The ultimate harness for coding agents and beyond. All your agents. All your machines. One command center. Start with code, then follow your curiosity and build across disciplines: CAD, circuits, robots, games and music.
TypeScript 3d-modelingagent-frameworkagent-harnessai-agentscadclaude-codecodexcoding-agentsdeveloper-toolsesp32game-developmentmcp-servermulti-agentopen-hardwareopencodeopenharnesspcbroboticsterminaltmux
import { codexGoalOf, codexTaskBoundary, startsCodexTurn } from './normalizer.js'
import type { AttachRules } from '../../lib/attachTranscript.js'
import type { RuntimeField } from '../../lib/runtimeProfile.js'
const bytes = (...markers: string[]): Buffer[] => markers.map((marker) => Buffer.from(marker))
/**
* Codex: a turn opens on a user message (`user_message`, or `UserMessage` inside `item_completed`) or a
* `/goal` injection, and its task began at the `task_started` before it. Model, effort and mode arrive
* together in `turn_context` and `thread_settings_applied`, written just before the turn's first
* message. A continuing goal's label depends on the goal before it.
*/
export function codexAttachRules(fields: (line: string) => readonly RuntimeField[]): AttachRules {
return {
startsTurn: startsCodexTurn,
turnMarkers: bytes('user_message', 'UserMessage', 'codex_internal_context'),
turnBegin: {
begins: (line) => codexTaskBoundary(line) === 'begins',
ends: (line) => codexTaskBoundary(line) === 'ends',
markers: bytes('task_started', 'task_complete', 'turn_aborted'),
},
fields,
fieldMarkers: bytes('turn_context', 'thread_settings_applied'),
required: ['model', 'effort', 'mode'],
seedFor: (opener) => (codexGoalOf(opener) === null ? null : (line) => codexGoalOf(line) !== null),
seedMarkers: bytes('codex_internal_context'),
}
}