Monitoring & Alerting Matrix (thresholds, latency baselines matching {{slo_latency_targets}}, and alert rules)
  4. Cost Governance & Token Accounting Model (budget tracking, burn-rate metrics, and circuit-breaker conditions) Length must be between 900 and 1400 words.

Self-review

  • Verify every branch in {{workflow_dag_topology}} has explicit parent-child trace context inheritance.
  • Ensure metric definitions align precisely with the capabilities of {{telemetry_backend}}.
  • Confirm alert conditions cover both performance degradation (SLOs) and financial overrun ({{cost_budget_ceiling}}).
AuraScore breakdown
81/100Provisional
Instruction clarity15/15 · Strong

Explicit role, a named task, and discrete steps the model can follow.

Context architecture12/12 · Strong

Background, inputs and variables the model needs before it starts.

Constraint engineering12/12 · Strong

Hard boundaries — what the model must and must not do.

Output specification6/14 · Thin

A named, field-level shape for the response.

Reasoning structure10/10 · Strong

Ordered work items that force analysis before an answer.

Model compatibility10/10 · Strong

Length and structure that travel across frontier models.

Token efficiency5/10 · Thin

Signal density — instruction weight without padding.

Reusability7/7 · Strong

Documented variables so the scaffold adapts to new inputs.

Robustness3/5 · Adequate

Quality bar, assumptions and behaviour when inputs are thin.

Observed performance1/5 · Thin

How much real usage the template has behind it.

ai-agents
agents-workflows
software-engineering-debugging
workflow-chains
observability
distributed-tracing