TAG
A typed delegation protocol that models every handoff between agents, tools, models, and humans as a structured object with explicit authority, scope, budgets, and evidence contracts.
Reliable agent systems need more than loose prompt chaining. TAG treats delegation as a governed protocol: every delegated task carries its principal, authority scope, budget, and evidence contract with it.
Problem Space
Multi-agent workflows become ambiguous when tasks, permissions, outputs, and responsibilities are passed through unstructured prompts. Nothing records who delegated, under what authority, with what budget, or what evidence must come back.
System Direction
TAG models delegation as a typed object: a principal delegates to a delegate with an objective, an authority scope, a workspace scope, a resource budget, an evidence contract, an approval policy, and a termination policy. Handoffs become structured, replayable records instead of ambient prompt state.
Public Capabilities
- 01Delegation as a typed object
- 02Principal and delegate identity
- 03Objective and authority scope
- 04Workspace scope
- 05Resource budgets
- 06Evidence contracts
- 07Approval policies
- 08Termination policies
- 09Replayable delegation records
TAG is presented publicly as a protocol-level architecture. Internal schemas, enforcement mechanics, and routing algorithms are not disclosed.
What Is Not Disclosed
Private implementation details, security-sensitive internals, and unreleased runtime architecture are intentionally not disclosed.