@tangle-network/agent-runtime 0.20.4 → 0.21.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["../src/loops/drivers/refine.ts","../src/loops/run-loop.ts"],"sourcesContent":["/**\n * @experimental\n *\n * Refine driver — single task per iteration, validator-gated.\n *\n * `plan` returns `[task]` (possibly transformed via `refineTask`) until the\n * prior verdict is valid OR the local cap is hit, then `[]`.\n * `decide` returns `'stop'` once the latest verdict is valid OR the cap is\n * reached. The kernel's `maxIterations` is an orthogonal safety cap;\n * whichever is lower wins.\n */\n\nimport { ValidationError } from '../../errors'\nimport type { DefaultVerdict, Driver, Iteration } from '../types'\n\nexport type RefineDecision = 'continue' | 'stop'\n\n/** @experimental */\nexport interface CreateRefineDriverOptions<Task> {\n /** Hard cap on iterations. Default 5. */\n maxIterations?: number\n /**\n * Optional task transform applied each round based on the prior verdict.\n * When omitted, the same task is replayed and the agent is expected to\n * inspect the sandbox session state for prior attempts.\n */\n refineTask?: (task: Task, prior: DefaultVerdict) => Task\n /** Stable identifier surfaced in trace events. Default `'refine'`. */\n name?: string\n}\n\n/** @experimental */\nexport function createRefineDriver<Task, Output>(\n options: CreateRefineDriverOptions<Task> = {},\n): Driver<Task, Output, RefineDecision> {\n const maxIterations = options.maxIterations ?? 5\n if (!Number.isFinite(maxIterations) || maxIterations <= 0) {\n throw new ValidationError('createRefineDriver: maxIterations must be > 0')\n }\n const refineTask = options.refineTask\n return {\n name: options.name ?? 'refine',\n async plan(task, history) {\n if (history.length >= maxIterations) return []\n if (history.length === 0) return [task]\n const prior = history.at(-1)\n if (!prior) return [task]\n if (prior.verdict?.valid === true) return []\n // Worker error: replay the same task so the agent can self-correct.\n // The driver has no signal beyond `verdict`; only the validator\n // controls \"good enough\".\n if (!refineTask || !prior.verdict) return [prior.task]\n return [refineTask(prior.task, prior.verdict)]\n },\n decide(history) {\n const last = history.at(-1)\n if (!last) return 'continue'\n if (last.verdict?.valid === true) return 'stop'\n if (history.length >= maxIterations) return 'stop'\n return 'continue'\n },\n }\n}\n\n/**\n * Test helper: select the last-valid iteration (or the last attempt if\n * none passed). Mirrors the kernel's default selector ordering for refine\n * topologies — the most recent successful attempt wins.\n *\n * @experimental\n */\nexport function refineWinnerIndex<Task, Output>(\n iterations: ReadonlyArray<Iteration<Task, Output>>,\n): number | undefined {\n for (let i = iterations.length - 1; i >= 0; i -= 1) {\n if (iterations[i]?.verdict?.valid) return i\n }\n return iterations.length > 0 ? iterations.length - 1 : undefined\n}\n","/**\n * @experimental\n *\n * `runLoop` — the topology-agnostic kernel built atop the sandbox SDK.\n *\n * Each iteration:\n * 1. `driver.plan(task, history)` → N tasks (1 = refine, N = fanout, 0 = stop)\n * 2. For each task (parallel, bounded by `maxConcurrency`):\n * a. round-robin an `AgentRunSpec` from `agentRuns`\n * b. `sandboxClient.create({ backend: { profile }, ...overrides })`\n * c. emit `loop.iteration.dispatch` with the placement\n * (`{ sibling, sandboxId }` or `{ fleet, fleetId, machineId, sandboxId }`)\n * d. iterate `box.streamPrompt(taskToPrompt(task))` and collect events\n * 3. `output.parse(events)` → typed `Output`\n * 4. `validator?.validate(output)` → `DefaultVerdict`\n * 5. Append `Iteration` to history; emit `loop.iteration.ended`\n * 6. `driver.decide(history)` → if terminal, return result + winner\n *\n * The kernel owns: iteration accounting, per-iteration timing, error\n * capture, abort propagation, concurrency cap, cost aggregation, and trace\n * emission. The kernel does NOT own: what the agent runs (sandbox SDK +\n * profile), how outputs are decoded (output adapter), how outputs are\n * scored (validator), or topology (driver).\n */\n\nimport type {\n AgentProfile,\n CreateSandboxOptions,\n SandboxEvent,\n SandboxInstance,\n} from '@tangle-network/sandbox'\nimport { ValidationError } from '../errors'\nimport type { RuntimeStreamEvent } from '../types'\nimport type {\n AgentRunSpec,\n Driver,\n ExecCtx,\n Iteration,\n LoopResult,\n LoopSandboxClient,\n LoopSandboxPlacement,\n LoopTraceEmitter,\n LoopTraceEvent,\n LoopWinner,\n OutputAdapter,\n Validator,\n} from './types'\n\nconst DEFAULT_MAX_ITERATIONS = 10\nconst DEFAULT_MAX_CONCURRENCY = 4\n\n/** @experimental */\nexport interface RunLoopOptions<Task, Output, Decision> {\n driver: Driver<Task, Output, Decision>\n /**\n * Single agent spec — every iteration uses this profile. Mutually\n * exclusive with `agentRuns`.\n */\n agentRun?: AgentRunSpec<Task>\n /**\n * Multiple specs for heterogeneous fanout. The kernel round-robins\n * through them when the driver plans N tasks. Mutually exclusive with\n * `agentRun`.\n */\n agentRuns?: AgentRunSpec<Task>[]\n output: OutputAdapter<Output>\n validator?: Validator<Output>\n task: Task\n ctx: ExecCtx\n /** Default 10. Hard cap on total iterations across all `plan()` rounds. */\n maxIterations?: number\n /** Default 4. In-flight worker cap within a single `plan()` batch. */\n maxConcurrency?: number\n /**\n * Pre-allocated id for trace correlation. Default = `loop-${random}`.\n * Surfaces as `runId` on every emitted `LoopTraceEvent`.\n */\n runId?: string\n /**\n * Clock override; default `Date.now`. Deterministic tests pass a\n * monotonic counter to stabilize iteration timing fields.\n */\n now?: () => number\n /**\n * Override the default winner selector (highest-valid-score, ties broken\n * by earliest iteration).\n */\n selectWinner?: (iterations: Iteration<Task, Output>[]) => LoopWinner<Task, Output> | undefined\n}\n\n/** @experimental */\nexport async function runLoop<Task, Output, Decision>(\n options: RunLoopOptions<Task, Output, Decision>,\n): Promise<LoopResult<Task, Output, Decision>> {\n const specs = resolveAgentRuns(options)\n const maxIterations = options.maxIterations ?? DEFAULT_MAX_ITERATIONS\n if (!Number.isFinite(maxIterations) || maxIterations <= 0) {\n throw new ValidationError('runLoop: maxIterations must be > 0')\n }\n const maxConcurrency = options.maxConcurrency ?? DEFAULT_MAX_CONCURRENCY\n if (!Number.isFinite(maxConcurrency) || maxConcurrency <= 0) {\n throw new ValidationError('runLoop: maxConcurrency must be > 0')\n }\n if (!options.ctx?.sandboxClient || typeof options.ctx.sandboxClient.create !== 'function') {\n throw new ValidationError('runLoop: ctx.sandboxClient.create is required')\n }\n const now = options.now ?? Date.now\n const runId = options.runId ?? `loop-${randomSuffix()}`\n const loopStart = now()\n const driverName = options.driver.name ?? 'driver'\n const iterations: Iteration<Task, Output>[] = []\n\n await emitTrace(options.ctx.traceEmitter, {\n kind: 'loop.started',\n runId,\n timestamp: now(),\n payload: {\n driver: driverName,\n agentRunNames: specs.map((spec) => spec.name ?? spec.profile.name ?? 'agent'),\n maxIterations,\n maxConcurrency,\n },\n })\n\n const controller = new AbortController()\n const onOuterAbort = () => controller.abort()\n if (options.ctx.signal) {\n if (options.ctx.signal.aborted) controller.abort()\n else options.ctx.signal.addEventListener('abort', onOuterAbort, { once: true })\n }\n\n try {\n while (iterations.length < maxIterations) {\n if (controller.signal.aborted) throwAbort()\n const planned = await options.driver.plan(options.task, iterations)\n if (planned.length === 0) break\n\n const remaining = maxIterations - iterations.length\n const slice = planned.slice(0, remaining)\n const baseIndex = iterations.length\n // Reserve slots up front so concurrent workers may mutate by index.\n for (let i = 0; i < slice.length; i += 1) {\n const spec = specs[(baseIndex + i) % specs.length]!\n iterations.push({\n index: baseIndex + i,\n task: slice[i] as Task,\n agentRunName: spec.name ?? spec.profile.name ?? 'agent',\n events: [],\n startedAt: now(),\n endedAt: 0,\n costUsd: 0,\n })\n }\n\n await runBatch({\n slice,\n baseIndex,\n iterations,\n specs,\n output: options.output,\n validator: options.validator,\n maxConcurrency,\n signal: controller.signal,\n ctx: options.ctx,\n runId,\n now,\n })\n\n if (controller.signal.aborted) throwAbort()\n\n const decision = await options.driver.decide(iterations)\n await emitTrace(options.ctx.traceEmitter, {\n kind: 'loop.decision',\n runId,\n timestamp: now(),\n payload: { decision: serializeDecision(decision), historyLength: iterations.length },\n })\n if (isTerminalDecision(decision)) {\n return finalize({\n options,\n decision,\n iterations,\n startMs: loopStart,\n now,\n runId,\n })\n }\n }\n\n if (iterations.length >= maxIterations) {\n // Cap reached without a terminal decision — ask the driver one more time\n // for its final state, then close out.\n const decision = await options.driver.decide(iterations)\n await emitTrace(options.ctx.traceEmitter, {\n kind: 'loop.decision',\n runId,\n timestamp: now(),\n payload: { decision: serializeDecision(decision), historyLength: iterations.length },\n })\n return finalize({ options, decision, iterations, startMs: loopStart, now, runId })\n }\n // `plan()` returned `[]` before `decide()` reached a terminal state.\n const decision = await options.driver.decide(iterations)\n await emitTrace(options.ctx.traceEmitter, {\n kind: 'loop.decision',\n runId,\n timestamp: now(),\n payload: { decision: serializeDecision(decision), historyLength: iterations.length },\n })\n return finalize({ options, decision, iterations, startMs: loopStart, now, runId })\n } finally {\n if (options.ctx.signal) options.ctx.signal.removeEventListener('abort', onOuterAbort)\n }\n}\n\ninterface RunBatchArgs<Task, Output> {\n slice: Task[]\n baseIndex: number\n iterations: Iteration<Task, Output>[]\n specs: AgentRunSpec<Task>[]\n output: OutputAdapter<Output>\n validator: Validator<Output> | undefined\n maxConcurrency: number\n signal: AbortSignal\n ctx: ExecCtx\n runId: string\n now: () => number\n}\n\nasync function runBatch<Task, Output>(args: RunBatchArgs<Task, Output>) {\n const queue = args.slice.map((task, offset) => ({ task, index: args.baseIndex + offset }))\n const inflight = new Set<Promise<void>>()\n while (queue.length > 0 || inflight.size > 0) {\n while (inflight.size < args.maxConcurrency && queue.length > 0) {\n const item = queue.shift()!\n const p = executeIteration({ ...args, item }).finally(() => inflight.delete(p))\n inflight.add(p)\n }\n if (inflight.size === 0) break\n await Promise.race(inflight)\n }\n}\n\ninterface ExecuteIterationArgs<Task, Output> extends RunBatchArgs<Task, Output> {\n item: { task: Task; index: number }\n}\n\nasync function executeIteration<Task, Output>(args: ExecuteIterationArgs<Task, Output>) {\n const slot = args.iterations[args.item.index]\n if (!slot)\n throw new ValidationError(`runLoop: missing iteration slot at index ${args.item.index}`)\n const spec = args.specs[args.item.index % args.specs.length]\n if (!spec) throw new ValidationError('runLoop: no AgentRunSpec available for iteration')\n slot.startedAt = args.now()\n slot.agentRunName = spec.name ?? spec.profile.name ?? 'agent'\n\n await emitTrace(args.ctx.traceEmitter, {\n kind: 'loop.iteration.started',\n runId: args.runId,\n timestamp: args.now(),\n payload: {\n iterationIndex: args.item.index,\n agentRunName: slot.agentRunName,\n taskHash: hashJson(args.item.task),\n },\n })\n\n try {\n const box = await createSandboxForSpec(args.ctx.sandboxClient, spec, args.signal)\n const placement = describePlacementSafe(args.ctx.sandboxClient, box)\n await emitTrace(args.ctx.traceEmitter, {\n kind: 'loop.iteration.dispatch',\n runId: args.runId,\n timestamp: args.now(),\n payload: {\n iterationIndex: args.item.index,\n agentRunName: slot.agentRunName,\n placement: placement.kind,\n sandboxId: placement.sandboxId,\n fleetId: placement.fleetId,\n machineId: placement.machineId,\n },\n })\n const message = spec.taskToPrompt(args.item.task)\n const events: SandboxEvent[] = []\n for await (const event of box.streamPrompt(message, { signal: args.signal })) {\n events.push(event)\n const llmCall = extractLlmCallEvent(event, slot.agentRunName)\n if (llmCall) {\n slot.costUsd += llmCall.costUsd ?? 0\n args.ctx.runHandle?.observe(llmCall)\n }\n }\n slot.events = events\n slot.output = args.output.parse(events)\n if (args.validator) {\n slot.verdict = await args.validator.validate(slot.output, {\n iteration: args.item.index,\n signal: args.signal,\n })\n }\n } catch (err) {\n slot.error = err instanceof Error ? err : new Error(String(err))\n } finally {\n slot.endedAt = args.now()\n await emitTrace(args.ctx.traceEmitter, {\n kind: 'loop.iteration.ended',\n runId: args.runId,\n timestamp: args.now(),\n payload: {\n iterationIndex: args.item.index,\n agentRunName: slot.agentRunName,\n outputHash: slot.output !== undefined ? hashJson(slot.output) : undefined,\n verdict: slot.verdict,\n error: slot.error?.message,\n costUsd: slot.costUsd,\n durationMs: slot.endedAt - slot.startedAt,\n },\n })\n }\n}\n\nfunction describePlacementSafe(\n client: LoopSandboxClient,\n box: SandboxInstance,\n): LoopSandboxPlacement {\n if (typeof client.describePlacement === 'function') {\n try {\n const result = client.describePlacement(box)\n if (\n result &&\n typeof result === 'object' &&\n (result.kind === 'sibling' || result.kind === 'fleet')\n ) {\n return {\n kind: result.kind,\n sandboxId: result.sandboxId ?? readSandboxId(box),\n fleetId: result.fleetId,\n machineId: result.machineId,\n }\n }\n } catch {\n // Adapter bug must not corrupt the iteration; fall through to default.\n }\n }\n return { kind: 'sibling', sandboxId: readSandboxId(box) }\n}\n\nfunction readSandboxId(box: SandboxInstance): string | undefined {\n const raw = (box as unknown as { id?: unknown }).id\n return typeof raw === 'string' && raw.length > 0 ? raw : undefined\n}\n\nasync function createSandboxForSpec<Task>(\n client: LoopSandboxClient,\n spec: AgentRunSpec<Task>,\n signal: AbortSignal,\n): Promise<SandboxInstance> {\n const overrides = spec.sandboxOverrides ?? {}\n const overrideBackend = overrides.backend\n const opts: CreateSandboxOptions = {\n ...overrides,\n backend: {\n type: overrideBackend?.type ?? inferBackendType(spec.profile),\n profile: spec.profile satisfies AgentProfile,\n ...(overrideBackend?.model ? { model: overrideBackend.model } : {}),\n ...(overrideBackend?.server ? { server: overrideBackend.server } : {}),\n },\n }\n // Cooperative cancellation: if the abort signal fires while .create is\n // pending, the promise itself is not abortable but the inflight prompt is.\n if (signal.aborted) throwAbort()\n return client.create(opts)\n}\n\nfunction inferBackendType(\n profile: AgentProfile,\n): CreateSandboxOptions['backend'] extends infer B\n ? B extends { type: infer T }\n ? T\n : never\n : never {\n // The sandbox SDK accepts profile-driven backend selection by name. When the\n // profile has no explicit hint we fall through to the SDK's default\n // ('opencode' on the platform side). Returning a literal here would lie\n // about provenance — let the SDK pick.\n type BackendType = NonNullable<CreateSandboxOptions['backend']>['type']\n const explicit = profile.metadata?.backendType\n if (typeof explicit === 'string') return explicit as BackendType\n return 'opencode' as BackendType\n}\n\ninterface FinalizeArgs<Task, Output, Decision> {\n options: RunLoopOptions<Task, Output, Decision>\n decision: Decision\n iterations: Iteration<Task, Output>[]\n startMs: number\n now: () => number\n runId: string\n}\n\nfunction finalize<Task, Output, Decision>(\n args: FinalizeArgs<Task, Output, Decision>,\n): LoopResult<Task, Output, Decision> {\n const winner = (args.options.selectWinner ?? defaultSelectWinner)(args.iterations)\n const costUsd = args.iterations.reduce((sum, iter) => sum + (iter.costUsd || 0), 0)\n const result: LoopResult<Task, Output, Decision> = {\n decision: args.decision,\n iterations: args.iterations,\n winner,\n durationMs: args.now() - args.startMs,\n costUsd,\n }\n void emitTrace(args.options.ctx.traceEmitter, {\n kind: 'loop.ended',\n runId: args.runId,\n timestamp: args.now(),\n payload: {\n winnerIterationIndex: winner?.iterationIndex,\n totalCostUsd: costUsd,\n durationMs: result.durationMs,\n iterations: args.iterations.length,\n },\n })\n return result\n}\n\nfunction defaultSelectWinner<Task, Output>(\n iterations: Iteration<Task, Output>[],\n): LoopWinner<Task, Output> | undefined {\n const candidates = iterations.filter((iter) => iter.output !== undefined && !iter.error)\n if (candidates.length === 0) return undefined\n const valid = candidates.filter((iter) => iter.verdict?.valid === true)\n const pool = valid.length > 0 ? valid : candidates\n const sorted = [...pool].sort(\n (a, b) => (b.verdict?.score ?? 0) - (a.verdict?.score ?? 0) || a.index - b.index,\n )\n const top = sorted[0]\n if (!top || top.output === undefined) return undefined\n return {\n task: top.task,\n output: top.output,\n verdict: top.verdict,\n iterationIndex: top.index,\n agentRunName: top.agentRunName,\n }\n}\n\nfunction resolveAgentRuns<Task, Output, Decision>(\n options: RunLoopOptions<Task, Output, Decision>,\n): AgentRunSpec<Task>[] {\n if (options.agentRun && options.agentRuns) {\n throw new ValidationError('runLoop: pass exactly one of `agentRun` or `agentRuns`')\n }\n if (options.agentRun) return [options.agentRun]\n if (options.agentRuns && options.agentRuns.length > 0) return options.agentRuns\n throw new ValidationError('runLoop: `agentRun` or non-empty `agentRuns` is required')\n}\n\nfunction isTerminalDecision(decision: unknown): boolean {\n return (\n decision === 'stop' || decision === 'pick-winner' || decision === 'fail' || decision === 'done'\n )\n}\n\nfunction serializeDecision(decision: unknown): string {\n if (typeof decision === 'string') return decision\n if (decision === null || decision === undefined) return 'null'\n try {\n return JSON.stringify(decision)\n } catch {\n return String(decision)\n }\n}\n\nasync function emitTrace(\n emitter: LoopTraceEmitter | undefined,\n event: LoopTraceEvent,\n): Promise<void> {\n if (!emitter) return\n await emitter.emit(event)\n}\n\nfunction randomSuffix(len = 8): string {\n return Math.random()\n .toString(36)\n .slice(2, 2 + len)\n}\n\nfunction throwAbort(): never {\n const err = new Error('aborted')\n err.name = 'AbortError'\n throw err\n}\n\n/**\n * Extract a `RuntimeStreamEvent`-shaped `llm_call` from a sandbox event when\n * the event carries usage/cost data. Returns `undefined` for non-cost events\n * so the kernel can iterate the full stream without branching.\n *\n * Sandbox SDK emits a polymorphic `SandboxEvent = { type, data, id? }`. The\n * canonical cost-carrying types observed in the wild:\n * - `llm_call` — `data: { model, tokensIn, tokensOut, costUsd, ... }`\n * - `message.completed` / `result` — `data: { usage: { inputTokens,\n * outputTokens, totalCostUsd? } }`\n * - `cost.usage` — same shape under a dedicated type\n *\n * Numeric coercion is strict: `Number.isFinite` gates every accumulator\n * write so a sentinel `NaN` from a misbehaving backend cannot poison the\n * ledger.\n */\nfunction extractLlmCallEvent(\n event: SandboxEvent,\n agentRunName: string,\n): (RuntimeStreamEvent & { type: 'llm_call' }) | undefined {\n if (!event || typeof event !== 'object') return undefined\n const type = String(event.type ?? '')\n const data =\n event.data && typeof event.data === 'object'\n ? (event.data as Record<string, unknown>)\n : ({} as Record<string, unknown>)\n\n if (type === 'llm_call' || type === 'cost.usage' || type === 'usage') {\n return buildLlmCall(data, agentRunName)\n }\n if (type === 'message.completed' || type === 'result' || type === 'final') {\n const usage = data.usage as Record<string, unknown> | undefined\n if (!usage || typeof usage !== 'object') return undefined\n return buildLlmCall({ ...usage, model: data.model ?? usage.model }, agentRunName)\n }\n return undefined\n}\n\nfunction buildLlmCall(\n data: Record<string, unknown>,\n agentRunName: string,\n): (RuntimeStreamEvent & { type: 'llm_call' }) | undefined {\n const tokensIn = pickFiniteNumber(data, ['tokensIn', 'inputTokens', 'prompt_tokens'])\n const tokensOut = pickFiniteNumber(data, ['tokensOut', 'outputTokens', 'completion_tokens'])\n const costUsd = pickFiniteNumber(data, ['costUsd', 'totalCostUsd', 'cost_usd', 'cost'])\n if (tokensIn === undefined && tokensOut === undefined && costUsd === undefined) {\n return undefined\n }\n const model = typeof data.model === 'string' && data.model.length > 0 ? data.model : agentRunName\n const event: RuntimeStreamEvent & { type: 'llm_call' } = {\n type: 'llm_call',\n model,\n }\n if (tokensIn !== undefined) event.tokensIn = tokensIn\n if (tokensOut !== undefined) event.tokensOut = tokensOut\n if (costUsd !== undefined) event.costUsd = costUsd\n return event\n}\n\nfunction pickFiniteNumber(data: Record<string, unknown>, keys: string[]): number | undefined {\n for (const key of keys) {\n const value = data[key]\n if (typeof value === 'number' && Number.isFinite(value)) return value\n }\n return undefined\n}\n\n/**\n * Stable hash for the trace payload. Not cryptographic — only used so\n * downstream eval pipelines can group iterations whose task / output is the\n * same. Bare structural hash; non-JSON values stringify via their `toString`.\n */\nfunction hashJson(value: unknown): string {\n let str: string\n try {\n str = JSON.stringify(value) ?? String(value)\n } catch {\n str = String(value)\n }\n // FNV-1a 32-bit — branch-free, dependency-free, good enough for grouping.\n let h = 0x811c9dc5\n for (let i = 0; i < str.length; i += 1) {\n h ^= str.charCodeAt(i)\n h = Math.imul(h, 0x01000193)\n }\n return (h >>> 0).toString(16).padStart(8, '0')\n}\n"],"mappings":";;;;;AAgCO,SAAS,mBACd,UAA2C,CAAC,GACN;AACtC,QAAM,gBAAgB,QAAQ,iBAAiB;AAC/C,MAAI,CAAC,OAAO,SAAS,aAAa,KAAK,iBAAiB,GAAG;AACzD,UAAM,IAAI,gBAAgB,+CAA+C;AAAA,EAC3E;AACA,QAAM,aAAa,QAAQ;AAC3B,SAAO;AAAA,IACL,MAAM,QAAQ,QAAQ;AAAA,IACtB,MAAM,KAAK,MAAM,SAAS;AACxB,UAAI,QAAQ,UAAU,cAAe,QAAO,CAAC;AAC7C,UAAI,QAAQ,WAAW,EAAG,QAAO,CAAC,IAAI;AACtC,YAAM,QAAQ,QAAQ,GAAG,EAAE;AAC3B,UAAI,CAAC,MAAO,QAAO,CAAC,IAAI;AACxB,UAAI,MAAM,SAAS,UAAU,KAAM,QAAO,CAAC;AAI3C,UAAI,CAAC,cAAc,CAAC,MAAM,QAAS,QAAO,CAAC,MAAM,IAAI;AACrD,aAAO,CAAC,WAAW,MAAM,MAAM,MAAM,OAAO,CAAC;AAAA,IAC/C;AAAA,IACA,OAAO,SAAS;AACd,YAAM,OAAO,QAAQ,GAAG,EAAE;AAC1B,UAAI,CAAC,KAAM,QAAO;AAClB,UAAI,KAAK,SAAS,UAAU,KAAM,QAAO;AACzC,UAAI,QAAQ,UAAU,cAAe,QAAO;AAC5C,aAAO;AAAA,IACT;AAAA,EACF;AACF;AASO,SAAS,kBACd,YACoB;AACpB,WAAS,IAAI,WAAW,SAAS,GAAG,KAAK,GAAG,KAAK,GAAG;AAClD,QAAI,WAAW,CAAC,GAAG,SAAS,MAAO,QAAO;AAAA,EAC5C;AACA,SAAO,WAAW,SAAS,IAAI,WAAW,SAAS,IAAI;AACzD;;;AC9BA,IAAM,yBAAyB;AAC/B,IAAM,0BAA0B;AA0ChC,eAAsB,QACpB,SAC6C;AAC7C,QAAM,QAAQ,iBAAiB,OAAO;AACtC,QAAM,gBAAgB,QAAQ,iBAAiB;AAC/C,MAAI,CAAC,OAAO,SAAS,aAAa,KAAK,iBAAiB,GAAG;AACzD,UAAM,IAAI,gBAAgB,oCAAoC;AAAA,EAChE;AACA,QAAM,iBAAiB,QAAQ,kBAAkB;AACjD,MAAI,CAAC,OAAO,SAAS,cAAc,KAAK,kBAAkB,GAAG;AAC3D,UAAM,IAAI,gBAAgB,qCAAqC;AAAA,EACjE;AACA,MAAI,CAAC,QAAQ,KAAK,iBAAiB,OAAO,QAAQ,IAAI,cAAc,WAAW,YAAY;AACzF,UAAM,IAAI,gBAAgB,+CAA+C;AAAA,EAC3E;AACA,QAAM,MAAM,QAAQ,OAAO,KAAK;AAChC,QAAM,QAAQ,QAAQ,SAAS,QAAQ,aAAa,CAAC;AACrD,QAAM,YAAY,IAAI;AACtB,QAAM,aAAa,QAAQ,OAAO,QAAQ;AAC1C,QAAM,aAAwC,CAAC;AAE/C,QAAM,UAAU,QAAQ,IAAI,cAAc;AAAA,IACxC,MAAM;AAAA,IACN;AAAA,IACA,WAAW,IAAI;AAAA,IACf,SAAS;AAAA,MACP,QAAQ;AAAA,MACR,eAAe,MAAM,IAAI,CAAC,SAAS,KAAK,QAAQ,KAAK,QAAQ,QAAQ,OAAO;AAAA,MAC5E;AAAA,MACA;AAAA,IACF;AAAA,EACF,CAAC;AAED,QAAM,aAAa,IAAI,gBAAgB;AACvC,QAAM,eAAe,MAAM,WAAW,MAAM;AAC5C,MAAI,QAAQ,IAAI,QAAQ;AACtB,QAAI,QAAQ,IAAI,OAAO,QAAS,YAAW,MAAM;AAAA,QAC5C,SAAQ,IAAI,OAAO,iBAAiB,SAAS,cAAc,EAAE,MAAM,KAAK,CAAC;AAAA,EAChF;AAEA,MAAI;AACF,WAAO,WAAW,SAAS,eAAe;AACxC,UAAI,WAAW,OAAO,QAAS,YAAW;AAC1C,YAAM,UAAU,MAAM,QAAQ,OAAO,KAAK,QAAQ,MAAM,UAAU;AAClE,UAAI,QAAQ,WAAW,EAAG;AAE1B,YAAM,YAAY,gBAAgB,WAAW;AAC7C,YAAM,QAAQ,QAAQ,MAAM,GAAG,SAAS;AACxC,YAAM,YAAY,WAAW;AAE7B,eAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK,GAAG;AACxC,cAAM,OAAO,OAAO,YAAY,KAAK,MAAM,MAAM;AACjD,mBAAW,KAAK;AAAA,UACd,OAAO,YAAY;AAAA,UACnB,MAAM,MAAM,CAAC;AAAA,UACb,cAAc,KAAK,QAAQ,KAAK,QAAQ,QAAQ;AAAA,UAChD,QAAQ,CAAC;AAAA,UACT,WAAW,IAAI;AAAA,UACf,SAAS;AAAA,UACT,SAAS;AAAA,QACX,CAAC;AAAA,MACH;AAEA,YAAM,SAAS;AAAA,QACb;AAAA,QACA;AAAA,QACA;AAAA,QACA;AAAA,QACA,QAAQ,QAAQ;AAAA,QAChB,WAAW,QAAQ;AAAA,QACnB;AAAA,QACA,QAAQ,WAAW;AAAA,QACnB,KAAK,QAAQ;AAAA,QACb;AAAA,QACA;AAAA,MACF,CAAC;AAED,UAAI,WAAW,OAAO,QAAS,YAAW;AAE1C,YAAMA,YAAW,MAAM,QAAQ,OAAO,OAAO,UAAU;AACvD,YAAM,UAAU,QAAQ,IAAI,cAAc;AAAA,QACxC,MAAM;AAAA,QACN;AAAA,QACA,WAAW,IAAI;AAAA,QACf,SAAS,EAAE,UAAU,kBAAkBA,SAAQ,GAAG,eAAe,WAAW,OAAO;AAAA,MACrF,CAAC;AACD,UAAI,mBAAmBA,SAAQ,GAAG;AAChC,eAAO,SAAS;AAAA,UACd;AAAA,UACA,UAAAA;AAAA,UACA;AAAA,UACA,SAAS;AAAA,UACT;AAAA,UACA;AAAA,QACF,CAAC;AAAA,MACH;AAAA,IACF;AAEA,QAAI,WAAW,UAAU,eAAe;AAGtC,YAAMA,YAAW,MAAM,QAAQ,OAAO,OAAO,UAAU;AACvD,YAAM,UAAU,QAAQ,IAAI,cAAc;AAAA,QACxC,MAAM;AAAA,QACN;AAAA,QACA,WAAW,IAAI;AAAA,QACf,SAAS,EAAE,UAAU,kBAAkBA,SAAQ,GAAG,eAAe,WAAW,OAAO;AAAA,MACrF,CAAC;AACD,aAAO,SAAS,EAAE,SAAS,UAAAA,WAAU,YAAY,SAAS,WAAW,KAAK,MAAM,CAAC;AAAA,IACnF;AAEA,UAAM,WAAW,MAAM,QAAQ,OAAO,OAAO,UAAU;AACvD,UAAM,UAAU,QAAQ,IAAI,cAAc;AAAA,MACxC,MAAM;AAAA,MACN;AAAA,MACA,WAAW,IAAI;AAAA,MACf,SAAS,EAAE,UAAU,kBAAkB,QAAQ,GAAG,eAAe,WAAW,OAAO;AAAA,IACrF,CAAC;AACD,WAAO,SAAS,EAAE,SAAS,UAAU,YAAY,SAAS,WAAW,KAAK,MAAM,CAAC;AAAA,EACnF,UAAE;AACA,QAAI,QAAQ,IAAI,OAAQ,SAAQ,IAAI,OAAO,oBAAoB,SAAS,YAAY;AAAA,EACtF;AACF;AAgBA,eAAe,SAAuB,MAAkC;AACtE,QAAM,QAAQ,KAAK,MAAM,IAAI,CAAC,MAAM,YAAY,EAAE,MAAM,OAAO,KAAK,YAAY,OAAO,EAAE;AACzF,QAAM,WAAW,oBAAI,IAAmB;AACxC,SAAO,MAAM,SAAS,KAAK,SAAS,OAAO,GAAG;AAC5C,WAAO,SAAS,OAAO,KAAK,kBAAkB,MAAM,SAAS,GAAG;AAC9D,YAAM,OAAO,MAAM,MAAM;AACzB,YAAM,IAAI,iBAAiB,EAAE,GAAG,MAAM,KAAK,CAAC,EAAE,QAAQ,MAAM,SAAS,OAAO,CAAC,CAAC;AAC9E,eAAS,IAAI,CAAC;AAAA,IAChB;AACA,QAAI,SAAS,SAAS,EAAG;AACzB,UAAM,QAAQ,KAAK,QAAQ;AAAA,EAC7B;AACF;AAMA,eAAe,iBAA+B,MAA0C;AACtF,QAAM,OAAO,KAAK,WAAW,KAAK,KAAK,KAAK;AAC5C,MAAI,CAAC;AACH,UAAM,IAAI,gBAAgB,4CAA4C,KAAK,KAAK,KAAK,EAAE;AACzF,QAAM,OAAO,KAAK,MAAM,KAAK,KAAK,QAAQ,KAAK,MAAM,MAAM;AAC3D,MAAI,CAAC,KAAM,OAAM,IAAI,gBAAgB,kDAAkD;AACvF,OAAK,YAAY,KAAK,IAAI;AAC1B,OAAK,eAAe,KAAK,QAAQ,KAAK,QAAQ,QAAQ;AAEtD,QAAM,UAAU,KAAK,IAAI,cAAc;AAAA,IACrC,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ,WAAW,KAAK,IAAI;AAAA,IACpB,SAAS;AAAA,MACP,gBAAgB,KAAK,KAAK;AAAA,MAC1B,cAAc,KAAK;AAAA,MACnB,UAAU,SAAS,KAAK,KAAK,IAAI;AAAA,IACnC;AAAA,EACF,CAAC;AAED,MAAI;AACF,UAAM,MAAM,MAAM,qBAAqB,KAAK,IAAI,eAAe,MAAM,KAAK,MAAM;AAChF,UAAM,YAAY,sBAAsB,KAAK,IAAI,eAAe,GAAG;AACnE,UAAM,UAAU,KAAK,IAAI,cAAc;AAAA,MACrC,MAAM;AAAA,MACN,OAAO,KAAK;AAAA,MACZ,WAAW,KAAK,IAAI;AAAA,MACpB,SAAS;AAAA,QACP,gBAAgB,KAAK,KAAK;AAAA,QAC1B,cAAc,KAAK;AAAA,QACnB,WAAW,UAAU;AAAA,QACrB,WAAW,UAAU;AAAA,QACrB,SAAS,UAAU;AAAA,QACnB,WAAW,UAAU;AAAA,MACvB;AAAA,IACF,CAAC;AACD,UAAM,UAAU,KAAK,aAAa,KAAK,KAAK,IAAI;AAChD,UAAM,SAAyB,CAAC;AAChC,qBAAiB,SAAS,IAAI,aAAa,SAAS,EAAE,QAAQ,KAAK,OAAO,CAAC,GAAG;AAC5E,aAAO,KAAK,KAAK;AACjB,YAAM,UAAU,oBAAoB,OAAO,KAAK,YAAY;AAC5D,UAAI,SAAS;AACX,aAAK,WAAW,QAAQ,WAAW;AACnC,aAAK,IAAI,WAAW,QAAQ,OAAO;AAAA,MACrC;AAAA,IACF;AACA,SAAK,SAAS;AACd,SAAK,SAAS,KAAK,OAAO,MAAM,MAAM;AACtC,QAAI,KAAK,WAAW;AAClB,WAAK,UAAU,MAAM,KAAK,UAAU,SAAS,KAAK,QAAQ;AAAA,QACxD,WAAW,KAAK,KAAK;AAAA,QACrB,QAAQ,KAAK;AAAA,MACf,CAAC;AAAA,IACH;AAAA,EACF,SAAS,KAAK;AACZ,SAAK,QAAQ,eAAe,QAAQ,MAAM,IAAI,MAAM,OAAO,GAAG,CAAC;AAAA,EACjE,UAAE;AACA,SAAK,UAAU,KAAK,IAAI;AACxB,UAAM,UAAU,KAAK,IAAI,cAAc;AAAA,MACrC,MAAM;AAAA,MACN,OAAO,KAAK;AAAA,MACZ,WAAW,KAAK,IAAI;AAAA,MACpB,SAAS;AAAA,QACP,gBAAgB,KAAK,KAAK;AAAA,QAC1B,cAAc,KAAK;AAAA,QACnB,YAAY,KAAK,WAAW,SAAY,SAAS,KAAK,MAAM,IAAI;AAAA,QAChE,SAAS,KAAK;AAAA,QACd,OAAO,KAAK,OAAO;AAAA,QACnB,SAAS,KAAK;AAAA,QACd,YAAY,KAAK,UAAU,KAAK;AAAA,MAClC;AAAA,IACF,CAAC;AAAA,EACH;AACF;AAEA,SAAS,sBACP,QACA,KACsB;AACtB,MAAI,OAAO,OAAO,sBAAsB,YAAY;AAClD,QAAI;AACF,YAAM,SAAS,OAAO,kBAAkB,GAAG;AAC3C,UACE,UACA,OAAO,WAAW,aACjB,OAAO,SAAS,aAAa,OAAO,SAAS,UAC9C;AACA,eAAO;AAAA,UACL,MAAM,OAAO;AAAA,UACb,WAAW,OAAO,aAAa,cAAc,GAAG;AAAA,UAChD,SAAS,OAAO;AAAA,UAChB,WAAW,OAAO;AAAA,QACpB;AAAA,MACF;AAAA,IACF,QAAQ;AAAA,IAER;AAAA,EACF;AACA,SAAO,EAAE,MAAM,WAAW,WAAW,cAAc,GAAG,EAAE;AAC1D;AAEA,SAAS,cAAc,KAA0C;AAC/D,QAAM,MAAO,IAAoC;AACjD,SAAO,OAAO,QAAQ,YAAY,IAAI,SAAS,IAAI,MAAM;AAC3D;AAEA,eAAe,qBACb,QACA,MACA,QAC0B;AAC1B,QAAM,YAAY,KAAK,oBAAoB,CAAC;AAC5C,QAAM,kBAAkB,UAAU;AAClC,QAAM,OAA6B;AAAA,IACjC,GAAG;AAAA,IACH,SAAS;AAAA,MACP,MAAM,iBAAiB,QAAQ,iBAAiB,KAAK,OAAO;AAAA,MAC5D,SAAS,KAAK;AAAA,MACd,GAAI,iBAAiB,QAAQ,EAAE,OAAO,gBAAgB,MAAM,IAAI,CAAC;AAAA,MACjE,GAAI,iBAAiB,SAAS,EAAE,QAAQ,gBAAgB,OAAO,IAAI,CAAC;AAAA,IACtE;AAAA,EACF;AAGA,MAAI,OAAO,QAAS,YAAW;AAC/B,SAAO,OAAO,OAAO,IAAI;AAC3B;AAEA,SAAS,iBACP,SAKQ;AAMR,QAAM,WAAW,QAAQ,UAAU;AACnC,MAAI,OAAO,aAAa,SAAU,QAAO;AACzC,SAAO;AACT;AAWA,SAAS,SACP,MACoC;AACpC,QAAM,UAAU,KAAK,QAAQ,gBAAgB,qBAAqB,KAAK,UAAU;AACjF,QAAM,UAAU,KAAK,WAAW,OAAO,CAAC,KAAK,SAAS,OAAO,KAAK,WAAW,IAAI,CAAC;AAClF,QAAM,SAA6C;AAAA,IACjD,UAAU,KAAK;AAAA,IACf,YAAY,KAAK;AAAA,IACjB;AAAA,IACA,YAAY,KAAK,IAAI,IAAI,KAAK;AAAA,IAC9B;AAAA,EACF;AACA,OAAK,UAAU,KAAK,QAAQ,IAAI,cAAc;AAAA,IAC5C,MAAM;AAAA,IACN,OAAO,KAAK;AAAA,IACZ,WAAW,KAAK,IAAI;AAAA,IACpB,SAAS;AAAA,MACP,sBAAsB,QAAQ;AAAA,MAC9B,cAAc;AAAA,MACd,YAAY,OAAO;AAAA,MACnB,YAAY,KAAK,WAAW;AAAA,IAC9B;AAAA,EACF,CAAC;AACD,SAAO;AACT;AAEA,SAAS,oBACP,YACsC;AACtC,QAAM,aAAa,WAAW,OAAO,CAAC,SAAS,KAAK,WAAW,UAAa,CAAC,KAAK,KAAK;AACvF,MAAI,WAAW,WAAW,EAAG,QAAO;AACpC,QAAM,QAAQ,WAAW,OAAO,CAAC,SAAS,KAAK,SAAS,UAAU,IAAI;AACtE,QAAM,OAAO,MAAM,SAAS,IAAI,QAAQ;AACxC,QAAM,SAAS,CAAC,GAAG,IAAI,EAAE;AAAA,IACvB,CAAC,GAAG,OAAO,EAAE,SAAS,SAAS,MAAM,EAAE,SAAS,SAAS,MAAM,EAAE,QAAQ,EAAE;AAAA,EAC7E;AACA,QAAM,MAAM,OAAO,CAAC;AACpB,MAAI,CAAC,OAAO,IAAI,WAAW,OAAW,QAAO;AAC7C,SAAO;AAAA,IACL,MAAM,IAAI;AAAA,IACV,QAAQ,IAAI;AAAA,IACZ,SAAS,IAAI;AAAA,IACb,gBAAgB,IAAI;AAAA,IACpB,cAAc,IAAI;AAAA,EACpB;AACF;AAEA,SAAS,iBACP,SACsB;AACtB,MAAI,QAAQ,YAAY,QAAQ,WAAW;AACzC,UAAM,IAAI,gBAAgB,wDAAwD;AAAA,EACpF;AACA,MAAI,QAAQ,SAAU,QAAO,CAAC,QAAQ,QAAQ;AAC9C,MAAI,QAAQ,aAAa,QAAQ,UAAU,SAAS,EAAG,QAAO,QAAQ;AACtE,QAAM,IAAI,gBAAgB,0DAA0D;AACtF;AAEA,SAAS,mBAAmB,UAA4B;AACtD,SACE,aAAa,UAAU,aAAa,iBAAiB,aAAa,UAAU,aAAa;AAE7F;AAEA,SAAS,kBAAkB,UAA2B;AACpD,MAAI,OAAO,aAAa,SAAU,QAAO;AACzC,MAAI,aAAa,QAAQ,aAAa,OAAW,QAAO;AACxD,MAAI;AACF,WAAO,KAAK,UAAU,QAAQ;AAAA,EAChC,QAAQ;AACN,WAAO,OAAO,QAAQ;AAAA,EACxB;AACF;AAEA,eAAe,UACb,SACA,OACe;AACf,MAAI,CAAC,QAAS;AACd,QAAM,QAAQ,KAAK,KAAK;AAC1B;AAEA,SAAS,aAAa,MAAM,GAAW;AACrC,SAAO,KAAK,OAAO,EAChB,SAAS,EAAE,EACX,MAAM,GAAG,IAAI,GAAG;AACrB;AAEA,SAAS,aAAoB;AAC3B,QAAM,MAAM,IAAI,MAAM,SAAS;AAC/B,MAAI,OAAO;AACX,QAAM;AACR;AAkBA,SAAS,oBACP,OACA,cACyD;AACzD,MAAI,CAAC,SAAS,OAAO,UAAU,SAAU,QAAO;AAChD,QAAM,OAAO,OAAO,MAAM,QAAQ,EAAE;AACpC,QAAM,OACJ,MAAM,QAAQ,OAAO,MAAM,SAAS,WAC/B,MAAM,OACN,CAAC;AAER,MAAI,SAAS,cAAc,SAAS,gBAAgB,SAAS,SAAS;AACpE,WAAO,aAAa,MAAM,YAAY;AAAA,EACxC;AACA,MAAI,SAAS,uBAAuB,SAAS,YAAY,SAAS,SAAS;AACzE,UAAM,QAAQ,KAAK;AACnB,QAAI,CAAC,SAAS,OAAO,UAAU,SAAU,QAAO;AAChD,WAAO,aAAa,EAAE,GAAG,OAAO,OAAO,KAAK,SAAS,MAAM,MAAM,GAAG,YAAY;AAAA,EAClF;AACA,SAAO;AACT;AAEA,SAAS,aACP,MACA,cACyD;AACzD,QAAM,WAAW,iBAAiB,MAAM,CAAC,YAAY,eAAe,eAAe,CAAC;AACpF,QAAM,YAAY,iBAAiB,MAAM,CAAC,aAAa,gBAAgB,mBAAmB,CAAC;AAC3F,QAAM,UAAU,iBAAiB,MAAM,CAAC,WAAW,gBAAgB,YAAY,MAAM,CAAC;AACtF,MAAI,aAAa,UAAa,cAAc,UAAa,YAAY,QAAW;AAC9E,WAAO;AAAA,EACT;AACA,QAAM,QAAQ,OAAO,KAAK,UAAU,YAAY,KAAK,MAAM,SAAS,IAAI,KAAK,QAAQ;AACrF,QAAM,QAAmD;AAAA,IACvD,MAAM;AAAA,IACN;AAAA,EACF;AACA,MAAI,aAAa,OAAW,OAAM,WAAW;AAC7C,MAAI,cAAc,OAAW,OAAM,YAAY;AAC/C,MAAI,YAAY,OAAW,OAAM,UAAU;AAC3C,SAAO;AACT;AAEA,SAAS,iBAAiB,MAA+B,MAAoC;AAC3F,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,KAAK,GAAG;AACtB,QAAI,OAAO,UAAU,YAAY,OAAO,SAAS,KAAK,EAAG,QAAO;AAAA,EAClE;AACA,SAAO;AACT;AAOA,SAAS,SAAS,OAAwB;AACxC,MAAI;AACJ,MAAI;AACF,UAAM,KAAK,UAAU,KAAK,KAAK,OAAO,KAAK;AAAA,EAC7C,QAAQ;AACN,UAAM,OAAO,KAAK;AAAA,EACpB;AAEA,MAAI,IAAI;AACR,WAAS,IAAI,GAAG,IAAI,IAAI,QAAQ,KAAK,GAAG;AACtC,SAAK,IAAI,WAAW,CAAC;AACrB,QAAI,KAAK,KAAK,GAAG,QAAU;AAAA,EAC7B;AACA,UAAQ,MAAM,GAAG,SAAS,EAAE,EAAE,SAAS,GAAG,GAAG;AAC/C;","names":["decision"]}
package/dist/loops.d.ts CHANGED
@@ -1,6 +1,6 @@
1
1
  export { AgentProfile, CreateSandboxOptions, SandboxEvent, SandboxInstance } from '@tangle-network/sandbox';
2
- import { D as DefaultVerdict, a as Driver, I as Iteration, A as AgentRunSpec, O as OutputAdapter, V as Validator, E as ExecCtx, L as LoopWinner, b as LoopResult } from './types-Bx-tArkc.js';
3
- export { c as LoopDecisionPayload, d as LoopEndedPayload, e as LoopIterationEndedPayload, f as LoopIterationStartedPayload, g as LoopSandboxClient, h as LoopStartedPayload, i as LoopTraceEmitter, j as LoopTraceEvent, k as ValidationCtx } from './types-Bx-tArkc.js';
2
+ import { D as DefaultVerdict, a as Driver, I as Iteration, A as AgentRunSpec, O as OutputAdapter, V as Validator, E as ExecCtx, L as LoopWinner, b as LoopResult } from './types-Cu-SkGa0.js';
3
+ export { c as LoopDecisionPayload, d as LoopEndedPayload, e as LoopIterationDispatchPayload, f as LoopIterationEndedPayload, g as LoopIterationStartedPayload, h as LoopSandboxClient, i as LoopSandboxPlacement, j as LoopStartedPayload, k as LoopTraceEmitter, l as LoopTraceEvent, m as ValidationCtx } from './types-Cu-SkGa0.js';
4
4
  import './runtime-run-B2j-hvBj.js';
5
5
  import './types-DvJIha6w.js';
6
6
  import '@tangle-network/agent-eval';
@@ -96,7 +96,9 @@ declare function refineWinnerIndex<Task, Output>(iterations: ReadonlyArray<Itera
96
96
  * 2. For each task (parallel, bounded by `maxConcurrency`):
97
97
  * a. round-robin an `AgentRunSpec` from `agentRuns`
98
98
  * b. `sandboxClient.create({ backend: { profile }, ...overrides })`
99
- * c. iterate `box.streamPrompt(taskToPrompt(task))` and collect events
99
+ * c. emit `loop.iteration.dispatch` with the placement
100
+ * (`{ sibling, sandboxId }` or `{ fleet, fleetId, machineId, sandboxId }`)
101
+ * d. iterate `box.streamPrompt(taskToPrompt(task))` and collect events
100
102
  * 3. `output.parse(events)` → typed `Output`
101
103
  * 4. `validator?.validate(output)` → `DefaultVerdict`
102
104
  * 5. Append `Iteration` to history; emit `loop.iteration.ended`
package/dist/loops.js CHANGED
@@ -2,7 +2,7 @@ import {
2
2
  createRefineDriver,
3
3
  refineWinnerIndex,
4
4
  runLoop
5
- } from "./chunk-VFUEE6DF.js";
5
+ } from "./chunk-EDVCVFQB.js";
6
6
  import {
7
7
  createFanoutVoteDriver,
8
8
  scoreFanoutVoteIterations
package/dist/mcp/bin.js CHANGED
@@ -1,11 +1,12 @@
1
1
  #!/usr/bin/env node
2
2
  import {
3
3
  createDefaultCoderDelegate,
4
- createMcpServer
5
- } from "../chunk-LPPM7EGS.js";
4
+ createMcpServer,
5
+ detectExecutor
6
+ } from "../chunk-4XA4TWKW.js";
6
7
  import {
7
8
  runLoop
8
- } from "../chunk-VFUEE6DF.js";
9
+ } from "../chunk-EDVCVFQB.js";
9
10
  import "../chunk-Z5LKAYAS.js";
10
11
  import "../chunk-XLWPTPRP.js";
11
12
  import "../chunk-RZAOYKCO.js";
@@ -17,8 +18,10 @@ async function main() {
17
18
  const maxConcurrency = parseConcurrency(process.env.MCP_MAX_CONCURRENT_SANDBOXES);
18
19
  const wantCoder = !process.env.MCP_DISABLE_CODER;
19
20
  const wantResearcher = !process.env.MCP_DISABLE_RESEARCHER;
21
+ const fleetId = parseFleetId(process.env.TANGLE_FLEET_ID);
20
22
  const needsSandbox = wantCoder || wantResearcher;
21
23
  let sandboxClient;
24
+ let executor;
22
25
  if (needsSandbox) {
23
26
  const apiKey = process.env.TANGLE_API_KEY;
24
27
  if (!apiKey && !process.env.AGENT_RUNTIME_MCP_ALLOW_NO_KEY) {
@@ -27,14 +30,27 @@ async function main() {
27
30
  );
28
31
  process.exit(2);
29
32
  }
33
+ if (fleetId && !apiKey) {
34
+ process.stderr.write(
35
+ "agent-runtime-mcp: TANGLE_FLEET_ID was set but TANGLE_API_KEY is missing; cannot resolve fleet handle. Provide an api key or unset TANGLE_FLEET_ID.\n"
36
+ );
37
+ process.exit(2);
38
+ }
30
39
  sandboxClient = await loadSandboxClient(apiKey);
40
+ executor = await detectExecutor({ sandboxClient });
41
+ if (fleetId) {
42
+ process.stderr.write(`agent-runtime-mcp: fleet-aware delegation: fleetId=${fleetId}
43
+ `);
44
+ }
45
+ process.stderr.write(`agent-runtime-mcp: delegation placement \u2192 ${executor.describe()}
46
+ `);
31
47
  }
32
- const coderDelegate = wantCoder && sandboxClient ? createDefaultCoderDelegate({
33
- sandboxClient,
48
+ const coderDelegate = wantCoder && executor ? createDefaultCoderDelegate({
49
+ executor,
34
50
  fanoutHarnesses,
35
51
  maxConcurrency
36
52
  }) : void 0;
37
- const researcherDelegate = wantResearcher && sandboxClient ? await loadResearcherDelegate(sandboxClient, maxConcurrency) : void 0;
53
+ const researcherDelegate = wantResearcher && executor ? await loadResearcherDelegate(executor.client, maxConcurrency) : void 0;
38
54
  const server = createMcpServer({ coderDelegate, researcherDelegate });
39
55
  process.on("SIGINT", () => {
40
56
  server.stop();
@@ -145,6 +161,11 @@ function parseHarnesses(raw) {
145
161
  const list = raw.split(",").map((entry) => entry.trim()).filter(Boolean);
146
162
  return list.length > 0 ? list : void 0;
147
163
  }
164
+ function parseFleetId(raw) {
165
+ if (typeof raw !== "string") return void 0;
166
+ const trimmed = raw.trim();
167
+ return trimmed.length > 0 ? trimmed : void 0;
168
+ }
148
169
  function parseConcurrency(raw) {
149
170
  if (!raw) return 4;
150
171
  const n = Number(raw);
@@ -1 +1 @@
1
- {"version":3,"sources":["../../src/mcp/bin.ts"],"sourcesContent":["#!/usr/bin/env node\n\n/**\n * @experimental\n *\n * `agent-runtime-mcp` — stdio MCP server entry point.\n *\n * Spins up a server with the default coder delegate (wired against the\n * real `@tangle-network/sandbox` client) and, when the optional\n * `@tangle-network/agent-knowledge` peer is installed, a researcher\n * delegate against `multiHarnessResearcherFanout`.\n *\n * Environment variables:\n * TANGLE_API_KEY required — passed to `new Sandbox({ apiKey })`\n * SANDBOX_BASE_URL optional — sandbox-SDK base URL override\n * MCP_MAX_CONCURRENT_SANDBOXES default 4 — kernel maxConcurrency cap\n * MCP_CODER_FANOUT_HARNESSES comma-separated harness ids to use for variants > 1\n * MCP_DISABLE_CODER set to `1` to omit `delegate_code`\n * MCP_DISABLE_RESEARCHER set to `1` to omit `delegate_research` even when peer is present\n */\n\nimport type { LoopSandboxClient } from '../loops'\nimport { runLoop } from '../loops'\nimport { createDefaultCoderDelegate, type ResearcherDelegate } from './delegates'\nimport { createMcpServer } from './server'\nimport type { ResearchOutputShape } from './types'\n\nasync function main(): Promise<void> {\n const fanoutHarnesses = parseHarnesses(process.env.MCP_CODER_FANOUT_HARNESSES)\n const maxConcurrency = parseConcurrency(process.env.MCP_MAX_CONCURRENT_SANDBOXES)\n const wantCoder = !process.env.MCP_DISABLE_CODER\n const wantResearcher = !process.env.MCP_DISABLE_RESEARCHER\n\n // Skip the sandbox client load entirely when no profile delegate needs it —\n // the feedback + status + history tools are queue-bound and require no\n // sandbox. Useful for tooling that mounts the MCP server purely for\n // self-introspection.\n const needsSandbox = wantCoder || wantResearcher\n let sandboxClient: LoopSandboxClient | undefined\n if (needsSandbox) {\n const apiKey = process.env.TANGLE_API_KEY\n if (!apiKey && !process.env.AGENT_RUNTIME_MCP_ALLOW_NO_KEY) {\n process.stderr.write(\n 'agent-runtime-mcp: TANGLE_API_KEY is required. Set AGENT_RUNTIME_MCP_ALLOW_NO_KEY=1 to run without it for diagnostics, or MCP_DISABLE_CODER=1 MCP_DISABLE_RESEARCHER=1 to run the queue-only subset.\\n',\n )\n process.exit(2)\n }\n sandboxClient = await loadSandboxClient(apiKey)\n }\n\n const coderDelegate =\n wantCoder && sandboxClient\n ? createDefaultCoderDelegate({\n sandboxClient,\n fanoutHarnesses,\n maxConcurrency,\n })\n : undefined\n\n const researcherDelegate =\n wantResearcher && sandboxClient\n ? await loadResearcherDelegate(sandboxClient, maxConcurrency)\n : undefined\n\n const server = createMcpServer({ coderDelegate, researcherDelegate })\n\n process.on('SIGINT', () => {\n server.stop()\n process.exit(0)\n })\n process.on('SIGTERM', () => {\n server.stop()\n process.exit(0)\n })\n\n await server.serve()\n}\n\nasync function loadSandboxClient(apiKey: string | undefined): Promise<LoopSandboxClient> {\n // Diagnostic mode: AGENT_RUNTIME_MCP_ALLOW_NO_KEY=1 enables tools/list + the\n // queue-bound tools (status / history / feedback) without sandbox creds.\n // Coder + researcher delegations require a real client; the stub fails loud\n // at create() so the agent observes the cause instead of silent success.\n if (!apiKey) {\n return {\n async create() {\n throw new Error(\n 'agent-runtime-mcp: TANGLE_API_KEY is unset; coder/researcher delegations are disabled in diagnostic mode. Set TANGLE_API_KEY or use MCP_DISABLE_CODER=1 MCP_DISABLE_RESEARCHER=1 to remove the unsupported tools from the tool list.',\n )\n },\n } satisfies LoopSandboxClient\n }\n // Dynamic import keeps the bin importable in environments that haven't\n // installed `@tangle-network/sandbox` yet (the runtime package lists it\n // as a peer dep, not a hard dep).\n const mod = await import('@tangle-network/sandbox').catch((err) => {\n process.stderr.write(\n `agent-runtime-mcp: failed to load @tangle-network/sandbox (${err.message}); install the peer dependency\\n`,\n )\n process.exit(2)\n })\n const SandboxCtor = (mod as { Sandbox?: new (config: unknown) => LoopSandboxClient }).Sandbox\n if (!SandboxCtor) {\n process.stderr.write(\n 'agent-runtime-mcp: @tangle-network/sandbox does not export Sandbox; cannot construct client\\n',\n )\n process.exit(2)\n }\n const baseUrl = process.env.SANDBOX_BASE_URL\n return new SandboxCtor({\n apiKey,\n ...(baseUrl ? { baseUrl } : {}),\n })\n}\n\ninterface ResearcherProfilePreset {\n agentRunSpec: Parameters<typeof runLoop>[0]['agentRun'] extends infer T ? NonNullable<T> : never\n output: Parameters<typeof runLoop>[0]['output']\n validator: Parameters<typeof runLoop>[0]['validator']\n}\n\ninterface ResearcherFanoutPreset {\n agentRuns: NonNullable<Parameters<typeof runLoop>[0]['agentRuns']>\n output: Parameters<typeof runLoop>[0]['output']\n validator: Parameters<typeof runLoop>[0]['validator']\n driver: Parameters<typeof runLoop>[0]['driver']\n}\n\nasync function loadResearcherDelegate(\n sandboxClient: LoopSandboxClient,\n maxConcurrency: number,\n): Promise<ResearcherDelegate | undefined> {\n // Optional peer — when `@tangle-network/agent-knowledge` isn't installed,\n // we silently omit the researcher tool from the advertisement. The\n // dynamic-import path is resolved at runtime; TypeScript cannot see the\n // peer, so we type the module structurally rather than via its own\n // declaration file.\n const profilesSpecifier = '@tangle-network/agent-knowledge/profiles'\n const mod = await import(profilesSpecifier).catch(() => undefined)\n if (!mod) return undefined\n type SingleFactory = (opts: { task: unknown }) => ResearcherProfilePreset\n type FanoutFactory = (opts: { task: unknown }) => ResearcherFanoutPreset\n const fanoutFactory = (mod as { multiHarnessResearcherFanout?: FanoutFactory })\n .multiHarnessResearcherFanout\n const singleFactory = (mod as { researcherProfile?: SingleFactory }).researcherProfile\n if (!fanoutFactory || !singleFactory) return undefined\n\n return async (args, ctx) => {\n const task = {\n question: args.question,\n knowledgeNamespace: args.namespace,\n scope: args.scope,\n sources: args.sources,\n recencyWindow: args.config?.recencyWindow\n ? {\n since: args.config.recencyWindow.since\n ? new Date(args.config.recencyWindow.since)\n : undefined,\n until: args.config.recencyWindow.until\n ? new Date(args.config.recencyWindow.until)\n : undefined,\n }\n : undefined,\n maxItems: args.config?.maxItems,\n minConfidence: args.config?.minConfidence,\n }\n const variants = Math.max(1, Math.trunc(args.variants ?? 1))\n ctx.report({ iteration: 0, phase: 'starting' })\n if (variants <= 1) {\n const preset = singleFactory({ task })\n const result = await runLoop({\n driver: {\n name: 'mcp-researcher-single',\n async plan(t, history) {\n return history.length === 0 ? [t] : []\n },\n decide(history) {\n return history.length > 0 ? 'pick-winner' : 'fail'\n },\n },\n agentRun: preset.agentRunSpec,\n output: preset.output,\n validator: preset.validator,\n task,\n ctx: { sandboxClient, signal: ctx.signal },\n maxIterations: 1,\n maxConcurrency,\n })\n const output = result.winner?.output\n if (!output) throw new Error('researcher delegate produced no winner')\n ctx.report({ iteration: 1, phase: 'completed' })\n return output as ResearchOutputShape\n }\n const fanout = fanoutFactory({ task })\n const result = await runLoop({\n driver: fanout.driver,\n agentRuns: fanout.agentRuns.slice(0, variants),\n output: fanout.output,\n validator: fanout.validator,\n task,\n ctx: { sandboxClient, signal: ctx.signal },\n maxIterations: variants,\n maxConcurrency: Math.min(maxConcurrency, variants),\n })\n const output = result.winner?.output\n if (!output) throw new Error('researcher delegate fanout produced no winner')\n ctx.report({ iteration: result.iterations.length, phase: 'completed' })\n return output as ResearchOutputShape\n }\n}\n\nfunction parseHarnesses(raw: string | undefined): string[] | undefined {\n if (!raw) return undefined\n const list = raw\n .split(',')\n .map((entry) => entry.trim())\n .filter(Boolean)\n return list.length > 0 ? list : undefined\n}\n\nfunction parseConcurrency(raw: string | undefined): number {\n if (!raw) return 4\n const n = Number(raw)\n if (!Number.isFinite(n) || n < 1) return 4\n return Math.min(Math.trunc(n), 32)\n}\n\nmain().catch((err) => {\n process.stderr.write(`agent-runtime-mcp: ${err instanceof Error ? err.stack : String(err)}\\n`)\n process.exit(1)\n})\n"],"mappings":";;;;;;;;;;;;;;AA2BA,eAAe,OAAsB;AACnC,QAAM,kBAAkB,eAAe,QAAQ,IAAI,0BAA0B;AAC7E,QAAM,iBAAiB,iBAAiB,QAAQ,IAAI,4BAA4B;AAChF,QAAM,YAAY,CAAC,QAAQ,IAAI;AAC/B,QAAM,iBAAiB,CAAC,QAAQ,IAAI;AAMpC,QAAM,eAAe,aAAa;AAClC,MAAI;AACJ,MAAI,cAAc;AAChB,UAAM,SAAS,QAAQ,IAAI;AAC3B,QAAI,CAAC,UAAU,CAAC,QAAQ,IAAI,gCAAgC;AAC1D,cAAQ,OAAO;AAAA,QACb;AAAA,MACF;AACA,cAAQ,KAAK,CAAC;AAAA,IAChB;AACA,oBAAgB,MAAM,kBAAkB,MAAM;AAAA,EAChD;AAEA,QAAM,gBACJ,aAAa,gBACT,2BAA2B;AAAA,IACzB;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC,IACD;AAEN,QAAM,qBACJ,kBAAkB,gBACd,MAAM,uBAAuB,eAAe,cAAc,IAC1D;AAEN,QAAM,SAAS,gBAAgB,EAAE,eAAe,mBAAmB,CAAC;AAEpE,UAAQ,GAAG,UAAU,MAAM;AACzB,WAAO,KAAK;AACZ,YAAQ,KAAK,CAAC;AAAA,EAChB,CAAC;AACD,UAAQ,GAAG,WAAW,MAAM;AAC1B,WAAO,KAAK;AACZ,YAAQ,KAAK,CAAC;AAAA,EAChB,CAAC;AAED,QAAM,OAAO,MAAM;AACrB;AAEA,eAAe,kBAAkB,QAAwD;AAKvF,MAAI,CAAC,QAAQ;AACX,WAAO;AAAA,MACL,MAAM,SAAS;AACb,cAAM,IAAI;AAAA,UACR;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAIA,QAAM,MAAM,MAAM,OAAO,yBAAyB,EAAE,MAAM,CAAC,QAAQ;AACjE,YAAQ,OAAO;AAAA,MACb,8DAA8D,IAAI,OAAO;AAAA;AAAA,IAC3E;AACA,YAAQ,KAAK,CAAC;AAAA,EAChB,CAAC;AACD,QAAM,cAAe,IAAiE;AACtF,MAAI,CAAC,aAAa;AAChB,YAAQ,OAAO;AAAA,MACb;AAAA,IACF;AACA,YAAQ,KAAK,CAAC;AAAA,EAChB;AACA,QAAM,UAAU,QAAQ,IAAI;AAC5B,SAAO,IAAI,YAAY;AAAA,IACrB;AAAA,IACA,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;AAAA,EAC/B,CAAC;AACH;AAeA,eAAe,uBACb,eACA,gBACyC;AAMzC,QAAM,oBAAoB;AAC1B,QAAM,MAAM,MAAM,OAAO,mBAAmB,MAAM,MAAM,MAAS;AACjE,MAAI,CAAC,IAAK,QAAO;AAGjB,QAAM,gBAAiB,IACpB;AACH,QAAM,gBAAiB,IAA8C;AACrE,MAAI,CAAC,iBAAiB,CAAC,cAAe,QAAO;AAE7C,SAAO,OAAO,MAAM,QAAQ;AAC1B,UAAM,OAAO;AAAA,MACX,UAAU,KAAK;AAAA,MACf,oBAAoB,KAAK;AAAA,MACzB,OAAO,KAAK;AAAA,MACZ,SAAS,KAAK;AAAA,MACd,eAAe,KAAK,QAAQ,gBACxB;AAAA,QACE,OAAO,KAAK,OAAO,cAAc,QAC7B,IAAI,KAAK,KAAK,OAAO,cAAc,KAAK,IACxC;AAAA,QACJ,OAAO,KAAK,OAAO,cAAc,QAC7B,IAAI,KAAK,KAAK,OAAO,cAAc,KAAK,IACxC;AAAA,MACN,IACA;AAAA,MACJ,UAAU,KAAK,QAAQ;AAAA,MACvB,eAAe,KAAK,QAAQ;AAAA,IAC9B;AACA,UAAM,WAAW,KAAK,IAAI,GAAG,KAAK,MAAM,KAAK,YAAY,CAAC,CAAC;AAC3D,QAAI,OAAO,EAAE,WAAW,GAAG,OAAO,WAAW,CAAC;AAC9C,QAAI,YAAY,GAAG;AACjB,YAAM,SAAS,cAAc,EAAE,KAAK,CAAC;AACrC,YAAMA,UAAS,MAAM,QAAQ;AAAA,QAC3B,QAAQ;AAAA,UACN,MAAM;AAAA,UACN,MAAM,KAAK,GAAG,SAAS;AACrB,mBAAO,QAAQ,WAAW,IAAI,CAAC,CAAC,IAAI,CAAC;AAAA,UACvC;AAAA,UACA,OAAO,SAAS;AACd,mBAAO,QAAQ,SAAS,IAAI,gBAAgB;AAAA,UAC9C;AAAA,QACF;AAAA,QACA,UAAU,OAAO;AAAA,QACjB,QAAQ,OAAO;AAAA,QACf,WAAW,OAAO;AAAA,QAClB;AAAA,QACA,KAAK,EAAE,eAAe,QAAQ,IAAI,OAAO;AAAA,QACzC,eAAe;AAAA,QACf;AAAA,MACF,CAAC;AACD,YAAMC,UAASD,QAAO,QAAQ;AAC9B,UAAI,CAACC,QAAQ,OAAM,IAAI,MAAM,wCAAwC;AACrE,UAAI,OAAO,EAAE,WAAW,GAAG,OAAO,YAAY,CAAC;AAC/C,aAAOA;AAAA,IACT;AACA,UAAM,SAAS,cAAc,EAAE,KAAK,CAAC;AACrC,UAAM,SAAS,MAAM,QAAQ;AAAA,MAC3B,QAAQ,OAAO;AAAA,MACf,WAAW,OAAO,UAAU,MAAM,GAAG,QAAQ;AAAA,MAC7C,QAAQ,OAAO;AAAA,MACf,WAAW,OAAO;AAAA,MAClB;AAAA,MACA,KAAK,EAAE,eAAe,QAAQ,IAAI,OAAO;AAAA,MACzC,eAAe;AAAA,MACf,gBAAgB,KAAK,IAAI,gBAAgB,QAAQ;AAAA,IACnD,CAAC;AACD,UAAM,SAAS,OAAO,QAAQ;AAC9B,QAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,+CAA+C;AAC5E,QAAI,OAAO,EAAE,WAAW,OAAO,WAAW,QAAQ,OAAO,YAAY,CAAC;AACtE,WAAO;AAAA,EACT;AACF;AAEA,SAAS,eAAe,KAA+C;AACrE,MAAI,CAAC,IAAK,QAAO;AACjB,QAAM,OAAO,IACV,MAAM,GAAG,EACT,IAAI,CAAC,UAAU,MAAM,KAAK,CAAC,EAC3B,OAAO,OAAO;AACjB,SAAO,KAAK,SAAS,IAAI,OAAO;AAClC;AAEA,SAAS,iBAAiB,KAAiC;AACzD,MAAI,CAAC,IAAK,QAAO;AACjB,QAAM,IAAI,OAAO,GAAG;AACpB,MAAI,CAAC,OAAO,SAAS,CAAC,KAAK,IAAI,EAAG,QAAO;AACzC,SAAO,KAAK,IAAI,KAAK,MAAM,CAAC,GAAG,EAAE;AACnC;AAEA,KAAK,EAAE,MAAM,CAAC,QAAQ;AACpB,UAAQ,OAAO,MAAM,sBAAsB,eAAe,QAAQ,IAAI,QAAQ,OAAO,GAAG,CAAC;AAAA,CAAI;AAC7F,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":["result","output"]}
1
+ {"version":3,"sources":["../../src/mcp/bin.ts"],"sourcesContent":["#!/usr/bin/env node\n\n/**\n * @experimental\n *\n * `agent-runtime-mcp` — stdio MCP server entry point.\n *\n * Spins up a server with the default coder delegate (wired against the\n * real `@tangle-network/sandbox` client) and, when the optional\n * `@tangle-network/agent-knowledge` peer is installed, a researcher\n * delegate against `multiHarnessResearcherFanout`.\n *\n * Environment variables:\n * TANGLE_API_KEY required — passed to `new Sandbox({ apiKey })`\n * SANDBOX_BASE_URL optional — sandbox-SDK base URL override\n * TANGLE_FLEET_ID optional — when set, delegations dispatch\n * INTO this fleet's shared workspace instead\n * of creating sibling sandboxes. Set by the\n * parent sandbox when launching this MCP\n * server so worker diffs land on the caller's\n * filesystem with no cross-sandbox boundary.\n * TANGLE_FLEET_EXCLUDE_MACHINES optional — comma-separated machine ids to\n * skip during fleet-mode round-robin\n * (typically the coordinator machine this\n * MCP server is running on).\n * MCP_MAX_CONCURRENT_SANDBOXES default 4 — kernel maxConcurrency cap\n * MCP_CODER_FANOUT_HARNESSES comma-separated harness ids to use for variants > 1\n * MCP_DISABLE_CODER set to `1` to omit `delegate_code`\n * MCP_DISABLE_RESEARCHER set to `1` to omit `delegate_research` even when peer is present\n */\n\nimport type { LoopSandboxClient } from '../loops'\nimport { runLoop } from '../loops'\nimport { detectExecutor } from './bin-helpers'\nimport { createDefaultCoderDelegate, type ResearcherDelegate } from './delegates'\nimport type { DelegationExecutor } from './executor'\nimport { createMcpServer } from './server'\nimport type { ResearchOutputShape } from './types'\n\nasync function main(): Promise<void> {\n const fanoutHarnesses = parseHarnesses(process.env.MCP_CODER_FANOUT_HARNESSES)\n const maxConcurrency = parseConcurrency(process.env.MCP_MAX_CONCURRENT_SANDBOXES)\n const wantCoder = !process.env.MCP_DISABLE_CODER\n const wantResearcher = !process.env.MCP_DISABLE_RESEARCHER\n const fleetId = parseFleetId(process.env.TANGLE_FLEET_ID)\n\n // Skip the sandbox client load entirely when no profile delegate needs it —\n // the feedback + status + history tools are queue-bound and require no\n // sandbox. Useful for tooling that mounts the MCP server purely for\n // self-introspection.\n const needsSandbox = wantCoder || wantResearcher\n let sandboxClient: LoopSandboxClient | undefined\n let executor: DelegationExecutor | undefined\n if (needsSandbox) {\n const apiKey = process.env.TANGLE_API_KEY\n if (!apiKey && !process.env.AGENT_RUNTIME_MCP_ALLOW_NO_KEY) {\n process.stderr.write(\n 'agent-runtime-mcp: TANGLE_API_KEY is required. Set AGENT_RUNTIME_MCP_ALLOW_NO_KEY=1 to run without it for diagnostics, or MCP_DISABLE_CODER=1 MCP_DISABLE_RESEARCHER=1 to run the queue-only subset.\\n',\n )\n process.exit(2)\n }\n // Fleet mode against a diagnostic stub is meaningless — the stub can't\n // resolve a real fleet handle. Refuse rather than silently degrading,\n // otherwise a fleet-mounted MCP would behave differently than configured.\n if (fleetId && !apiKey) {\n process.stderr.write(\n 'agent-runtime-mcp: TANGLE_FLEET_ID was set but TANGLE_API_KEY is missing; cannot resolve fleet handle. Provide an api key or unset TANGLE_FLEET_ID.\\n',\n )\n process.exit(2)\n }\n sandboxClient = await loadSandboxClient(apiKey)\n executor = await detectExecutor({ sandboxClient })\n if (fleetId) {\n process.stderr.write(`agent-runtime-mcp: fleet-aware delegation: fleetId=${fleetId}\\n`)\n }\n process.stderr.write(`agent-runtime-mcp: delegation placement → ${executor.describe()}\\n`)\n }\n\n const coderDelegate =\n wantCoder && executor\n ? createDefaultCoderDelegate({\n executor,\n fanoutHarnesses,\n maxConcurrency,\n })\n : undefined\n\n const researcherDelegate =\n wantResearcher && executor\n ? await loadResearcherDelegate(executor.client, maxConcurrency)\n : undefined\n\n const server = createMcpServer({ coderDelegate, researcherDelegate })\n\n process.on('SIGINT', () => {\n server.stop()\n process.exit(0)\n })\n process.on('SIGTERM', () => {\n server.stop()\n process.exit(0)\n })\n\n await server.serve()\n}\n\nasync function loadSandboxClient(apiKey: string | undefined): Promise<LoopSandboxClient> {\n // Diagnostic mode: AGENT_RUNTIME_MCP_ALLOW_NO_KEY=1 enables tools/list + the\n // queue-bound tools (status / history / feedback) without sandbox creds.\n // Coder + researcher delegations require a real client; the stub fails loud\n // at create() so the agent observes the cause instead of silent success.\n if (!apiKey) {\n return {\n async create() {\n throw new Error(\n 'agent-runtime-mcp: TANGLE_API_KEY is unset; coder/researcher delegations are disabled in diagnostic mode. Set TANGLE_API_KEY or use MCP_DISABLE_CODER=1 MCP_DISABLE_RESEARCHER=1 to remove the unsupported tools from the tool list.',\n )\n },\n } satisfies LoopSandboxClient\n }\n // Dynamic import keeps the bin importable in environments that haven't\n // installed `@tangle-network/sandbox` yet (the runtime package lists it\n // as a peer dep, not a hard dep).\n const mod = await import('@tangle-network/sandbox').catch((err) => {\n process.stderr.write(\n `agent-runtime-mcp: failed to load @tangle-network/sandbox (${err.message}); install the peer dependency\\n`,\n )\n process.exit(2)\n })\n const SandboxCtor = (mod as { Sandbox?: new (config: unknown) => LoopSandboxClient }).Sandbox\n if (!SandboxCtor) {\n process.stderr.write(\n 'agent-runtime-mcp: @tangle-network/sandbox does not export Sandbox; cannot construct client\\n',\n )\n process.exit(2)\n }\n const baseUrl = process.env.SANDBOX_BASE_URL\n return new SandboxCtor({\n apiKey,\n ...(baseUrl ? { baseUrl } : {}),\n })\n}\n\ninterface ResearcherProfilePreset {\n agentRunSpec: Parameters<typeof runLoop>[0]['agentRun'] extends infer T ? NonNullable<T> : never\n output: Parameters<typeof runLoop>[0]['output']\n validator: Parameters<typeof runLoop>[0]['validator']\n}\n\ninterface ResearcherFanoutPreset {\n agentRuns: NonNullable<Parameters<typeof runLoop>[0]['agentRuns']>\n output: Parameters<typeof runLoop>[0]['output']\n validator: Parameters<typeof runLoop>[0]['validator']\n driver: Parameters<typeof runLoop>[0]['driver']\n}\n\nasync function loadResearcherDelegate(\n sandboxClient: LoopSandboxClient,\n maxConcurrency: number,\n): Promise<ResearcherDelegate | undefined> {\n // Optional peer — when `@tangle-network/agent-knowledge` isn't installed,\n // we silently omit the researcher tool from the advertisement. The\n // dynamic-import path is resolved at runtime; TypeScript cannot see the\n // peer, so we type the module structurally rather than via its own\n // declaration file.\n const profilesSpecifier = '@tangle-network/agent-knowledge/profiles'\n const mod = await import(profilesSpecifier).catch(() => undefined)\n if (!mod) return undefined\n type SingleFactory = (opts: { task: unknown }) => ResearcherProfilePreset\n type FanoutFactory = (opts: { task: unknown }) => ResearcherFanoutPreset\n const fanoutFactory = (mod as { multiHarnessResearcherFanout?: FanoutFactory })\n .multiHarnessResearcherFanout\n const singleFactory = (mod as { researcherProfile?: SingleFactory }).researcherProfile\n if (!fanoutFactory || !singleFactory) return undefined\n\n return async (args, ctx) => {\n const task = {\n question: args.question,\n knowledgeNamespace: args.namespace,\n scope: args.scope,\n sources: args.sources,\n recencyWindow: args.config?.recencyWindow\n ? {\n since: args.config.recencyWindow.since\n ? new Date(args.config.recencyWindow.since)\n : undefined,\n until: args.config.recencyWindow.until\n ? new Date(args.config.recencyWindow.until)\n : undefined,\n }\n : undefined,\n maxItems: args.config?.maxItems,\n minConfidence: args.config?.minConfidence,\n }\n const variants = Math.max(1, Math.trunc(args.variants ?? 1))\n ctx.report({ iteration: 0, phase: 'starting' })\n if (variants <= 1) {\n const preset = singleFactory({ task })\n const result = await runLoop({\n driver: {\n name: 'mcp-researcher-single',\n async plan(t, history) {\n return history.length === 0 ? [t] : []\n },\n decide(history) {\n return history.length > 0 ? 'pick-winner' : 'fail'\n },\n },\n agentRun: preset.agentRunSpec,\n output: preset.output,\n validator: preset.validator,\n task,\n ctx: { sandboxClient, signal: ctx.signal },\n maxIterations: 1,\n maxConcurrency,\n })\n const output = result.winner?.output\n if (!output) throw new Error('researcher delegate produced no winner')\n ctx.report({ iteration: 1, phase: 'completed' })\n return output as ResearchOutputShape\n }\n const fanout = fanoutFactory({ task })\n const result = await runLoop({\n driver: fanout.driver,\n agentRuns: fanout.agentRuns.slice(0, variants),\n output: fanout.output,\n validator: fanout.validator,\n task,\n ctx: { sandboxClient, signal: ctx.signal },\n maxIterations: variants,\n maxConcurrency: Math.min(maxConcurrency, variants),\n })\n const output = result.winner?.output\n if (!output) throw new Error('researcher delegate fanout produced no winner')\n ctx.report({ iteration: result.iterations.length, phase: 'completed' })\n return output as ResearchOutputShape\n }\n}\n\nfunction parseHarnesses(raw: string | undefined): string[] | undefined {\n if (!raw) return undefined\n const list = raw\n .split(',')\n .map((entry) => entry.trim())\n .filter(Boolean)\n return list.length > 0 ? list : undefined\n}\n\nfunction parseFleetId(raw: string | undefined): string | undefined {\n if (typeof raw !== 'string') return undefined\n const trimmed = raw.trim()\n return trimmed.length > 0 ? trimmed : undefined\n}\n\nfunction parseConcurrency(raw: string | undefined): number {\n if (!raw) return 4\n const n = Number(raw)\n if (!Number.isFinite(n) || n < 1) return 4\n return Math.min(Math.trunc(n), 32)\n}\n\nmain().catch((err) => {\n process.stderr.write(`agent-runtime-mcp: ${err instanceof Error ? err.stack : String(err)}\\n`)\n process.exit(1)\n})\n"],"mappings":";;;;;;;;;;;;;;;AAuCA,eAAe,OAAsB;AACnC,QAAM,kBAAkB,eAAe,QAAQ,IAAI,0BAA0B;AAC7E,QAAM,iBAAiB,iBAAiB,QAAQ,IAAI,4BAA4B;AAChF,QAAM,YAAY,CAAC,QAAQ,IAAI;AAC/B,QAAM,iBAAiB,CAAC,QAAQ,IAAI;AACpC,QAAM,UAAU,aAAa,QAAQ,IAAI,eAAe;AAMxD,QAAM,eAAe,aAAa;AAClC,MAAI;AACJ,MAAI;AACJ,MAAI,cAAc;AAChB,UAAM,SAAS,QAAQ,IAAI;AAC3B,QAAI,CAAC,UAAU,CAAC,QAAQ,IAAI,gCAAgC;AAC1D,cAAQ,OAAO;AAAA,QACb;AAAA,MACF;AACA,cAAQ,KAAK,CAAC;AAAA,IAChB;AAIA,QAAI,WAAW,CAAC,QAAQ;AACtB,cAAQ,OAAO;AAAA,QACb;AAAA,MACF;AACA,cAAQ,KAAK,CAAC;AAAA,IAChB;AACA,oBAAgB,MAAM,kBAAkB,MAAM;AAC9C,eAAW,MAAM,eAAe,EAAE,cAAc,CAAC;AACjD,QAAI,SAAS;AACX,cAAQ,OAAO,MAAM,sDAAsD,OAAO;AAAA,CAAI;AAAA,IACxF;AACA,YAAQ,OAAO,MAAM,kDAA6C,SAAS,SAAS,CAAC;AAAA,CAAI;AAAA,EAC3F;AAEA,QAAM,gBACJ,aAAa,WACT,2BAA2B;AAAA,IACzB;AAAA,IACA;AAAA,IACA;AAAA,EACF,CAAC,IACD;AAEN,QAAM,qBACJ,kBAAkB,WACd,MAAM,uBAAuB,SAAS,QAAQ,cAAc,IAC5D;AAEN,QAAM,SAAS,gBAAgB,EAAE,eAAe,mBAAmB,CAAC;AAEpE,UAAQ,GAAG,UAAU,MAAM;AACzB,WAAO,KAAK;AACZ,YAAQ,KAAK,CAAC;AAAA,EAChB,CAAC;AACD,UAAQ,GAAG,WAAW,MAAM;AAC1B,WAAO,KAAK;AACZ,YAAQ,KAAK,CAAC;AAAA,EAChB,CAAC;AAED,QAAM,OAAO,MAAM;AACrB;AAEA,eAAe,kBAAkB,QAAwD;AAKvF,MAAI,CAAC,QAAQ;AACX,WAAO;AAAA,MACL,MAAM,SAAS;AACb,cAAM,IAAI;AAAA,UACR;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF;AAIA,QAAM,MAAM,MAAM,OAAO,yBAAyB,EAAE,MAAM,CAAC,QAAQ;AACjE,YAAQ,OAAO;AAAA,MACb,8DAA8D,IAAI,OAAO;AAAA;AAAA,IAC3E;AACA,YAAQ,KAAK,CAAC;AAAA,EAChB,CAAC;AACD,QAAM,cAAe,IAAiE;AACtF,MAAI,CAAC,aAAa;AAChB,YAAQ,OAAO;AAAA,MACb;AAAA,IACF;AACA,YAAQ,KAAK,CAAC;AAAA,EAChB;AACA,QAAM,UAAU,QAAQ,IAAI;AAC5B,SAAO,IAAI,YAAY;AAAA,IACrB;AAAA,IACA,GAAI,UAAU,EAAE,QAAQ,IAAI,CAAC;AAAA,EAC/B,CAAC;AACH;AAeA,eAAe,uBACb,eACA,gBACyC;AAMzC,QAAM,oBAAoB;AAC1B,QAAM,MAAM,MAAM,OAAO,mBAAmB,MAAM,MAAM,MAAS;AACjE,MAAI,CAAC,IAAK,QAAO;AAGjB,QAAM,gBAAiB,IACpB;AACH,QAAM,gBAAiB,IAA8C;AACrE,MAAI,CAAC,iBAAiB,CAAC,cAAe,QAAO;AAE7C,SAAO,OAAO,MAAM,QAAQ;AAC1B,UAAM,OAAO;AAAA,MACX,UAAU,KAAK;AAAA,MACf,oBAAoB,KAAK;AAAA,MACzB,OAAO,KAAK;AAAA,MACZ,SAAS,KAAK;AAAA,MACd,eAAe,KAAK,QAAQ,gBACxB;AAAA,QACE,OAAO,KAAK,OAAO,cAAc,QAC7B,IAAI,KAAK,KAAK,OAAO,cAAc,KAAK,IACxC;AAAA,QACJ,OAAO,KAAK,OAAO,cAAc,QAC7B,IAAI,KAAK,KAAK,OAAO,cAAc,KAAK,IACxC;AAAA,MACN,IACA;AAAA,MACJ,UAAU,KAAK,QAAQ;AAAA,MACvB,eAAe,KAAK,QAAQ;AAAA,IAC9B;AACA,UAAM,WAAW,KAAK,IAAI,GAAG,KAAK,MAAM,KAAK,YAAY,CAAC,CAAC;AAC3D,QAAI,OAAO,EAAE,WAAW,GAAG,OAAO,WAAW,CAAC;AAC9C,QAAI,YAAY,GAAG;AACjB,YAAM,SAAS,cAAc,EAAE,KAAK,CAAC;AACrC,YAAMA,UAAS,MAAM,QAAQ;AAAA,QAC3B,QAAQ;AAAA,UACN,MAAM;AAAA,UACN,MAAM,KAAK,GAAG,SAAS;AACrB,mBAAO,QAAQ,WAAW,IAAI,CAAC,CAAC,IAAI,CAAC;AAAA,UACvC;AAAA,UACA,OAAO,SAAS;AACd,mBAAO,QAAQ,SAAS,IAAI,gBAAgB;AAAA,UAC9C;AAAA,QACF;AAAA,QACA,UAAU,OAAO;AAAA,QACjB,QAAQ,OAAO;AAAA,QACf,WAAW,OAAO;AAAA,QAClB;AAAA,QACA,KAAK,EAAE,eAAe,QAAQ,IAAI,OAAO;AAAA,QACzC,eAAe;AAAA,QACf;AAAA,MACF,CAAC;AACD,YAAMC,UAASD,QAAO,QAAQ;AAC9B,UAAI,CAACC,QAAQ,OAAM,IAAI,MAAM,wCAAwC;AACrE,UAAI,OAAO,EAAE,WAAW,GAAG,OAAO,YAAY,CAAC;AAC/C,aAAOA;AAAA,IACT;AACA,UAAM,SAAS,cAAc,EAAE,KAAK,CAAC;AACrC,UAAM,SAAS,MAAM,QAAQ;AAAA,MAC3B,QAAQ,OAAO;AAAA,MACf,WAAW,OAAO,UAAU,MAAM,GAAG,QAAQ;AAAA,MAC7C,QAAQ,OAAO;AAAA,MACf,WAAW,OAAO;AAAA,MAClB;AAAA,MACA,KAAK,EAAE,eAAe,QAAQ,IAAI,OAAO;AAAA,MACzC,eAAe;AAAA,MACf,gBAAgB,KAAK,IAAI,gBAAgB,QAAQ;AAAA,IACnD,CAAC;AACD,UAAM,SAAS,OAAO,QAAQ;AAC9B,QAAI,CAAC,OAAQ,OAAM,IAAI,MAAM,+CAA+C;AAC5E,QAAI,OAAO,EAAE,WAAW,OAAO,WAAW,QAAQ,OAAO,YAAY,CAAC;AACtE,WAAO;AAAA,EACT;AACF;AAEA,SAAS,eAAe,KAA+C;AACrE,MAAI,CAAC,IAAK,QAAO;AACjB,QAAM,OAAO,IACV,MAAM,GAAG,EACT,IAAI,CAAC,UAAU,MAAM,KAAK,CAAC,EAC3B,OAAO,OAAO;AACjB,SAAO,KAAK,SAAS,IAAI,OAAO;AAClC;AAEA,SAAS,aAAa,KAA6C;AACjE,MAAI,OAAO,QAAQ,SAAU,QAAO;AACpC,QAAM,UAAU,IAAI,KAAK;AACzB,SAAO,QAAQ,SAAS,IAAI,UAAU;AACxC;AAEA,SAAS,iBAAiB,KAAiC;AACzD,MAAI,CAAC,IAAK,QAAO;AACjB,QAAM,IAAI,OAAO,GAAG;AACpB,MAAI,CAAC,OAAO,SAAS,CAAC,KAAK,IAAI,EAAG,QAAO;AACzC,SAAO,KAAK,IAAI,KAAK,MAAM,CAAC,GAAG,EAAE;AACnC;AAEA,KAAK,EAAE,MAAM,CAAC,QAAQ;AACpB,UAAQ,OAAO,MAAM,sBAAsB,eAAe,QAAQ,IAAI,QAAQ,OAAO,GAAG,CAAC;AAAA,CAAI;AAC7F,UAAQ,KAAK,CAAC;AAChB,CAAC;","names":["result","output"]}
@@ -1,10 +1,131 @@
1
+ import { h as LoopSandboxClient } from '../types-Cu-SkGa0.js';
2
+ import { SandboxInstance } from '@tangle-network/sandbox';
1
3
  import { CoderOutput } from '../profiles.js';
2
- import { g as LoopSandboxClient } from '../types-Bx-tArkc.js';
3
- import '@tangle-network/sandbox';
4
4
  import '../runtime-run-B2j-hvBj.js';
5
5
  import '../types-DvJIha6w.js';
6
6
  import '@tangle-network/agent-eval';
7
7
 
8
+ /**
9
+ * @experimental
10
+ *
11
+ * Delegation executors — the layer between MCP delegates and the sandbox
12
+ * substrate. Each executor exposes a {@link LoopSandboxClient} the kernel
13
+ * consumes plus a placement tag so the trace pipeline can correlate workers
14
+ * with their physical placement.
15
+ *
16
+ * Two implementations ship in-box:
17
+ *
18
+ * - {@link createSiblingSandboxExecutor} — every delegation spawns a fresh
19
+ * sandbox sibling to the caller. Default when the MCP server runs as a
20
+ * standalone CLI mounted outside a fleet.
21
+ *
22
+ * - {@link createFleetWorkspaceExecutor} — delegations dispatch onto machines
23
+ * in the caller's existing fleet so worker diffs land directly on the
24
+ * caller's filesystem (the fleet's shared workspace). Selected when the
25
+ * parent sandbox passes `TANGLE_FLEET_ID` into the MCP server's env.
26
+ */
27
+
28
+ /** @experimental */
29
+ interface DelegationExecutor {
30
+ /** Sandbox client the kernel calls. Returned with `describePlacement` set. */
31
+ readonly client: LoopSandboxClient;
32
+ /** Best-effort one-liner used in stderr boot logs and diagnostics. */
33
+ describe(): string;
34
+ }
35
+ /** @experimental */
36
+ interface SiblingSandboxExecutorOptions {
37
+ client: LoopSandboxClient;
38
+ }
39
+ /**
40
+ * Wrap a raw sandbox SDK client so the kernel emits
41
+ * `loop.iteration.dispatch` events with `{ placement: 'sibling', sandboxId }`.
42
+ *
43
+ * The returned client `.create()` delegates to the underlying client; the
44
+ * only added behavior is a `describePlacement` tag the kernel reads.
45
+ *
46
+ * @experimental
47
+ */
48
+ declare function createSiblingSandboxExecutor(options: SiblingSandboxExecutorOptions): DelegationExecutor;
49
+ /**
50
+ * Minimal `SandboxFleet` surface the fleet executor calls. Declared
51
+ * structurally so tests can pass an in-memory stub without instantiating the
52
+ * sandbox SDK.
53
+ *
54
+ * @experimental
55
+ */
56
+ interface FleetHandle {
57
+ readonly fleetId: string;
58
+ /** Machine ids in dispatch-eligible order. The executor round-robins. */
59
+ readonly ids: ReadonlyArray<string>;
60
+ /** Resolve a machine id to its `SandboxInstance` — that machine is mounted
61
+ * on the fleet's shared workspace, so any diff the worker writes lands on
62
+ * every other fleet machine's filesystem too. */
63
+ sandbox(machineId: string): Promise<SandboxInstance>;
64
+ }
65
+ /** @experimental */
66
+ interface FleetWorkspaceExecutorOptions {
67
+ fleet: FleetHandle;
68
+ /**
69
+ * Override the machine-selection policy. Default = round-robin across
70
+ * `fleet.ids`, skipping the optional `excludeMachineIds` set (typically the
71
+ * coordinator machine the MCP server is running on).
72
+ */
73
+ selectMachine?: (call: {
74
+ callIndex: number;
75
+ ids: ReadonlyArray<string>;
76
+ }) => string;
77
+ /**
78
+ * Machine ids to skip during default round-robin. Set to the caller's own
79
+ * machineId so workers don't compete with the orchestrator on the same VM.
80
+ */
81
+ excludeMachineIds?: ReadonlyArray<string>;
82
+ }
83
+ /**
84
+ * Build an executor that resolves each delegated iteration to an existing
85
+ * machine in `fleet`. The fleet's shared-workspace policy means the worker
86
+ * machine sees the caller's filesystem — diffs land in-place with no
87
+ * cross-sandbox copy step.
88
+ *
89
+ * @experimental
90
+ */
91
+ declare function createFleetWorkspaceExecutor(options: FleetWorkspaceExecutorOptions): DelegationExecutor;
92
+
93
+ /**
94
+ * @experimental
95
+ *
96
+ * Helpers extracted from `bin.ts` so the env-detection + executor-selection
97
+ * logic is unit-testable without spawning a subprocess. The bin imports from
98
+ * here; tests import from here directly.
99
+ */
100
+
101
+ /** @experimental */
102
+ interface DetectExecutorArgs {
103
+ sandboxClient: LoopSandboxClient;
104
+ /** Raw env (defaults to `process.env`). Pass an explicit map for tests. */
105
+ env?: Record<string, string | undefined>;
106
+ /**
107
+ * Override how a fleet handle is resolved from the client + fleet id. The
108
+ * default reads `client.fleets.get(fleetId)` and validates the returned
109
+ * shape against the structural `FleetHandle` contract.
110
+ */
111
+ resolveFleet?: (client: LoopSandboxClient, fleetId: string) => Promise<FleetHandle>;
112
+ }
113
+ /**
114
+ * Pick the right executor for an MCP server invocation based on env vars.
115
+ *
116
+ * - `TANGLE_FLEET_ID` set → fleet-workspace placement; resolves the handle
117
+ * via `sandboxClient.fleets.get(...)`.
118
+ * - Otherwise → sibling-sandbox placement; each delegation creates a fresh
119
+ * sandbox via `sandboxClient.create(...)`.
120
+ *
121
+ * Fails loud (throws) when fleet mode is requested but the SDK shape is
122
+ * incompatible — the operator chose fleet semantics, silently degrading to
123
+ * sibling mode would lie about workspace topology.
124
+ *
125
+ * @experimental
126
+ */
127
+ declare function detectExecutor(args: DetectExecutorArgs): Promise<DelegationExecutor>;
128
+
8
129
  /**
9
130
  * @experimental
10
131
  *
@@ -215,7 +336,18 @@ type CoderDelegate = (args: DelegateCodeArgs, ctx: DelegateRunCtx) => Promise<Co
215
336
  type ResearcherDelegate = (args: DelegateResearchArgs, ctx: DelegateRunCtx) => Promise<ResearchOutputShape>;
216
337
  /** @experimental */
217
338
  interface CreateDefaultCoderDelegateOptions {
218
- sandboxClient: LoopSandboxClient;
339
+ /**
340
+ * Execution placement. Pass a {@link DelegationExecutor} (sibling or fleet)
341
+ * to control where worker iterations land. `sandboxClient` is a
342
+ * convenience shorthand that wraps the client in a sibling executor — pass
343
+ * one or the other, not both.
344
+ */
345
+ executor?: DelegationExecutor;
346
+ /**
347
+ * Convenience shorthand for sibling placement. Equivalent to
348
+ * `executor: createSiblingSandboxExecutor({ client: sandboxClient })`.
349
+ */
350
+ sandboxClient?: LoopSandboxClient;
219
351
  /** Default `['claude-code', 'codex', 'opencode/zai-coding-plan/glm-5.1']` when variants > 1. */
220
352
  fanoutHarnesses?: string[];
221
353
  /** Hard cap on the kernel's per-batch concurrency. Default 4. */
@@ -824,4 +956,4 @@ interface DelegationStatusHandlerOptions {
824
956
  /** @experimental */
825
957
  declare function createDelegationStatusHandler(options: DelegationStatusHandlerOptions): (raw: unknown) => Promise<DelegationStatusResult>;
826
958
 
827
- export { type CoderDelegate, type CreateDefaultCoderDelegateOptions, DELEGATE_CODE_DESCRIPTION, DELEGATE_CODE_INPUT_SCHEMA, DELEGATE_CODE_TOOL_NAME, DELEGATE_FEEDBACK_DESCRIPTION, DELEGATE_FEEDBACK_INPUT_SCHEMA, DELEGATE_FEEDBACK_TOOL_NAME, DELEGATE_RESEARCH_DESCRIPTION, DELEGATE_RESEARCH_INPUT_SCHEMA, DELEGATE_RESEARCH_TOOL_NAME, DELEGATION_HISTORY_DESCRIPTION, DELEGATION_HISTORY_INPUT_SCHEMA, DELEGATION_HISTORY_TOOL_NAME, DELEGATION_STATUS_DESCRIPTION, DELEGATION_STATUS_INPUT_SCHEMA, DELEGATION_STATUS_TOOL_NAME, type DelegateCodeArgs, type DelegateCodeConfig, type DelegateCodeResult, type DelegateFeedbackArgs, type DelegateFeedbackResult, type DelegateResearchArgs, type DelegateResearchConfig, type DelegateResearchResult, type DelegateRunCtx, type DelegationError, type DelegationFeedbackSnapshot, type DelegationHistoryArgs, type DelegationHistoryEntry, type DelegationHistoryResult, type DelegationProfile, type DelegationProgress, type DelegationRecord, type DelegationResultPayload, type DelegationStatus, type DelegationStatusArgs, type DelegationStatusResult, DelegationTaskQueue, type DelegationTaskQueueOptions, type FeedbackEvent, type FeedbackRating, type FeedbackRefersTo, type FeedbackStore, InMemoryFeedbackStore, type JsonRpcMessage, type JsonRpcResponse, type McpServer, type McpServerOptions, type McpToolDescriptor, type McpTransport, type ResearchOutputShape, type ResearchSource, type ResearcherDelegate, type SubmitInput, type SubmitOutput, createDefaultCoderDelegate, createDelegateCodeHandler, createDelegateFeedbackHandler, createDelegateResearchHandler, createDelegationHistoryHandler, createDelegationStatusHandler, createInProcessTransport, createMcpServer, eventToSnapshot, hashIdempotencyInput, validateDelegateCodeArgs, validateDelegateFeedbackArgs, validateDelegateResearchArgs, validateDelegationHistoryArgs, validateDelegationStatusArgs };
959
+ export { type CoderDelegate, type CreateDefaultCoderDelegateOptions, DELEGATE_CODE_DESCRIPTION, DELEGATE_CODE_INPUT_SCHEMA, DELEGATE_CODE_TOOL_NAME, DELEGATE_FEEDBACK_DESCRIPTION, DELEGATE_FEEDBACK_INPUT_SCHEMA, DELEGATE_FEEDBACK_TOOL_NAME, DELEGATE_RESEARCH_DESCRIPTION, DELEGATE_RESEARCH_INPUT_SCHEMA, DELEGATE_RESEARCH_TOOL_NAME, DELEGATION_HISTORY_DESCRIPTION, DELEGATION_HISTORY_INPUT_SCHEMA, DELEGATION_HISTORY_TOOL_NAME, DELEGATION_STATUS_DESCRIPTION, DELEGATION_STATUS_INPUT_SCHEMA, DELEGATION_STATUS_TOOL_NAME, type DelegateCodeArgs, type DelegateCodeConfig, type DelegateCodeResult, type DelegateFeedbackArgs, type DelegateFeedbackResult, type DelegateResearchArgs, type DelegateResearchConfig, type DelegateResearchResult, type DelegateRunCtx, type DelegationError, type DelegationExecutor, type DelegationFeedbackSnapshot, type DelegationHistoryArgs, type DelegationHistoryEntry, type DelegationHistoryResult, type DelegationProfile, type DelegationProgress, type DelegationRecord, type DelegationResultPayload, type DelegationStatus, type DelegationStatusArgs, type DelegationStatusResult, DelegationTaskQueue, type DelegationTaskQueueOptions, type DetectExecutorArgs, type FeedbackEvent, type FeedbackRating, type FeedbackRefersTo, type FeedbackStore, type FleetHandle, type FleetWorkspaceExecutorOptions, InMemoryFeedbackStore, type JsonRpcMessage, type JsonRpcResponse, type McpServer, type McpServerOptions, type McpToolDescriptor, type McpTransport, type ResearchOutputShape, type ResearchSource, type ResearcherDelegate, type SiblingSandboxExecutorOptions, type SubmitInput, type SubmitOutput, createDefaultCoderDelegate, createDelegateCodeHandler, createDelegateFeedbackHandler, createDelegateResearchHandler, createDelegationHistoryHandler, createDelegationStatusHandler, createFleetWorkspaceExecutor, createInProcessTransport, createMcpServer, createSiblingSandboxExecutor, detectExecutor, eventToSnapshot, hashIdempotencyInput, validateDelegateCodeArgs, validateDelegateFeedbackArgs, validateDelegateResearchArgs, validateDelegationHistoryArgs, validateDelegationStatusArgs };
package/dist/mcp/index.js CHANGED
@@ -22,8 +22,11 @@ import {
22
22
  createDelegateResearchHandler,
23
23
  createDelegationHistoryHandler,
24
24
  createDelegationStatusHandler,
25
+ createFleetWorkspaceExecutor,
25
26
  createInProcessTransport,
26
27
  createMcpServer,
28
+ createSiblingSandboxExecutor,
29
+ detectExecutor,
27
30
  eventToSnapshot,
28
31
  hashIdempotencyInput,
29
32
  validateDelegateCodeArgs,
@@ -31,8 +34,8 @@ import {
31
34
  validateDelegateResearchArgs,
32
35
  validateDelegationHistoryArgs,
33
36
  validateDelegationStatusArgs
34
- } from "../chunk-LPPM7EGS.js";
35
- import "../chunk-VFUEE6DF.js";
37
+ } from "../chunk-4XA4TWKW.js";
38
+ import "../chunk-EDVCVFQB.js";
36
39
  import "../chunk-Z5LKAYAS.js";
37
40
  import "../chunk-XLWPTPRP.js";
38
41
  import "../chunk-RZAOYKCO.js";
@@ -61,8 +64,11 @@ export {
61
64
  createDelegateResearchHandler,
62
65
  createDelegationHistoryHandler,
63
66
  createDelegationStatusHandler,
67
+ createFleetWorkspaceExecutor,
64
68
  createInProcessTransport,
65
69
  createMcpServer,
70
+ createSiblingSandboxExecutor,
71
+ detectExecutor,
66
72
  eventToSnapshot,
67
73
  hashIdempotencyInput,
68
74
  validateDelegateCodeArgs,
@@ -1,5 +1,5 @@
1
1
  import { AgentProfile } from '@tangle-network/sandbox';
2
- import { O as OutputAdapter, V as Validator, A as AgentRunSpec, a as Driver } from './types-Bx-tArkc.js';
2
+ import { O as OutputAdapter, V as Validator, A as AgentRunSpec, a as Driver } from './types-Cu-SkGa0.js';
3
3
  import './runtime-run-B2j-hvBj.js';
4
4
  import './types-DvJIha6w.js';
5
5
  import '@tangle-network/agent-eval';
@@ -135,10 +135,24 @@ interface LoopResult<Task, Output, Decision> {
135
135
  * `new Sandbox({ apiKey, baseUrl })` — declared as a structural type so
136
136
  * tests can pass a stub without instantiating the SDK.
137
137
  *
138
+ * `describePlacement` is optional. When present, the kernel calls it after
139
+ * each `create()` so the `loop.iteration.dispatch` trace event carries fleet
140
+ * coordinates (fleetId + machineId) instead of just the sibling sandboxId.
141
+ * Fleet-aware adapters set this; the raw `Sandbox` SDK class does not, and
142
+ * the kernel falls back to `{ placement: 'sibling', sandboxId: box.id }`.
143
+ *
138
144
  * @experimental
139
145
  */
140
146
  interface LoopSandboxClient {
141
147
  create(options?: CreateSandboxOptions): Promise<SandboxInstance>;
148
+ describePlacement?(box: SandboxInstance): LoopSandboxPlacement;
149
+ }
150
+ /** @experimental */
151
+ interface LoopSandboxPlacement {
152
+ kind: 'sibling' | 'fleet';
153
+ sandboxId?: string;
154
+ fleetId?: string;
155
+ machineId?: string;
142
156
  }
143
157
  /** @experimental */
144
158
  interface LoopTraceEmitter {
@@ -155,6 +169,11 @@ type LoopTraceEvent = {
155
169
  runId: string;
156
170
  timestamp: number;
157
171
  payload: LoopIterationStartedPayload;
172
+ } | {
173
+ kind: 'loop.iteration.dispatch';
174
+ runId: string;
175
+ timestamp: number;
176
+ payload: LoopIterationDispatchPayload;
158
177
  } | {
159
178
  kind: 'loop.iteration.ended';
160
179
  runId: string;
@@ -184,6 +203,25 @@ interface LoopIterationStartedPayload {
184
203
  agentRunName: string;
185
204
  taskHash: string;
186
205
  }
206
+ /**
207
+ * Where the iteration's worker was placed. `sibling` = a fresh sandbox the
208
+ * kernel created via `sandboxClient.create`. `fleet` = an existing machine in
209
+ * a shared-workspace fleet — workers see the caller's filesystem and any diff
210
+ * they write lands on it directly.
211
+ *
212
+ * @experimental
213
+ */
214
+ interface LoopIterationDispatchPayload {
215
+ iterationIndex: number;
216
+ agentRunName: string;
217
+ placement: 'sibling' | 'fleet';
218
+ /** Set on every placement. Lets analyst loops correlate per-iteration logs. */
219
+ sandboxId?: string;
220
+ /** Set only when `placement === 'fleet'`. */
221
+ fleetId?: string;
222
+ /** Set only when `placement === 'fleet'`. */
223
+ machineId?: string;
224
+ }
187
225
  /** @experimental */
188
226
  interface LoopIterationEndedPayload {
189
227
  iterationIndex: number;
@@ -222,4 +260,4 @@ interface ExecCtx {
222
260
  signal?: AbortSignal;
223
261
  }
224
262
 
225
- export type { AgentRunSpec as A, DefaultVerdict as D, ExecCtx as E, Iteration as I, LoopWinner as L, OutputAdapter as O, Validator as V, Driver as a, LoopResult as b, LoopDecisionPayload as c, LoopEndedPayload as d, LoopIterationEndedPayload as e, LoopIterationStartedPayload as f, LoopSandboxClient as g, LoopStartedPayload as h, LoopTraceEmitter as i, LoopTraceEvent as j, ValidationCtx as k };
263
+ export type { AgentRunSpec as A, DefaultVerdict as D, ExecCtx as E, Iteration as I, LoopWinner as L, OutputAdapter as O, Validator as V, Driver as a, LoopResult as b, LoopDecisionPayload as c, LoopEndedPayload as d, LoopIterationDispatchPayload as e, LoopIterationEndedPayload as f, LoopIterationStartedPayload as g, LoopSandboxClient as h, LoopSandboxPlacement as i, LoopStartedPayload as j, LoopTraceEmitter as k, LoopTraceEvent as l, ValidationCtx as m };
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@tangle-network/agent-runtime",
3
- "version": "0.20.4",
3
+ "version": "0.21.0",
4
4
  "description": "Reusable runtime lifecycle for domain-specific agents.",
5
5
  "homepage": "https://github.com/tangle-network/agent-runtime#readme",
6
6
  "repository": {