AGENTS.md Template: Logistics Control Tower Agents

Overview

This AGENTS.md template for Logistics Control Tower Agents defines the operating context for AI coding agents coordinating multi-agent orchestration across planning, dispatch, inventory management, and carrier interactions. It supports both single-agent workstreams and multi-agent collaboration with explicit handoffs, memory rules, and supervisor-driven governance. Direct answer: use this AGENTS.md template to codify roles, rules, and handoffs for a logistics control tower workflow, enabling deterministic decisions and auditable collaboration among agents.

When to Use This AGENTS.md Template

• When building a logistics control tower that coordinates routes, inventory, and carrier handoffs across multiple agents.
• When explicit handoffs, tool governance, and escalation paths for human review are required.
• When you want a copyable, project-level operating manual that teams can paste into AGENTS.md for consistency.

Copyable AGENTS.md Template

# AGENTS.md
Project: Logistics Control Tower - Agent Operating Model

Project Role
- Objective: Coordinate end-to-end logistics from order intake to delivery using a unified agent operating model.
- Scope: Multi-agent orchestration, handoffs, tool governance, and human review.

Agent roster and responsibilities
- planner: define shipment plan, demand forecast, routing, and sequencing.
- dispatcher: assign legs to carriers, monitor ETA, replan on disruptions.
- inventory_manager: track stock, inbound/outbound, safety stock, and MOQs.
- carrier_connector: integrate with carrier systems, obtain capacity, book trips.
- compliance_checker: verify regulations, insurance, permits; flag violations.
- auditor: verify outputs, logs, metrics, and post-mortems.

Supervisor / orchestrator behavior
- The orchestrator maintains a single source of truth, memory, and decision log.
- It assigns tasks, schedules handoffs, and triggers human review when thresholds are crossed.
- It enforces timeouts, guardrails, and escalation rules.

Handoff rules between agents
- Planner → Dispatcher: once plan is generated, include ETA windows and critical constraints.
- Dispatcher → Inventory Manager: on plan confirmation, update stock projections and delivery windows.
- Inventory Manager → Carrier Connector: when capacity is secured, share booking details.
- Carrier Connector → Compliance Checker: share regulatory requirements and approvals.
- Compliance Checker → Auditor: provide compliance evidence and audit trail.

Context, memory, and source-of-truth rules
- Memory stores last 24 hours of decisions and related artifacts in the memory layer.
- Source of truth is the order management system and inventory system.
- All agent outputs must reference source IDs (order_id, sku_id, carrier_id).

Tool access and permission rules
- Agents may invoke read/write operations on systems they own (order, inventory, carriers) with read-only access to others unless explicitly granted.
- Secrets must be retrieved via a secure vault; never hard-code credentials.

Architecture rules
- Use a centralized orchestrator to manage state and inter-agent messaging.
- Stateless agents fetch fresh context per request unless memory is required.

File structure rules
- Keep templates under templates/logistics-control-tower/
- Use clear, deterministic file names (plan.json, execution.log, audit.json)

Data, API, or integration rules
- All outbound calls must be logged with timestamps and correlation IDs.
- Guard sensitive PII; mask where possible.

Validation rules
- Validate plan feasibility, capacity, and regulatory flags before execution.
- Validate that handoffs include all required context.

Security rules
- Enforce least privilege; rotate credentials every 90 days.
- Enforce audit trails for all critical actions.

Testing rules
- Include unit tests per agent, integration tests for handoffs, and end-to-end tests for typical delivery cycles.
- Security tests for access controls and secret handling.
- Canaries and health checks for deployment.

Deployment rules
- Deploy orchestrator changes with canary testing and feature flags.
- Rollback on failed health checks.

Human review and escalation rules
- Trigger human review for critical deviations or regulatory flags.
- Provide a concise incident summary and root cause trace.

Failure handling and rollback rules
- If a step fails, rollback to a safe state and notify the supervisor.
- Re-run with idempotent operations where possible.

Things Agents must not do
- Do not modify order data outside approved channels.
- Do not perform actions without an explicit plan or without handoffs.
- Do not circumvent governance or sensitive data restrictions.

Recommended Agent Operating Model

The planning, execution, and oversight loop relies on clearly defined roles and decision boundaries. Planner defines the route and plan; Dispatcher executes or re-routes; Inventory Manager maintains stock; Carrier Connector handles capacity; Compliance Checker enforces rules; Auditor validates the run. The supervisor coordinates handoffs, enforces guardrails, and escalates to human reviewers when needed.

Recommended Project Structure

logistics-control-tower/
├── agents/
│   ├── planner/
│   ├── dispatcher/
│   ├── inventory_manager/
│   ├── carrier_connector/
│   ├── compliance_checker/
│   └── auditor/
├── orchestrator/
├── memory/
├── tools/
├── data/
├── workflows/
│   └── order-to-delivery/
├── tests/
└── docs/

Core Operating Principles

• Operate with a single source of truth and auditable decisions.
• Isolate responsibilities; avoid cross-agent side effects.
• Enforce explicit handoffs with required context for every transition.
• Guard against data leakage and ensure secure access controls.
• Prefer idempotent actions and deterministic outputs.
• Provide clear failure signals and human-friendly escalation paths.

Agent Handoff and Collaboration Rules

• Planner to Dispatcher: include complete plan, constraints, and confidence.
• Dispatcher to Inventory Manager: include updated stock projections and delivery windows.
• Inventory Manager to Carrier Connector: share capacity, route, and booking details.
• Carrier Connector to Compliance Checker: provide regulatory requirements, permits, and approvals.
• Compliance Checker to Auditor: attach compliance evidence and audit trail.

Tool Governance and Permission Rules

• Commands may only be executed within approved tool boundaries.
• Edits to critical data (orders, inventory, or carriers) require orchestration approval.
• API calls must be logged with correlation IDs and secrets retrieved from a vault.
• Secrets must never be embedded in code or logs.
• Production systems require explicit approvals and canary rollout for changes.
• External services must be accessed through approved proxies with access controls.
• Handoff is the only contract for state changes; no surprise edits.

Code Construction Rules

• Write modular, idempotent code with clear input/output contracts.
• Do not bypass the orchestrator for state-changing actions.
• Validate outputs against the source-of-truth before applying changes.
• Document interfaces and dependencies for each agent.

Security and Production Rules

• Use least-privilege access to all systems; rotate credentials.
• Encrypt sensitive data in transit and at rest.
• Audit all critical actions; keep immutable logs.
• Test in staging with production-like data before live rollout.

Testing Checklist

• Unit tests for each agent with representative inputs.
• Integration tests for handoffs and orchestration paths.
• End-to-end tests for typical order-to-delivery scenarios.
• Security tests for access controls and secret handling.
• Canary deployment checks and rollback procedures.

Common Mistakes to Avoid

• Skipping explicit handoffs; relying on implicit state sharing.
• Hard-coding secrets or credentials in templates.
• Overloading agents with unrelated responsibilities.
• Forgetting to log correlation IDs for traceability.
• Neglecting human review for regulatory-critical actions.

FAQ

What is the purpose of this AGENTS.md template for Logistics Control Tower Agents?

It provides a copyable, project-level operating manual that codifies roles, rules, and handoffs for a logistics control tower workflow using AI coding agents.

Who supervises the multi-agent orchestration?

The orchestrator supervises, enforces guardrails, manages state, and triggers human review when thresholds are exceeded.

How are tool access and secrets managed?

Secrets are retrieved from a secure vault; tools and data access are restricted by least-privilege policies and audited actions.

What happens during a handoff between agents?

Handoffs pass complete context, including the plan, constraints, and correlation IDs; the receiving agent resumes from the new context with a clear boundary.

How is memory kept current and trustworthy?

Memory stores recent decisions with references to source-of-truth IDs; outputs must cite order_id, sku_id, or carrier_id to maintain traceability.