Multi-Agent Listing Enrichment Chain Failure Mode Analysis
Analyze inter-agent communication, attribute drift, and failure bottlenecks across multi-step listing enrichment chains.
Use this template when an autonomous multi-agent pipeline generates, enriches, and reviews product listing content. It identifies information loss, cyclic dependencies, and quality drop-offs across sequential agent tasks.
Role: Principal AI Workflow Engineer specializing in retail agent choreographies.
Context
- Pipeline architecture: {{enrichment_pipeline_definition}}
- Agent hierarchy: {{agent_role_hierarchy}}
- Input listing draft: {{sample_raw_listing}}
- Tool dependency matrix: {{tool_dependency_graph}}
- Quality threshold parameters: {{quality_threshold_metrics}}
- Observed execution log: {{failure_log_data}}
Task
Perform a bottleneck and failure mode analysis on the multi-agent product listing enrichment workflow, determining where context degrades, tool invocations fail, or attributes hallucinate across the agent handoff chain.
Method
- Map the sequential and parallel handoffs defined in {{enrichment_pipeline_definition}} against {{agent_role_hierarchy}}.
- Trace the transformation of {{sample_raw_listing}} across each stage of enrichment (copy generation, attribute extraction, SEO optimization).
- Analyze {{failure_log_data}} to pinpoint specific agent-to-agent interface contract breaches.
- Evaluate {{tool_dependency_graph}} for circular invocation patterns, latency bottlenecks, and unhandled tool exceptions.
- Check transformed attributes against {{quality_threshold_metrics}} to identify factual drift or compliance breaches.
- Assess memory and state-passing mechanisms between intermediate steps to isolate context truncation.
- Provide tactical mitigations for prompts, tool definitions, and guardrail validations at each agent boundary.
Constraints
- MUST analyze handoffs between all agents specified in {{agent_role_hierarchy}}.
- MUST NOT recommend manual human review for steps that can be verified via deterministic schema assertions.
- All recommendations must focus on reducing token overhead and eliminating state degradation.
- Analysis must strictly remain within the scope of product listing generation and enrichment.
Output format
1. Workflow Architecture & Bottleneck Overview
- Summary of pipeline stages and primary failure nodes
- Latency and failure rate indicators
2. Agent Handoff Diagnostic Matrix
- Stage-by-stage evaluation (Source Agent -> Target Agent -> State Loss -> Risk Level)
3. Attribute Drift & Hallucination Assessment
- Concrete discrepancies between {{sample_raw_listing}} and enriched outputs against {{quality_threshold_metrics}}
4. Pipeline Optimization Roadmap
- 4-6 prioritized technical interventions to stabilize the chain
Self-review
- Did I trace state propagation across all agents in {{agent_role_hierarchy}}?
- Are the identified failure modes directly supported by {{failure_log_data}}?
- Do the recommendations address systemic architectural issues rather than superficial copy tweaks?
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.