Reconfigurable Action Theories: A Formal Foundation for Dynamically Composable Agent Harnesses
Organizations: University of Science and Technology of China, Hefei, 230026, China
Abstract
Action theories typically fix their vocabulary and causal laws, whereas agent harnesses can change their tools, providers, and policies during execution. We introduce Reconfigurable Action Theories (RAT), in which actions can change the action theory itself. A theory module in RAT contains a vocabulary, causal laws, and an initializer. A RAT configuration records the active modules and the world state. Module loading and unloading are represented by actions governed by ordinary causal laws. RAT retains the standard causal action semantics of its language fragment when the active module set is fixed. We give sufficient conditions for at most one next configuration, preservation of existing behavior when a module is added, and equal results when modules are loaded in either order. We establish the computational complexity of reasoning tasks in RAT and translate finite RAT systems into nonmonotonic causal theories and Answer Set Programming (ASP), preserving bounded executions. Experiments compare finite RAT models with executions of official DeepSeek Harness, Pi, and OpenHands implementations. RAT provides a formal basis for checking how component changes affect a harness's available actions, behavior, and reachable goals.
Figures & tables
| Stage | Operations | Observed result |
|---|---|---|
| 1 | Load the observation-policy plugin, then write "alpha=2" . | Loading succeeds. Write rejected because the file has not been observed. File remains "alpha=1" . |
| 2 | Read, then edit "alpha=1" to "alpha=2" . | Both calls succeed. File becomes "alpha=2" . |
| 3 | External write of "alpha=external" , then write "alpha=3" . | External write succeeds, making the file "alpha=external" . The subsequent write is rejected as the observation is outdated. |
| 4 | Read again, then write "alpha=3" . | Both calls succeed. File becomes "alpha=3" . |
| 5 | Unload the observation-policy plugin, perform an external write of "alpha=external-again" , then write "alpha=4" . | Unloading succeeds. External write makes the file "alpha=external-again" . The subsequent write succeeds without another read, making the file "alpha=4" . |
| 6 | Reload the observation-policy plugin, then write "alpha=5" . | Reloading clears the observation record. Write rejected. File remains "alpha=4" . |
| Family | Task | Cases | Repeats | Runs | SAT | UNSAT | Timeout | Memory |
|---|---|---|---|---|---|---|---|---|
| Tools | C | 5 | 3 | 15 | 12 | 0 | 0 | 3 |
| Tools | P | 5 | 3 | 15 | 9 | 0 | 3 | 3 |
| Providers | C | 50 | 1 | 50 | 40 | 0 | 0 | 10 |
| Providers | P | 50 | 1 | 50 | 19 | 12 | 1 | 18 |
| Resources, feasible | C | 12 | 3 | 36 | 36 | 0 | 0 | 0 |
| Resources, feasible | P | 8 | 3 | 24 | 24 | 0 | 0 | 0 |
LLM Agent Harnesses
LLM agent harnesses are the software and runtime systems that organize an agent’s tools, context, state, execution, and safeguards.
See all 428 papers