Utility Skill
Overview
A decision framework for agent orchestration based on Liu et al.,
"Utility-Guided Agent Orchestration for Efficient LLM Tool Use"
(arXiv:2603.19896).
Each candidate action is scored by subtracting weighted costs from
expected gain, producing a single utility value that guides action
selection.
The framework prevents over-calling tools and premature stopping by
making both errors costly.
Utility range is [-2.3, 1.0].
When To Use
- Deciding whether to dispatch another agent or tool call
- Gating expensive tool calls (search, code execution, delegation)
- Selecting the right model tier for a sub-task
- Continuation decisions after receiving partial results
- Verification gating before writing or committing output
When NOT to Use
- Single-step operations with one obvious action
- Trivial tasks where cost of scoring exceeds benefit
- Already-committed actions that cannot be undone
Action Space
A = {respond, retrieve, tool_call, verify, delegate, stop}
| Action |
Description |
| respond |
Emit a final answer from current context |
| retrieve |
Fetch additional information (search, read, lookup) |
| tool_call |
Execute a tool (code runner, API, file write) |
| verify |
Check a prior result for correctness or completeness |
| delegate |
Spawn a sub-agent or hand off to a specialist |
| stop |
Terminate the loop and return current state |
Utility Function
U(a | s_t) = Gain(a | s_t)
- λ₁ · StepCost(a | s_t)
- λ₂ · Uncertainty(a | s_t)
- λ₃ · Redundancy(a | s_t)
| Parameter |
Default |
Rationale |
| λ₁ |
1.0 |
Cost baseline; all other weights relative to this |
| λ₂ |
0.5 |
Weak empirical correlation with outcome (r=0.0131) |
| λ₃ |
0.8 |
Redundancy pruning yields ~10% token savings |
Utility range: [-2.3, 1.0].
Positive values indicate the action is worth taking.
Values below the floor (-0.5 default) indicate the action should
be skipped.
Termination Conditions
Stop the loop when any of the following is true:
- (a) Selected action is
stop
- (b) Step budget exhausted (default: 10 steps)
- (c) All non-
stop actions score below the floor (default: -0.5)
High-gain override: If Gain >= 0.7 for any action, condition
(c) may be overridden.
Document the override and the gain value in your reasoning trace.
Quick Start
Minimal 4-step advisory pattern:
- Construct state: gather task context per
modules/state-builder.md
- Score candidates: evaluate each action in
A per
modules/action-selector.md
- Prefer highest utility: select the action with the
maximum
U(a | s_t), subject to termination conditions
- Log score and decision: record the winning action,
its utility value, and step count before executing
Detailed Resources
- State Builder:
modules/state-builder.md, how to
populate s_t from task context
- Gain:
modules/gain.md, estimating expected information
or progress gain
- Step Cost:
modules/step-cost.md, token, latency, and
monetary cost tables
- Uncertainty:
modules/uncertainty.md, confidence
estimation and calibration
- Redundancy:
modules/redundancy.md, detecting duplicate
or low-delta actions
- Action Selector:
modules/action-selector.md, scoring
loop and tie-breaking rules
- Integration:
modules/integration.md, wiring utility
scoring into existing orchestration loops
Exit Criteria
1---2name: utility-23description: Scores agent actions by expected gain, cost, uncertainty, and redundancy. Use when deciding whether to dispatch an agent or invoke a tool.4alwaysApply: false5category: infrastructure6tags:7- orchestration8- cost-control9- decision-making10- agent-dispatch11dependencies: []12provides:13 infrastructure:14 - utility-scoring15 - action-selection16 - termination-control17 patterns:18 - gain-estimation19 - cost-computation20 - redundancy-detection21usage_patterns:22- agent-dispatch-gating23- tool-call-decisions24- continuation-decisions25- model-tier-selection26complexity: intermediate27model_hint: standard28estimated_tokens: 60029progressive_loading: true30modules:31- modules/state-builder.md32- modules/gain.md33- modules/step-cost.md34- modules/uncertainty.md35- modules/redundancy.md36- modules/action-selector.md37- modules/integration.md38metadata:39 source: https://skillmd.com/skills/athola/utility-240---4142# Utility Skill4344## Overview4546A decision framework for agent orchestration based on Liu et al.,47"Utility-Guided Agent Orchestration for Efficient LLM Tool Use"48(arXiv:2603.19896).49Each candidate action is scored by subtracting weighted costs from50expected gain, producing a single utility value that guides action51selection.52The framework prevents over-calling tools and premature stopping by53making both errors costly.54Utility range is [-2.3, 1.0].5556## When To Use5758- Deciding whether to dispatch another agent or tool call59- Gating expensive tool calls (search, code execution, delegation)60- Selecting the right model tier for a sub-task61- Continuation decisions after receiving partial results62- Verification gating before writing or committing output6364## When NOT to Use6566- Single-step operations with one obvious action67- Trivial tasks where cost of scoring exceeds benefit68- Already-committed actions that cannot be undone6970## Action Space7172`A = {respond, retrieve, tool_call, verify, delegate, stop}`7374| Action | Description |75|-----------|------------------------------------------------------|76| respond | Emit a final answer from current context |77| retrieve | Fetch additional information (search, read, lookup) |78| tool_call | Execute a tool (code runner, API, file write) |79| verify | Check a prior result for correctness or completeness |80| delegate | Spawn a sub-agent or hand off to a specialist |81| stop | Terminate the loop and return current state |8283## Utility Function8485```86U(a | s_t) = Gain(a | s_t)87 - λ₁ · StepCost(a | s_t)88 - λ₂ · Uncertainty(a | s_t)89 - λ₃ · Redundancy(a | s_t)90```9192| Parameter | Default | Rationale |93|-----------|---------|---------------------------------------------------|94| λ₁ | 1.0 | Cost baseline; all other weights relative to this |95| λ₂ | 0.5 | Weak empirical correlation with outcome (r=0.0131) |96| λ₃ | 0.8 | Redundancy pruning yields ~10% token savings |9798Utility range: **[-2.3, 1.0]**.99Positive values indicate the action is worth taking.100Values below the floor (-0.5 default) indicate the action should101be skipped.102103## Termination Conditions104105Stop the loop when **any** of the following is true:106107- (a) Selected action is `stop`108- (b) Step budget exhausted (default: 10 steps)109- (c) All non-`stop` actions score below the floor (default: -0.5)110111**High-gain override:** If `Gain >= 0.7` for any action, condition112(c) may be overridden.113Document the override and the gain value in your reasoning trace.114115## Quick Start116117Minimal 4-step advisory pattern:1181191. **Construct state**: gather task context per120 `modules/state-builder.md`1212. **Score candidates**: evaluate each action in `A` per122 `modules/action-selector.md`1233. **Prefer highest utility**: select the action with the124 maximum `U(a | s_t)`, subject to termination conditions1254. **Log score and decision**: record the winning action,126 its utility value, and step count before executing127128## Detailed Resources129130- **State Builder**: `modules/state-builder.md`, how to131 populate `s_t` from task context132- **Gain**: `modules/gain.md`, estimating expected information133 or progress gain134- **Step Cost**: `modules/step-cost.md`, token, latency, and135 monetary cost tables136- **Uncertainty**: `modules/uncertainty.md`, confidence137 estimation and calibration138- **Redundancy**: `modules/redundancy.md`, detecting duplicate139 or low-delta actions140- **Action Selector**: `modules/action-selector.md`, scoring141 loop and tie-breaking rules142- **Integration**: `modules/integration.md`, wiring utility143 scoring into existing orchestration loops144145## Exit Criteria146147- [ ] State constructed with task goal and prior steps148- [ ] All six actions scored before selecting one149- [ ] Termination condition checked after each step150- [ ] Score and decision logged for each step taken151- [ ] High-gain overrides documented with gain value