Skip to main content
Multi-agent systems involve an orchestrator delegating tasks to specialized agents, which may further delegate to sub-agents. Without governance, delegation chains can escalate privileges, bypass tool restrictions, and lose traceability. This flow shows how TealTiger enforces delegation governance, privilege boundaries, and correlation ID tracing across cooperating agents.

Orchestration Flow

Orchestrator receives task

The orchestrator agent receives a complex task: “Research competitor pricing and update our pricing spreadsheet.” This requires two capabilities — web research and spreadsheet editing — handled by different specialized agents.

TealEngine — Evaluate delegation to Agent A

TealEngine evaluates whether the orchestrator is allowed to delegate the research subtask to Agent A (Research Agent). The engine checks: Does the orchestrator have delegation permission? Is Agent A registered? Is the delegation depth within limits (max depth: 3)?

Agent A delegates to Agent B

Agent A (Research Agent) determines it needs Agent B (Web Scraper) to fetch competitor websites. Agent A attempts to delegate a sub-task to Agent B.

TealEngine — Verify Agent B permissions

TealEngine evaluates the delegation from Agent A to Agent B. It checks: Does Agent A have permission to delegate? Is Agent B registered? Current delegation depth is 2 (Orchestrator → A → B) — within the limit of 3. Critically, Agent B inherits only the permissions of Agent A, not the orchestrator. Privilege cannot escalate through delegation.

Agent B calls tool

Agent B attempts to call the web_scrape tool to fetch competitor pricing pages.

TealRegistry — Tool authorization for Agent B

TealRegistry checks whether web_scrape is allowlisted for Agent B specifically. Each agent has its own tool allowlist — Agent B can use web_scrape and web_search, but cannot use spreadsheet_edit (that’s only allowlisted for the orchestrator).

Agent B returns result

Agent B completes the web scraping and returns competitor pricing data to Agent A, which aggregates the research and returns it to the orchestrator.

TealMemory — Shared memory governance

TealMemory governs the shared memory space. Agent B writes research results to shared memory with classification INTERNAL. The orchestrator can read this data, but Agent B cannot read the orchestrator’s CONFIDENTIAL pricing strategy documents. Memory access is governed by both scope and classification.

TealAudit — Correlation ID traces full chain

TealAudit logs every event in the delegation chain with a shared correlation ID. The audit trail shows: Orchestrator → delegated to Agent A → Agent A delegated to Agent B → Agent B called web_scrape → Agent B returned result → Agent A returned result → Orchestrator updated spreadsheet. A single query on the correlation ID reconstructs the entire chain.

Flow Diagram

Multi-Agent System governance flow

Code Example

Key Governance Points

Without delegation governance, a low-privilege agent can delegate to a higher-privilege agent and effectively escalate its own permissions. TealEngine enforces that delegated agents inherit the permissions of their parent, never the root orchestrator.
Correlation IDs propagate automatically through createChildContext. Every event in the delegation chain — across all agents — shares the same root correlation ID, making it possible to reconstruct the full chain with a single audit query.

Agentic Patterns

Governance patterns for multi-agent architectures

Execution Identity

How correlation IDs and execution contexts work

TealRegistry API

Agent and tool registration configuration

TealMemory API

Shared memory governance and classification