Autonomous Research Analyst AGENTS.md Template

Overview

Direct answer: This AGENTS.md template provides a concrete operating manual for autonomous research analyst agents and a multi-agent orchestration pattern. It defines roles, guardrails, and handoffs to ensure reliable research outputs from AI coding agents.

The AGENTS.md template governs the autonomous research analyst workflow from data gathering, hypothesis testing, to synthesis, including handoffs to researchers, evaluators, and domain specialists as needed.

When to Use This AGENTS.md Template

• When you need repeatable, audit-friendly research automation with AI coding agents.
• When attempt a multi-agent approach to data gathering, hypothesis testing, and synthesis is required.
• When governance, memory, and source-of-truth controls are necessary to prevent context drift.

Copyable AGENTS.md Template

Copy the block below into AGENTS.md to establish the project-wide operating context.

# AGENTS.md

Project Role: Autonomous Research Analyst ensemble
Agents: Planner, Researcher, Evaluator, SourceOfTruthGuard, Domain Specialist

Supervisor/Orchestrator: The orchestrator coordinates task assignments, memory, and handoffs; enforces guardrails and approval gates for sensitive operations.

Handoff Rules: Transitions occur at defined decision points with full context and memory transfer to the next agent; ensure source-of-truth references are available.

Context, Memory, Source of Truth: Maintain a single memory store; use citations and data provenance for all claims; update memories only via approved channels.

Tool Access & Permissions: Access limited by role; secrets stored in vault; rotate keys; approvals required for privileged tool usage.

Architecture Rules: Microservice-like components with clear interfaces; avoid tight coupling across agents.

File Structure Rules: Organize by workflow: /src, /agents, /data, /tests, /docs.

Data/API/Integrations: Use standardized adapters; validate inputs/outputs; respect rate limits and auth scopes.

Validation Rules: Validate data quality, reproducibility, and traceability; log all decisions.

Security Rules: Encrypt at rest/in transit; restrict outbound connections; monitor suspicious activity.

Testing Rules: Unit, integration, and end-to-end tests for agent outputs; mock external systems in tests.

Deployment Rules: Deploy via staged promotions; require human review for critical changes.

Human Review & Escalation: All high-risk actions require human verification before execution.

Failure Handling & Rollback: Define rollback plan and recovery steps for every task.

Things Agents Must Not Do: Do not circumvent guardrails; do not operate on production systems without approval; never ignore source-of-truth.

Recommended Agent Operating Model

Roles and decision boundaries for a robust autonomous research workflow: Planner, Researcher, Evaluator, and Domain Specialist are coordinated by a central Orchestrator. Escalations route through human-in-the-loop when uncertain or high-risk tasks arise.

Recommended Project Structure

autonomous-research-project/
├── agents/
│   ├── planner/
│   ├── researcher/
│   ├── evaluator/
│   └── domain-specialist/
├── data/
├── src/
├── tests/
└── docs/

Core Operating Principles

• Respect source-of-truth and data provenance for all claims.
• Maintain memory discipline with explicit memory sources and timestamps.
• Favor human review for high-risk or ambiguous outputs.

Agent Handoff and Collaboration Rules

Planner assigns tasks; Researcher collects data; Evaluator verifies results; Domain Specialist adds domain nuance. Handoff includes context, results, and memory pointers.

Tool Governance and Permission Rules

Tool use is governed by role-scoped permissions; secrets in vault; approvals required for privileged tools.

Code Construction Rules

Produce modular, testable code; follow repository conventions; artifacts traceable to memory/context.

Security and Production Rules

Secure by design: encryption, access controls, audit logs, and safe rollout strategies.

Testing Checklist

• Unit tests for each agent function; integration tests for handoffs; end-to-end tests of research pipeline.

Common Mistakes to Avoid

• Drifting context across agents; skipping memory transfer; bypassing the source-of-truth.

FAQ

What is the purpose of this AGENTS.md Template for Autonomous Research Analysts?

This template defines a concrete operating manual for autonomous AI coding agents performing structured research tasks, with clear roles, guardrails, and multi-agent handoffs to ensure reliable research outputs.

How does multi-agent orchestration work in this template?

The template specifies a planner, researchers, evaluators, and domain specialists. The orchestrator coordinates tasks, triggers handoffs, and maintains memory and sources of truth to prevent drift.

What are the essential handoff rules between agents?

Handoffs occur at clearly defined decision points with context, completion criteria, and memory transfer. The receiving agent must reference the source of truth and confirm task readiness.

How are tools, APIs, and secrets governed?

Access is restricted by role and scope. Secrets are stored in a vault, API keys are rotated, and tool usage is audited. Handoffs require supervisor approval for sensitive operations.

What is the expected lifecycle for validation, testing, and deployment?

Code and results pass unit tests and integration checks. Deployment follows staged promotion with human review on critical paths.