Infrastructure Types

The Strategos.Infrastructure package provides production-ready implementations of core abstractions.

Thompson Sampling

Adaptive agent selection using contextual multi-armed bandits.

ContextualAgentSelector

Selects the best agent for a task using Thompson Sampling with learned beliefs.

Methods

Method	Parameters	Returns	Description
SelectAgentAsync	string taskDescription, CancellationToken ct	Task<AgentId>	Selects best agent for task
RecordOutcomeAsync	AgentId agent, TaskCategory category, bool success, CancellationToken ct	Task	Updates beliefs based on outcome

Example

public class AssignAgentStep : IWorkflowStep<TaskState>
{
    private readonly IAgentSelector _selector;

    public async Task<StepResult<TaskState>> ExecuteAsync(
        TaskState state,
        StepContext context,
        CancellationToken ct)
    {
        var agent = await _selector.SelectAgentAsync(state.TaskDescription, ct);
        return state.With(s => s.AssignedAgent, agent).AsResult();
    }
}

---

AgentBelief

Represents performance belief as a Beta distribution.

Properties

Property	Type	Description
Alpha	double	Success count + prior
Beta	double	Failure count + prior
Mean	double	Expected success rate (alpha / (alpha + beta))
Variance	double	Uncertainty in estimate

Methods

Method	Parameters	Returns	Description
Sample	Random random	double	Draws sample from distribution
Update	bool success	AgentBelief	Returns updated belief

Example

// Initial belief with uniform prior
var belief = new AgentBelief(alpha: 1, beta: 1);

// After 7 successes and 3 failures
belief = belief.Update(success: true);  // Repeat for each outcome
// belief.Mean ~= 0.727

---

TaskCategory

Enumeration of task categories for contextual selection.

Value	Description
Analysis	Data analysis and interpretation
Coding	Code generation and modification
Research	Information gathering and synthesis
Writing	Content creation and editing
Planning	Strategy and roadmap development
Review	Evaluation and feedback
Unknown	Unclassified task type

---

IBeliefStore

Interface for persisting agent performance beliefs.

Methods

Method	Parameters	Returns	Description
GetBeliefAsync	AgentId agent, TaskCategory category	Task<AgentBelief>	Gets belief for agent/category
SetBeliefAsync	AgentId agent, TaskCategory category, AgentBelief belief	Task	Persists updated belief

Implementations

Type	Description
InMemoryBeliefStore	In-memory storage (dev/testing)

---

Loop Detection

Detects stuck workflows using multiple strategies.

ExactRepetitionDetector

Detects identical action sequences.

Configuration

Property	Type	Default	Description
WindowSize	int	10	Number of recent actions to analyze
Threshold	int	3	Repetitions to trigger detection

Example

services.AddLoopDetection(options => options
    .AddExactRepetition(config => config
        .WithWindowSize(10)
        .WithThreshold(3)));

---

SemanticRepetitionDetector

Detects similar outputs using cosine similarity.

Configuration

Property	Type	Default	Description
SimilarityThreshold	double	0.95	Minimum similarity to consider repetition
WindowSize	int	5	Number of recent outputs to compare

Example

services.AddLoopDetection(options => options
    .AddSemanticRepetition(config => config
        .WithSimilarityThreshold(0.95)
        .WithWindowSize(5)));

---

OscillationDetector

Detects A-B-A-B patterns.

Configuration

Property	Type	Default	Description
PatternLength	int	2	Length of oscillation pattern
Repetitions	int	2	Times pattern must repeat

Example

services.AddLoopDetection(options => options
    .AddOscillation(config => config
        .WithPatternLength(2)
        .WithRepetitions(2)));

---

NoProgressDetector

Detects activity without meaningful state change.

Configuration

Property	Type	Default	Description
StepThreshold	int	5	Steps without progress to trigger
ProgressEvaluator	Func<TState, TState, bool>	-	Custom progress evaluation

Example

services.AddLoopDetection(options => options
    .AddNoProgress(config => config
        .WithStepThreshold(5)
        .WithProgressEvaluator((prev, curr) =>
            prev.ProcessedCount != curr.ProcessedCount)));

---

Budget Guard

Enforces resource limits on workflow execution.

BudgetGuard

Main type for budget enforcement.

Methods

Method	Parameters	Returns	Description
CheckBudgetAsync	StepContext context	Task<BudgetStatus>	Checks current budget status
RecordUsageAsync	ResourceUsage usage	Task	Records resource consumption

---

BudgetOptions

Configuration for budget thresholds.

Properties

Property	Type	Default	Description
MaxSteps	int	100	Maximum step executions
MaxTokens	int	50,000	Maximum LLM tokens
MaxWallTime	TimeSpan	30 minutes	Maximum elapsed time
ScarceThreshold	double	0.75	Percentage to enter Scarce state
CriticalThreshold	double	0.90	Percentage to enter Critical state

Example

services.AddBudgetGuard(options => options
    .WithMaxSteps(100)
    .WithMaxTokens(50_000)
    .WithMaxWallTime(TimeSpan.FromMinutes(30))
    .WithScarceThreshold(0.75)
    .WithCriticalThreshold(0.90));

---

ScarcityLevel

Budget consumption state enumeration.

Value	Description	Typical Response
Abundant	< 50% consumed	Normal operation
Normal	50-75% consumed	Monitor usage
Scarce	75-90% consumed	Reduce exploration
Critical	> 90% consumed	Minimal operations only

---

BudgetStatus

Current budget state returned from checks.

Properties

Property	Type	Description
Level	ScarcityLevel	Current scarcity level
StepsRemaining	int	Steps until limit
TokensRemaining	int	Tokens until limit
TimeRemaining	TimeSpan	Time until limit
IsExhausted	bool	Whether any budget is exhausted

---

Registration

Complete infrastructure setup:

services.AddStrategos()
    .AddThompsonSampling(options => options
        .WithPrior(alpha: 2, beta: 2)
        .WithBeliefStore<InMemoryBeliefStore>())
    .AddLoopDetection(options => options
        .AddExactRepetition()
        .AddSemanticRepetition()
        .AddOscillation()
        .AddNoProgress())
    .AddBudgetGuard(options => options
        .WithMaxSteps(100)
        .WithMaxTokens(50_000)
        .WithMaxWallTime(TimeSpan.FromMinutes(30)));