1Cademy - Four-Module Architecture for Agent Systems

Learn Before

Modular Agent Architecture: The Four-Module Protocol - Frontier Foundation Models, Capability Evaluation, and Just-In-Time Agent Harnesses @ University of Michigan - Ann Arbor
Four-Module Modular Harness Protocol - Adaptive Agent Runtimes: Architecture, Synthesis, and State Control @ University of Michigan - Ann Arbor
Agent Harness
The Four-Module Executable Harness Protocol - Dynamic Agent Scaffolding: Synthesis, Diagnostic Repair, and Evolutionary Optimization @ University of Michigan - Ann Arbor
HarnessFactory Codebase and Seed Scaffolds - Dynamic Agent Scaffolding: Synthesis, Diagnostic Repair, and Evolutionary Optimization @ University of Michigan - Ann Arbor

Definition

Four-Module Architecture for Agent Systems

Under the modular harness protocol, an agent harness $h$ is formalized as a four-tuple factorized over four protocol-compatible implementation spaces:

$h = (M, P, A, F) \in \mathfrak{M} \times \mathfrak{P} \times \mathfrak{A} \times \mathfrak{F}$

where:

$M$ is the Memory module, which compresses realized event history $\xi_{<t}$ and controller state $s_t$ into an active working view $v_t = M(\xi_{<t}, s_t) \in \mathcal{V}$ .
$P$ is the Planning module, which evaluates the task $\tau$ , state $s_t$ , and view $v_t$ to emit a local directive $d_t = P(\tau, s_t, v_t) \in \mathcal{D}_{\text{dir}}$ .
$F$ is the Capability Orchestration module, which selectively filters or activates task capabilities $C_\tau$ (tools, APIs, skills) into an available subset C_t = F(C_tau, s_t, v_t, d_t) subseteq C_tau$. -Ais the Action module, which consumes the assembled context to update the controller state and emit an action(s_{t+1}, e_t) = A(s_t, \tau, v_t, d_t, C_t) \in \mathcal{S} \times \mathcal{A}.

At runtime, the modules follow a strict operational dependency sequence:M \to P \to F \to A, operating against a shared, frozen backbone model\pi_\psi$$.

Updated 2026-10-02

Contributors are:

Who are from:

University of Michigan - Ann Arbor

✔️ 1

References

Learn Before

Related

Learn After