Academic Literature Review Agent Orchestration Plan
Architect a sequential multi-agent pipeline for automated academic literature screening, data extraction, and synthesis.
Use this template when setting up multi-agent research pipelines to systematically screen academic databases, extract empirical findings, and validate claims. It is designed for university research centers and informatics departments looking to automate literature discovery workflows.
Role: Principal Academic Informatics Architect with deep expertise in multi-agent orchestration and systematic review methodologies.
Context
- Target Research Domain: {{research_domain}}
- Source Databases and APIs: {{target_databases}}
- Screening and Inclusion Criteria: {{inclusion_criteria}}
- Target Agent Orchestration Framework: {{orchestration_engine}}
- Reference Validation Benchmark: {{validation_benchmark}}
- Human-in-the-Loop Verification Stage: {{human_review_stage}}
Task
Develop a comprehensive architectural workflow plan that sequences specialized autonomous agents to execute an end-to-end systematic literature review, establishing deterministic validation gates between literature ingestion, deduplication, abstract screening, full-text extraction, and synthesis.
Method
- Map data ingestion agents against {{target_databases}}, defining query formatting, pagination handling, and metadata normalization.
- Construct the primary screening agent prompt topology using {{inclusion_criteria}} for binary and confidence-scored triage.
- Design a deduplication and record linkage chain to reconcile preprints with peer-reviewed versions.
- Define the full-text extraction agent schema to parse methodology, sample size, effect sizes, and primary conclusions.
- Engineer a secondary critique agent that tests extracted findings against {{validation_benchmark}} to flag hallucinations.
- Specify the trigger parameters for {{human_review_stage}} when agent confidence falls below predefined statistical thresholds.
- Detail inter-agent state management and token payload handoffs within {{orchestration_engine}}.
- Establish telemetry and logging specifications for academic auditability and reproducibility.
Constraints
- MUST maintain an immutable audit trail linking every extracted datum to source DOI and page coordinates.
- MUST NOT allow downstream synthesis agents to execute without passing confidence validation gates.
- Hand-off payloads between agents must be strictly typed JSON schemas.
- State persistence must support asynchronous resume after manual review intervention.
Output format
Provide a structured implementation plan containing:
- Section 1: End-to-End Pipeline Topology (narrative flow and state diagram description)
- Section 2: Agent Node Specifications (roles, prompt models, input/output schemas for at least 4 distinct agents)
- Section 3: Human-in-the-Loop Gateways and Failure Recovery Matrix
- Section 4: Telemetry, Verification, and Reproducibility Protocol Total length should be between 600 and 900 words.
Self-review
- Did I define clear input and output schema boundaries for every agent in the chain?
- Are the confidence score cutoffs for human handoff explicitly specified?
- Does the plan address API rate-limiting and asynchronous failure states across databases?
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.