shared

overrides, and the handler an eval entry node runs its worker under.

Types

ReasoningEffort

export type ReasoningEffort = "low" | "medium" | "high"

(source)

Functions

docsOnlyHandler

docsOnlyHandler(intr)

Approves reads of the docs that ship in this package and rejects every other interrupt. Run a worker under it when it may look up syntax but must not write, run commands, or escape a budget trip. The stdlib eval entry nodes use it.

Parameters:

Name	Type	Default
intr

(source)

withContext

withContext(task: string, context: string): string

Fold optional context material into a task prompt. Returns the task unchanged when no context is given.

@param task - The task text @param context - Extra material, or "" for none

Parameters:

Name	Type	Default
task	string
context	string

Returns: string

(source)

llmOptions

llmOptions(
  model: string,
  provider: string,
  tools: any[] = [],
  hostedTools: string[] = [],
  reasoningEffort: ReasoningEffort | null = null,
  thinking: boolean = false,
  validationRetries: number = 0,
  maxToolCallRounds: number = 0,
): any

Build an llm options object with an optional model override. Returns a fresh object every call and never mutates its arguments. The model and provider fields are only set when a model is named, so an empty override never clobbers the default model.

@param model - Model name, or "" for the ambient model @param provider - Provider for the model, or "" to auto-resolve @param tools - Tools to offer the LLM @param hostedTools - Provider-hosted tools to enable @param reasoningEffort - Reasoning effort to request from the model, or null for none @param thinking - Whether to enable thinking mode, which allows the model to use more time and tokens to reason about its answer @param validationRetries - How many times to resend a structured-output call whose answer failed schema validation, or 0 for none @param maxToolCallRounds - Tool-call rounds the call may use, or 0 for the configured default

Parameters:

Name	Type	Default
model	string
provider	string
tools	any[]	[]
hostedTools	string[]	[]
reasoningEffort	ReasoningEffort | null	null
thinking	boolean	false
validationRetries	number	0
maxToolCallRounds	number	0

Returns: any

(source)