coding

typechecks, compiles, and (by default) has a node main.

This agent's contract ends at VALID source. To also check the program's behavior, compose it with agencyVerifierAgent: generate source here, run agencyVerifierAgent on the result, and retry with its findings folded into the task.

Types

WriteFailure

A failed generation: the last source attempted, with the problems found in it (review findings, compile errors, or the reason generation stopped).

/** A failed generation: the last source attempted, with the problems found
  in it (review findings, compile errors, or the reason generation stopped). */
export type WriteFailure = {
  source: string;
  problems: string[]
}

(source)

Functions

buildTools

buildTools(strict: boolean = false, projectTools: boolean = true): any[]

Return the Agency writer's tools: the bundled documentation, the source inspectors it checks drafts with, read-only access to the project it is writing for, and fetches for a named external resource. Not testFile: the writer returns source it has not saved, so there is nothing on disk for a harness to test yet.

@param strict - Offer a typecheck that reports what a sandboxed compile would, so the writer sees those errors on its own draft @param projectTools - False leaves out the file and git tools, for a draft that depends on nothing in the project

Parameters:

Name	Type	Default
strict	boolean	false
projectTools	boolean	true

Returns: any[]

(source)

agencyCodingAgent

agencyCodingAgent(
  task: string,
  context: string = "",
  maxAttempts: number = 3,
  maxCost: number = $20.00,
  maxTime: number = 15m,
  model: string = "",
  provider: string = "",
  session: string = "",
  extraTools: any[] = [],
  requireMain: boolean = true,
  dir: string = "",
  strict: boolean = false,
  projectTools: boolean = true,
): Result<string, WriteFailure>

Write an Agency program for the task. Iterates until the source parses, typechecks, compiles, and (when requireMain) has a node main, or attempts run out. On failure returns the last attempted source with the problems found in it.

@param task - What the program should do @param context - Extra material folded into the prompt, or "" @param maxAttempts - Generation attempts before giving up @param maxCost - Hard spend cap @param maxTime - Hard wall-clock cap @param model - Model override, or "" for the ambient model @param provider - Provider for the model override @param session - Session name to share a thread across calls, or "" for isolated @param extraTools - Extra tools to offer the LLM, appended to the built-in set @param requireMain - Demand a node main entry point (default). Pass false when the deliverable is a library module whose exports the caller places. @param dir - Directory the draft's relative imports resolve against, so a draft that imports a sibling file typechecks. "" when it imports none. @param strict - Check each draft the way a sandboxed (--agency-only) compile would, and give the writer a typecheck that does the same. Pass true when the deliverable will run sandboxed. @param projectTools - False leaves out the writer's file and git tools. Pass false when nothing in the caller's project bears on the task.

Parameters:

Name	Type	Default
task	string
context	string	""
maxAttempts	number	3
maxCost	number	$20.00
maxTime	number	15m
model	string	""
provider	string	""
session	string	""
extraTools	any[]	[]
requireMain	boolean	true
dir	string	""
strict	boolean	false
projectTools	boolean	true

Returns: Result<string, WriteFailure>

Throws: std::guard

(source)