Multi-Agent Execution Pipeline Failure and Circuit Breaker Analysis
Audit autonomous agent workflow chains to detect failure propagation, recovery loops, and circuit breaker vulnerabilities.
Use this template when designing or reviewing multi-step AI agent workflows that interact with production engineering toolchains. It produces an architectural analysis of failure points, retry storms, and fallback isolation mechanisms.
Role: Principal Site Reliability and Agent Architecture Specialist
Context
- Workflow specification: {{agent_pipeline_spec}}
- Execution infrastructure: {{runtime_environment}}
- External dependencies: {{dependency_topology}}
- Resilience policies: {{error_budget_policy}}
- Known failure traces: {{historical_failure_logs}}
Task
Produce an exhaustive failure propagation and circuit breaker analysis for the multi-step autonomous agent workflow, pinpointing cascade vulnerabilities and defining deterministic fallback topologies.
Method
- Deconstruct the workflow DAG defined in {{agent_pipeline_spec}} into discrete transition nodes, identifying state mutations at each step.
- Map the blast radius of external dependency failures across {{dependency_topology}} under network degradation or token rate-limiting.
- Evaluate {{error_budget_policy}} against recursive retry behaviors to detect hallucination amplification or infinite retry loops.
- Correlate historical anomalies in {{historical_failure_logs}} with pipeline choke points and concurrency limits in {{runtime_environment}}.
- Design circuit breaker trigger conditions for agent tool calls based on consecutive parsing errors, schema violations, and execution timeouts.
- Formulate deterministic fallback mechanisms (e.g., graceful degradation, human-in-the-loop escalation, static cache retrieval) for each stage.
- Assess state rollbacks and transaction roll-forward viability across multi-agent handoffs.
Constraints
- MUST evaluate every node in the workflow for deadlocks and unhandled exception states.
- MUST NOT recommend unbounded exponential retries on non-idempotent tool execution steps.
- Analysis must distinguish between LLM reasoning drift and deterministic runtime infrastructure failures.
- Recommendations must preserve data consistency within {{runtime_environment}}.
Output format
- Executive Summary: High-level resilience rating and key vulnerability hotspots (max 200 words).
- Failure Mode and Effects Analysis (FMEA) Table: Columns for Step ID, Failure Mode, Root Cause, Severity (1-5), and Blast Radius.
- Circuit Breaker Architecture: Explicit thresholds, trip conditions, and half-open reset parameters per agent transition.
- Fallback Routing Strategies: Concrete fallback routing pseudocode or state machine diagrams.
- Mitigation Roadmap: Prioritized action items ranked by implementation complexity and risk reduction.
Self-review
- Did I audit all external tool dependencies listed in {{dependency_topology}}?
- Are all fallback paths deterministic and free of recursive agent invocation?
- Does the analysis address state rollbacks for partially completed transactions?
Explicit role, a named task, and discrete steps the model can follow.
Background, inputs and variables the model needs before it starts.
Hard boundaries — what the model must and must not do.
A named, field-level shape for the response.
Ordered work items that force analysis before an answer.
Length and structure that travel across frontier models.
Signal density — instruction weight without padding.
Documented variables so the scaffold adapts to new inputs.
Quality bar, assumptions and behaviour when inputs are thin.
How much real usage the template has behind it.