Autonomous Agent Multi-Step Workflow Chain Failure-Mode Analysis

Role: Staff Reliability Engineer for Agentic Systems with deep specialization in stateful workflow orchestration and autonomous loop stability.

Context

• Orchestration Engine: {{orchestration_engine}}
• Workflow Chain Graph: {{chain_definition_payload}}
• State Persistence Backend: {{state_store_type}}
• Maximum Execution Depth: {{max_iteration_depth}}
• Idempotency Strategy: {{idempotency_strategy}}
• Authorized Side Effects: {{allowed_side_effects}}

Task

Deliver an in-depth reliability and failure-mode analysis report evaluating the autonomous workflow chain defined in {{chain_definition_payload}}, identifying single points of failure, runaway recursion risks, state synchronization bugs, and recovery gaps.

Method

1. Map the execution DAG defined in {{chain_definition_payload}}, establishing transition conditions between agent thoughts, tool invocations, and environment observation steps.
2. Stress-test loop termination boundaries against {{max_iteration_depth}} to detect cyclic dependencies where tool failure causes unrecoverable re-prompting loops.
3. Analyze state serialization between steps within {{state_store_type}}, checking for context window degradation, memory leaks, and race conditions during concurrent runs.
4. Audit step-level idempotency under {{idempotency_strategy}} to identify partial-failure scenarios where non-idempotent side effects could be duplicated upon retry.
5. Evaluate side-effect boundaries against {{allowed_side_effects}}, pinpointing where dynamic tool routing could bypass workflow constraints.
6. Trace fallback mechanisms for handling network timeouts, schema mismatches, and malformed tool outputs across external web service calls.
7. Formulate a deterministic state machine recovery model with concrete retry budgets, backoff policies, and dead-letter queue routing.

Constraints

• MUST evaluate specific failure vectors for every node identified in {{chain_definition_payload}}.
• MUST NOT suggest removing agentic autonomy as a substitute for resilient error handling.
• State transitions MUST include deterministic fallback states for non-deterministic model outputs.
• Analysis MUST explicitly account for storage mechanics in {{state_store_type}}.

Output format

Produce a technical resilience report organized into:

1. Execution Graph & Critical Path Analysis
2. Failure-Mode Effects Analysis (FMEA table: Node, Failure Mode, Probability, Impact, Mitigation)
3. State Drift and Idempotency Breakdown
4. Deterministic Recovery & Circuit-Breaker Architecture (with Mermaid sequence or state diagram)

Self-review

• Verify that the FMEA table covers every transition in {{chain_definition_payload}}.
• Check that retry strategies enforce the limits set in {{max_iteration_depth}} without risking infinite loops.
• Confirm that idempotency edge cases directly reflect the declared {{idempotency_strategy}}.