koog-agents-optimization/ai.koog.agents.optimization.optimizers.ace/ACEPlaybook

ACEPlaybook

class ACEPlaybook(storagePath: ResilientPath)

Mutable collection of curated PlaybookSections persisted to storagePath.

The playbook is the artifact optimized by ACEOptimizer: insights are accumulated as bullets, refined via DeltaUpdates, and injected into the agent's system prompt at run time. Backed by a JSON file that is loaded with load and persisted with save.

Constructors

ACEPlaybook

constructor(storagePath: ResilientPath)

Types

UpdatedBullets

data class UpdatedBullets(val helpfulBullets: List<String>, val harmfulBullets: List<String>, val neutralBullets: List<String>, val notFoundBullets: List<String>, val unknownBulletTags: List<ACEOptimizer.BulletTag>)

Outcome of an updateCounters call, partitioning the processed bullet tags by how they were handled.

Functions

applyDelta

fun applyDelta(delta: DeltaUpdate): Boolean

Returns true if delta was applied; false otherwise.

load

fun load(): ACEPlaybook

Reads the playbook contents from storagePath into this instance and returns it.

save

fun save()

Persists the current sections to storagePath as pretty-printed JSON, creating parent directories as needed.

toJson

fun toJson(json: Json): String

Serializes the playbook sections to a JSON string using the provided json instance.

toPromptRepresentation

fun toPromptRepresentation(): String

Renders the playbook as a human-readable, sectioned text block for embedding in an LLM prompt.

updateCounters

fun updateCounters(bulletTags: List<ACEOptimizer.BulletTag>): ACEPlaybook.UpdatedBullets

Applies bulletTags to the playbook, incrementing each referenced bullet's helpful/harmful counter, and returns a breakdown of what happened as UpdatedBullets.