Skip to main content

Access Tiers

OSSA v0.3.2 Access Tiers and Separation of Duties specification for agent privilege management

Access Tiers

OSSA v0.3.2 introduces a comprehensive Access Tiers system that defines hierarchical privilege levels and separation of duties for AI agents. This specification ensures that agents operate within well-defined permission boundaries, supporting enterprise governance and security requirements.

Overview

The Access Tiers system provides:

  • Hierarchical privilege levels - Four tiers from read-only to policy governance
  • Separation of duties - Rules preventing conflicts of interest
  • Approval chains - Workflows for privileged operations
  • Delegation rules - How agents can delegate to lower tiers
  • Audit requirements - Logging and compliance tracking per tier

Quick Start

Every agent must declare exactly one access tier in its manifest:

apiVersion: ossa/v0.4.9 kind: Agent metadata: name: my-agent version: 1.0.0 labels: access_tier: tier_1_read spec: access: tier: tier_1_read permissions: - read_code - read_configs prohibited: - write_* audit_level: standard

The Four Tiers

TierNameDescriptionApproval Required
tier_1_readRead Only (Analyzers)Analyze, audit, scan - cannot modify stateNo
tier_2_write_limitedWrite Limited (Workers)Create/modify in sandboxed areas onlyNo
tier_3_write_elevatedWrite Elevated (Operators)Modify production systems with approvalYes
tier_4_policyPolicy (Governors)Define and enforce policies - isolated from executionNo

See Tiers Reference for detailed documentation on each tier.

Separation of Duties

OSSA enforces separation of duties to prevent conflicts of interest:

RuleDescription
Critic-Executor SeparationReviewers cannot approve; critics cannot execute
Governor-Executor SeparationPolicy-defining agents cannot execute policies
Read-Write SeparationIn security/compliance domains, readers cannot write
Production IsolationProduction modifications require elevated privileges

See Separation Rules for implementation details.

Approval Chains

Elevated operations require approval workflows:

ChainDescriptionSteps
StandardSingle approval1 human or governor
ElevatedTwo-step approvalGovernor + human
CriticalMulti-party approvalAutomated checks + governor + 2 humans

See Approval Chains for workflow definitions.

CLI Validation

Validate access tier configuration with the OSSA CLI:

# Install the CLI npm install -g @bluefly/openstandardagents # Validate agent manifest ossa validate my-agent.ossa.yaml # Check access tier compliance ossa validate my-agent.ossa.yaml --check access-tier

Example: Complete Agent with Access Tier

apiVersion: ossa/v0.4.9 kind: Agent metadata: name: security-scanner version: 1.0.0 description: Security vulnerability scanner - read-only analysis agent labels: team: security access_tier: tier_1_read domain: security spec: type: analyzer role: | You are a security vulnerability scanner that analyzes code for potential security issues. You can read code, scan dependencies, and report findings. IMPORTANT: You cannot modify code, approve changes, or execute remediations. llm: provider: anthropic model: claude-sonnet-4-20250514 temperature: 0.1 # Access Tier Configuration access: tier: tier_1_read permissions: - read_code - read_configs - read_logs - execute_queries prohibited: - write_* - delete_* - execute_commands audit_level: standard requires_approval: false # Separation of Duties separation: role: scanner conflicts_with: - remediator - executor - approver prohibited_actions: - execute - merge - approve

Specification Reference

The full Access Tiers specification is available in the OSSA repository: