@tangle-network/agent-runtime 0.94.6 → 0.94.8

package/dist/chunk-M22PL4GH.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../src/runtime/sandbox-events.ts"],"sourcesContent":["/**\n * Sandbox-event → runtime-event mapping.\n *\n * The sandbox SDK emits a polymorphic `SandboxEvent = { type, data, id? }`\n * whose `type` vocabulary is backend-determined (opencode, etc.) rather than\n * enumerated by the SDK. Two consumers project it:\n *   - the loop kernel's cost ledger (`extractLlmCallEvent`) — sums usage off\n *     every cost-bearing event, regardless of stream shape;\n *   - the `AgentRuntime.act` streaming contract (`mapSandboxEvent`) — projects\n *     incremental events to the `RuntimeStreamEvent` chat-UX vocabulary.\n *\n * Both live here so the empirically-observed `type` vocabulary has one home.\n */\n\nimport type { SandboxEvent } from '@tangle-network/sandbox'\nimport type { RuntimeStreamEvent } from '../types'\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 * 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` / `usage` — same shape under a dedicated type\n *\n * Numeric coercion is strict: `Number.isFinite` gates every accumulator write\n * so a sentinel `NaN` from a misbehaving backend cannot poison the ledger.\n */\nexport function 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  // sandbox 0.4.0 terminal event: `data = { tokenUsage: { inputTokens, outputTokens,\n  // reasoningTokens, cacheReadInputTokens }, totalCostUsd }`. Usage lives under\n  // `tokenUsage` (not `usage`) and the cost is top-level — neither matched the\n  // branches above, so an in-process loopDispatch run reported {0,0} and the\n  // backend-integrity guard misread a real run as a stub. Reasoning tokens are\n  // billed output (reasoning models), so they fold into the output count.\n  if (type === 'done') {\n    const usage = data.tokenUsage as Record<string, unknown> | undefined\n    if (!usage || typeof usage !== 'object') return undefined\n    const out = pickFiniteNumber(usage, ['outputTokens', 'completion_tokens', 'tokensOut'])\n    const reasoning = pickFiniteNumber(usage, ['reasoningTokens'])\n    const mergedOut =\n      out !== undefined || reasoning !== undefined ? (out ?? 0) + (reasoning ?? 0) : undefined\n    return buildLlmCall(\n      {\n        inputTokens: usage.inputTokens,\n        outputTokens: mergedOut,\n        totalCostUsd: data.totalCostUsd,\n        model: data.model ?? usage.model,\n      },\n      agentRunName,\n    )\n  }\n  return undefined\n}\n\n/**\n * Sum the token usage + USD cost of a sandbox turn's events — the one honest way to meter an\n * `openSandboxRun` cell. Folds `extractLlmCallEvent` over the stream (which reads usage off EVERY backend\n * event shape), so a `runProfileMatrix` dispatch can report it to `ctx.cost`:\n *\n *     receipt: (turn) => {\n *       const u = sumSandboxUsage(turn.events)\n *       return { model, inputTokens: u.input, outputTokens: u.output,\n *         ...(u.costUsd > 0 ? { actualCostUsd: u.costUsd } : {}) }\n *     }\n *\n * Without this a cell reads `{tokens:0, cost:0}` and the backend-integrity guard correctly aborts the\n * matrix as a stub. `agentRunName` is the fallback model label for cost-only events (default `'agent'`).\n */\nexport function sumSandboxUsage(\n  events: readonly SandboxEvent[],\n  agentRunName = 'agent',\n): { input: number; output: number; costUsd: number } {\n  let input = 0\n  let output = 0\n  let costUsd = 0\n  for (const ev of events) {\n    const call = extractLlmCallEvent(ev, agentRunName)\n    if (!call) continue\n    input += call.tokensIn ?? 0\n    output += call.tokensOut ?? 0\n    costUsd += call.costUsd ?? 0\n  }\n  return { input, output, costUsd }\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 * Cross-event state for {@link mapSandboxToolEvent}. Sandbox backends emit a\n * tool invocation as MANY `message.part.updated` frames on the same call id\n * (pending → running → completed), so faithful projection needs per-call\n * status memory: one `tool_call` on first sighting, at most one `tool_result`\n * on the terminal transition, nothing on intermediate re-frames. Create one\n * state per turn via {@link createSandboxToolPartState}.\n *\n * @experimental\n */\nexport interface SandboxToolPartState {\n  /** Last seen status per tool call id. A terminal status is sticky — later\n   *  frames on a settled call project to nothing. */\n  statusByCall: Map<string, string>\n  /** Sequence for synthesized call ids when an event carries none. */\n  seq: number\n}\n\n/**\n * Fresh per-turn {@link SandboxToolPartState} for {@link mapSandboxToolEvent} — an\n * empty call-status map so each turn projects tool frames independently.\n *\n * @experimental\n */\nexport function createSandboxToolPartState(): SandboxToolPartState {\n  return { statusByCall: new Map(), seq: 0 }\n}\n\n/** Terminal tool statuses that are failures (everything here settles the call). */\nconst TERMINAL_TOOL_FAILURE =\n  /^(error|errored|failed|failure|cancelled|canceled|timeout|timed_out)$/i\n\n/**\n * Project one `SandboxEvent` onto the `tool_call` / `tool_result` variants of\n * `RuntimeStreamEvent` — the tool-part projection `mapSandboxEvent`\n * deliberately does NOT perform. Opt-in and additive: `mapSandboxEvent`'s\n * default vocabulary (text/reasoning deltas + `llm_call`) is unchanged;\n * consumers that need the tool surface (chat UIs rendering tool activity)\n * compose this projector alongside it — `streamAgentTurn` does exactly that\n * under its `preserveToolParts` option.\n *\n * Handled shapes (observed on the opencode / claude-code sandbox backends):\n *   - `message.part.updated` with `part.type === 'tool'` — stateful: a\n *     `tool_call` on the call id's first frame (args from `state.input` or\n *     `state.metadata.input`), a `tool_result` when the status transitions to\n *     `completed` (result from `state.output` / `metadata.output`) or to a\n *     terminal failure (result is `{ error, status, output? }` — the error\n *     surfaced in-band, never dropped).\n *   - bare `tool*` event types (`tool.call`, `tool_result`, …) — stateless:\n *     `*result*` types project to `tool_result`, the rest to `tool_call`.\n *\n * Returns `[]` for every non-tool event.\n *\n * @experimental\n */\nexport function mapSandboxToolEvent(\n  event: SandboxEvent,\n  state: SandboxToolPartState,\n): (RuntimeStreamEvent & { type: 'tool_call' | 'tool_result' })[] {\n  if (!event || typeof event !== 'object') return []\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 === 'message.part.updated') {\n    const part =\n      data.part && typeof data.part === 'object' ? (data.part as Record<string, unknown>) : {}\n    if (String(part.type ?? '') !== 'tool') return []\n    return projectToolPart(part, state, typeof event.id === 'string' ? event.id : undefined)\n  }\n\n  if (type.includes('tool')) {\n    const callId =\n      pickString(data, ['toolCallId', 'tool_use_id', 'id']) ??\n      (typeof event.id === 'string' ? event.id : undefined) ??\n      `sandbox-tool-${++state.seq}`\n    const toolName = pickString(data, ['name', 'toolName', 'tool']) ?? 'sandbox_tool'\n    if (type.includes('result')) {\n      return [\n        {\n          type: 'tool_result',\n          toolName,\n          toolCallId: callId,\n          result: data.output ?? data.result ?? data.content ?? data,\n        },\n      ]\n    }\n    return [\n      { type: 'tool_call', toolName, toolCallId: callId, args: data.input ?? data.args ?? {} },\n    ]\n  }\n\n  return []\n}\n\nfunction projectToolPart(\n  part: Record<string, unknown>,\n  state: SandboxToolPartState,\n  eventId: string | undefined,\n): (RuntimeStreamEvent & { type: 'tool_call' | 'tool_result' })[] {\n  const callId =\n    pickString(part, ['callID', 'callId', 'toolCallId', 'id']) ??\n    eventId ??\n    `sandbox-tool-${++state.seq}`\n  const toolName = pickString(part, ['tool', 'toolName', 'name']) ?? 'sandbox_tool'\n  const toolState =\n    part.state && typeof part.state === 'object' ? (part.state as Record<string, unknown>) : {}\n  const metadata =\n    toolState.metadata && typeof toolState.metadata === 'object'\n      ? (toolState.metadata as Record<string, unknown>)\n      : {}\n  const status = pickString(toolState, ['status']) ?? 'updated'\n\n  const previous = state.statusByCall.get(callId)\n  const settled =\n    previous === 'completed' || (previous !== undefined && TERMINAL_TOOL_FAILURE.test(previous))\n  if (settled) return []\n\n  const out: (RuntimeStreamEvent & { type: 'tool_call' | 'tool_result' })[] = []\n  if (previous === undefined) {\n    out.push({\n      type: 'tool_call',\n      toolName,\n      toolCallId: callId,\n      args: toolState.input ?? metadata.input ?? {},\n    })\n  }\n  state.statusByCall.set(callId, status)\n\n  if (status === 'completed') {\n    out.push({\n      type: 'tool_result',\n      toolName,\n      toolCallId: callId,\n      result: toolState.output ?? metadata.output ?? '',\n    })\n  } else if (TERMINAL_TOOL_FAILURE.test(status)) {\n    const message =\n      pickString(toolState, ['error', 'message']) ??\n      pickString(metadata, ['error', 'message']) ??\n      `sandbox tool ended with status ${status}`\n    const output = toolState.output ?? metadata.output\n    out.push({\n      type: 'tool_result',\n      toolName,\n      toolCallId: callId,\n      result: { error: message, status, ...(output !== undefined ? { output } : {}) },\n    })\n  }\n  return out\n}\n\nfunction pickString(data: Record<string, unknown>, keys: string[]): string | undefined {\n  for (const key of keys) {\n    const value = data[key]\n    if (typeof value === 'string' && value.length > 0) return value\n  }\n  return undefined\n}\n\n/**\n * Project one `SandboxEvent` onto the `RuntimeStreamEvent` chat-UX vocabulary,\n * for runtimes that bridge a sandbox `streamPrompt` into the\n * `AgentRuntime.act` streaming contract. Returns `undefined` for events that\n * have no faithful projection — the raw stream is preserved separately for the\n * `OutputAdapter`, so an unmapped event never loses data.\n *\n * Mapped (the task-optional incremental variants — no synthesized task\n * lifecycle, no guessed tool-part shapes):\n *   - `message.part.updated` text part → `text_delta`\n *   - `message.part.updated` reasoning/thinking part → `reasoning_delta`\n *   - cost-bearing events → `llm_call` (shared with the ledger extractor)\n *\n * Tool parts are deliberately NOT mapped here (unchanged default) — compose\n * {@link mapSandboxToolEvent} alongside when a consumer needs them.\n *\n * The opencode backend emits incremental text as\n * `{ type: 'message.part.updated', data: { part: { type, text }, delta } }`;\n * `delta` is the increment, `part.text` the running accumulation.\n */\nexport function mapSandboxEvent(\n  event: SandboxEvent,\n  opts: { agentRunName?: string } = {},\n): RuntimeStreamEvent | 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 === 'message.part.updated') {\n    const part =\n      data.part && typeof data.part === 'object' ? (data.part as Record<string, unknown>) : {}\n    const partType = String(part.type ?? '')\n    const delta = typeof data.delta === 'string' ? data.delta : undefined\n    const text = delta ?? (typeof part.text === 'string' ? part.text : undefined)\n    if (text === undefined) return undefined\n    if (partType === 'text') return { type: 'text_delta', text }\n    if (partType === 'reasoning' || partType === 'thinking')\n      return { type: 'reasoning_delta', text }\n    return undefined\n  }\n\n  return extractLlmCallEvent(event, opts.agentRunName ?? 'agent')\n}\n"],"mappings":";AA+BO,SAAS,oBACd,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;AAOA,MAAI,SAAS,QAAQ;AACnB,UAAM,QAAQ,KAAK;AACnB,QAAI,CAAC,SAAS,OAAO,UAAU,SAAU,QAAO;AAChD,UAAM,MAAM,iBAAiB,OAAO,CAAC,gBAAgB,qBAAqB,WAAW,CAAC;AACtF,UAAM,YAAY,iBAAiB,OAAO,CAAC,iBAAiB,CAAC;AAC7D,UAAM,YACJ,QAAQ,UAAa,cAAc,UAAa,OAAO,MAAM,aAAa,KAAK;AACjF,WAAO;AAAA,MACL;AAAA,QACE,aAAa,MAAM;AAAA,QACnB,cAAc;AAAA,QACd,cAAc,KAAK;AAAA,QACnB,OAAO,KAAK,SAAS,MAAM;AAAA,MAC7B;AAAA,MACA;AAAA,IACF;AAAA,EACF;AACA,SAAO;AACT;AAgBO,SAAS,gBACd,QACA,eAAe,SACqC;AACpD,MAAI,QAAQ;AACZ,MAAI,SAAS;AACb,MAAI,UAAU;AACd,aAAW,MAAM,QAAQ;AACvB,UAAM,OAAO,oBAAoB,IAAI,YAAY;AACjD,QAAI,CAAC,KAAM;AACX,aAAS,KAAK,YAAY;AAC1B,cAAU,KAAK,aAAa;AAC5B,eAAW,KAAK,WAAW;AAAA,EAC7B;AACA,SAAO,EAAE,OAAO,QAAQ,QAAQ;AAClC;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;AA0BO,SAAS,6BAAmD;AACjE,SAAO,EAAE,cAAc,oBAAI,IAAI,GAAG,KAAK,EAAE;AAC3C;AAGA,IAAM,wBACJ;AAyBK,SAAS,oBACd,OACA,OACgE;AAChE,MAAI,CAAC,SAAS,OAAO,UAAU,SAAU,QAAO,CAAC;AACjD,QAAM,OAAO,OAAO,MAAM,QAAQ,EAAE;AACpC,QAAM,OACJ,MAAM,QAAQ,OAAO,MAAM,SAAS,WAC/B,MAAM,OACN,CAAC;AAER,MAAI,SAAS,wBAAwB;AACnC,UAAM,OACJ,KAAK,QAAQ,OAAO,KAAK,SAAS,WAAY,KAAK,OAAmC,CAAC;AACzF,QAAI,OAAO,KAAK,QAAQ,EAAE,MAAM,OAAQ,QAAO,CAAC;AAChD,WAAO,gBAAgB,MAAM,OAAO,OAAO,MAAM,OAAO,WAAW,MAAM,KAAK,MAAS;AAAA,EACzF;AAEA,MAAI,KAAK,SAAS,MAAM,GAAG;AACzB,UAAM,SACJ,WAAW,MAAM,CAAC,cAAc,eAAe,IAAI,CAAC,MACnD,OAAO,MAAM,OAAO,WAAW,MAAM,KAAK,WAC3C,gBAAgB,EAAE,MAAM,GAAG;AAC7B,UAAM,WAAW,WAAW,MAAM,CAAC,QAAQ,YAAY,MAAM,CAAC,KAAK;AACnE,QAAI,KAAK,SAAS,QAAQ,GAAG;AAC3B,aAAO;AAAA,QACL;AAAA,UACE,MAAM;AAAA,UACN;AAAA,UACA,YAAY;AAAA,UACZ,QAAQ,KAAK,UAAU,KAAK,UAAU,KAAK,WAAW;AAAA,QACxD;AAAA,MACF;AAAA,IACF;AACA,WAAO;AAAA,MACL,EAAE,MAAM,aAAa,UAAU,YAAY,QAAQ,MAAM,KAAK,SAAS,KAAK,QAAQ,CAAC,EAAE;AAAA,IACzF;AAAA,EACF;AAEA,SAAO,CAAC;AACV;AAEA,SAAS,gBACP,MACA,OACA,SACgE;AAChE,QAAM,SACJ,WAAW,MAAM,CAAC,UAAU,UAAU,cAAc,IAAI,CAAC,KACzD,WACA,gBAAgB,EAAE,MAAM,GAAG;AAC7B,QAAM,WAAW,WAAW,MAAM,CAAC,QAAQ,YAAY,MAAM,CAAC,KAAK;AACnE,QAAM,YACJ,KAAK,SAAS,OAAO,KAAK,UAAU,WAAY,KAAK,QAAoC,CAAC;AAC5F,QAAM,WACJ,UAAU,YAAY,OAAO,UAAU,aAAa,WAC/C,UAAU,WACX,CAAC;AACP,QAAM,SAAS,WAAW,WAAW,CAAC,QAAQ,CAAC,KAAK;AAEpD,QAAM,WAAW,MAAM,aAAa,IAAI,MAAM;AAC9C,QAAM,UACJ,aAAa,eAAgB,aAAa,UAAa,sBAAsB,KAAK,QAAQ;AAC5F,MAAI,QAAS,QAAO,CAAC;AAErB,QAAM,MAAsE,CAAC;AAC7E,MAAI,aAAa,QAAW;AAC1B,QAAI,KAAK;AAAA,MACP,MAAM;AAAA,MACN;AAAA,MACA,YAAY;AAAA,MACZ,MAAM,UAAU,SAAS,SAAS,SAAS,CAAC;AAAA,IAC9C,CAAC;AAAA,EACH;AACA,QAAM,aAAa,IAAI,QAAQ,MAAM;AAErC,MAAI,WAAW,aAAa;AAC1B,QAAI,KAAK;AAAA,MACP,MAAM;AAAA,MACN;AAAA,MACA,YAAY;AAAA,MACZ,QAAQ,UAAU,UAAU,SAAS,UAAU;AAAA,IACjD,CAAC;AAAA,EACH,WAAW,sBAAsB,KAAK,MAAM,GAAG;AAC7C,UAAM,UACJ,WAAW,WAAW,CAAC,SAAS,SAAS,CAAC,KAC1C,WAAW,UAAU,CAAC,SAAS,SAAS,CAAC,KACzC,kCAAkC,MAAM;AAC1C,UAAM,SAAS,UAAU,UAAU,SAAS;AAC5C,QAAI,KAAK;AAAA,MACP,MAAM;AAAA,MACN;AAAA,MACA,YAAY;AAAA,MACZ,QAAQ,EAAE,OAAO,SAAS,QAAQ,GAAI,WAAW,SAAY,EAAE,OAAO,IAAI,CAAC,EAAG;AAAA,IAChF,CAAC;AAAA,EACH;AACA,SAAO;AACT;AAEA,SAAS,WAAW,MAA+B,MAAoC;AACrF,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,KAAK,GAAG;AACtB,QAAI,OAAO,UAAU,YAAY,MAAM,SAAS,EAAG,QAAO;AAAA,EAC5D;AACA,SAAO;AACT;AAsBO,SAAS,gBACd,OACA,OAAkC,CAAC,GACH;AAChC,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,wBAAwB;AACnC,UAAM,OACJ,KAAK,QAAQ,OAAO,KAAK,SAAS,WAAY,KAAK,OAAmC,CAAC;AACzF,UAAM,WAAW,OAAO,KAAK,QAAQ,EAAE;AACvC,UAAM,QAAQ,OAAO,KAAK,UAAU,WAAW,KAAK,QAAQ;AAC5D,UAAM,OAAO,UAAU,OAAO,KAAK,SAAS,WAAW,KAAK,OAAO;AACnE,QAAI,SAAS,OAAW,QAAO;AAC/B,QAAI,aAAa,OAAQ,QAAO,EAAE,MAAM,cAAc,KAAK;AAC3D,QAAI,aAAa,eAAe,aAAa;AAC3C,aAAO,EAAE,MAAM,mBAAmB,KAAK;AACzC,WAAO;AAAA,EACT;AAEA,SAAO,oBAAoB,OAAO,KAAK,gBAAgB,OAAO;AAChE;","names":[]}

package/dist/{chunk-4VMRCEI6.js → chunk-MSUWXCHD.js}

@@ -8,7 +8,7 @@ import {
   DELEGATION_STATUS_DESCRIPTION,
   DELEGATION_STATUS_INPUT_SCHEMA,
   DELEGATION_STATUS_TOOL_NAME
-} from "./chunk-LWNMCOEA.js";
+} from "./chunk-FRXDOBFP.js";
 
 // src/mcp/openai-tools.ts
 function buildTool(name, description, parameters) {
@@ -45,4 +45,4 @@ export {
   mcpToolsForRuntimeMcp,
   mcpToolsForRuntimeMcpSubset
 };
-//# sourceMappingURL=chunk-4VMRCEI6.js.map
+//# sourceMappingURL=chunk-MSUWXCHD.js.map

package/dist/{chunk-ZQZX77MM.js → chunk-QDSOD7RC.js}

@@ -1,6 +1,6 @@
 import {
   extractLlmCallEvent
-} from "./chunk-FVJ7M3DA.js";
+} from "./chunk-M22PL4GH.js";
 import {
   AnalystError
 } from "./chunk-YEJR7IXO.js";
@@ -579,4 +579,4 @@ export {
   iterationsToTraceStore,
   runAnalystLoop
 };
-//# sourceMappingURL=chunk-ZQZX77MM.js.map
+//# sourceMappingURL=chunk-QDSOD7RC.js.map

package/dist/{chunk-RNWDH5Z3.js → chunk-TNUBPPZA.js}

@@ -3,7 +3,7 @@ import {
 } from "./chunk-5QOB7H74.js";
 import {
   supervise
-} from "./chunk-LWNMCOEA.js";
+} from "./chunk-FRXDOBFP.js";
 
 // src/knowledge/improvement-job.ts
 import {
@@ -211,4 +211,4 @@ export {
   createAgentKnowledgeReadinessCheck,
   runKnowledgeImprovementJob
 };
-//# sourceMappingURL=chunk-RNWDH5Z3.js.map
+//# sourceMappingURL=chunk-TNUBPPZA.js.map

package/dist/{completion-gate-CfX2thvL.d.ts → completion-gate-tzwyyD-E.d.ts}

@@ -1,5 +1,5 @@
-import { L as LocalHarness, C as CodexTokenUsage, a as CodexExecutionPolicy } from './local-harness-CePtFrwu.js';
-import { c as Executor } from './types-DHLXdLVN.js';
+import { L as LocalHarness, C as CodexTokenUsage, b as CodexExecutionPolicy } from './local-harness-ZLMkKdZc.js';
+import { c as Executor } from './types-CogNCaw7.js';
 
 /**
  *

package/dist/{coordination-pOGZuYS7.d.ts → coordination-o0TzS7Ms.d.ts}

@@ -1,12 +1,12 @@
-import { E as ExecutorFactory, e as ExecutorRegistry, i as Spend, A as Agent, S as Scope, a as ResultBlobStore, B as Budget } from './types-DHLXdLVN.js';
+import { E as ExecutorFactory, e as ExecutorRegistry, i as Spend, A as Agent, S as Scope, a as ResultBlobStore, B as Budget } from './types-CogNCaw7.js';
 import { AgentProfile } from '@tangle-network/agent-interface';
 import { U as UiLens, a as UiFinding, C as CoderTask } from './substrate-DO2GHNg2.js';
-import { S as SandboxClient, E as ExecCtx, g as LoopTraceEmitter, e as LoopTraceEvent, a as RuntimeStreamEvent, d as AgentRunSpec } from './types-Kjr5W6OF.js';
+import { S as SandboxClient, E as ExecCtx, g as LoopTraceEmitter, e as LoopTraceEvent, a as RuntimeStreamEvent, d as AgentRunSpec } from './types-CKmyZ9TP.js';
 import { BackendType, SandboxEvent, SandboxInstance } from '@tangle-network/sandbox';
 import { AgentEvalError } from '@tangle-network/agent-eval';
-import { b as ToolSpec, c as RuntimeTelemetryOptions, R as RouterConfig } from './sanitize-Dr3SliRC.js';
-import { G as GitRunner, a as WorktreeCheckRunner, D as DeliverableSpec } from './completion-gate-CfX2thvL.js';
-import { L as LocalHarness } from './local-harness-CePtFrwu.js';
+import { b as ToolSpec, c as RuntimeTelemetryOptions, R as RouterConfig } from './sanitize-Dcbjp0tU.js';
+import { G as GitRunner, a as WorktreeCheckRunner, D as DeliverableSpec } from './completion-gate-tzwyyD-E.js';
+import { L as LocalHarness } from './local-harness-ZLMkKdZc.js';
 import { ProviderExecutorOptions, AgentEnvironmentProviderRegistry } from './environment-provider.js';
 import { AgentEnvironmentProvider } from '@tangle-network/agent-interface/environment-provider';
 

package/dist/environment-provider.d.ts

@@ -2,8 +2,8 @@ import { AgentProfile, AgentProfileValidationResult } from '@tangle-network/agen
 import { CreateAgentEnvironmentInput, AgentTurnInput, AgentEnvironmentProvider, AgentEnvironmentCapabilities, AgentProfileRef } from '@tangle-network/agent-interface/environment-provider';
 export { AgentEnvironment, AgentEnvironmentCapabilities, AgentEnvironmentEvent, AgentEnvironmentProvider, AgentEnvironmentQuery, AgentEnvironmentStatus, AgentEnvironmentSummary, AgentProfileRef, AgentSession, AgentSessionRef, AgentSessionStatus, AgentTurnInput, AgentTurnResult, CheckpointRef, CheckpointRequest, CreateAgentEnvironmentInput, ExecRequest, ExecResult, ForkRequest, PlacementInfo, ResourceRequest, WorkspaceRequest } from '@tangle-network/agent-interface/environment-provider';
 import { CreateSandboxOptions, BackendType } from '@tangle-network/sandbox';
-import { R as Runtime, E as ExecutorFactory } from './types-DHLXdLVN.js';
-import { S as SandboxClient } from './types-Kjr5W6OF.js';
+import { R as Runtime, E as ExecutorFactory } from './types-CogNCaw7.js';
+import { S as SandboxClient } from './types-CKmyZ9TP.js';
 import '@tangle-network/agent-eval';
 
 /** Provider object or registry name accepted by runtime provider adapters.

package/dist/{improve-Ciuvu2jh.d.ts → improve-DZeatcs7.d.ts}

@@ -1,7 +1,8 @@
+import { WorktreeAdapter } from '@tangle-network/agent-eval/campaign';
 import { Scenario, SelfImproveOptions, SurfaceProposer, SelfImproveResult, MutableSurface } from '@tangle-network/agent-eval/contract';
 import { AgentProfile } from '@tangle-network/agent-interface';
-import { L as LocalHarness } from './local-harness-CePtFrwu.js';
-import { V as Verifier, C as CandidateGenerator } from './agentic-generator-CL1q0H91.js';
+import { L as LocalHarness } from './local-harness-ZLMkKdZc.js';
+import { V as Verifier, C as CandidateGenerator } from './agentic-generator-1g7GDba7.js';
 
 /**
  *
@@ -21,13 +22,14 @@ import { V as Verifier, C as CandidateGenerator } from './agentic-generator-CL1q
  *     lesson document supplied through `opts.memory`.
  *   - `surface: 'agent-profile'` → caller-supplied proposer mutates the complete
  *     canonical AgentProfile JSON in one candidate.
- *   - `surface` ∈ {`tools`, `mcp`, `hooks`, `subagents`, `agent-profile`, `code`} → no zero-config default
+ *   - `surface` ∈ {`tools`, `mcp`, `hooks`, `subagents`, `agent-profile`} → no zero-config default
  *     proposer exists (a code/config proposer needs caller-supplied wiring — a
  *     worktree repo root, a candidate generator, a serializer). The facade
  *     requires an explicit `opts.generator` for these and throws a `ConfigError`
  *     otherwise. This is a designed boundary, not a missing default: there is
- *     no safe value the facade could invent for those surfaces. Code also
- *     requires `opts.code.repoRoot` so its incumbent is a real isolated checkout.
+ *     no safe value the facade could invent for those surfaces. Code instead
+ *     requires `opts.code.repoRoot` and accepts only the runtime-owned
+ *     `opts.code.generator` path so every isolated checkout can be released.
  *
  * Everything else (`scenarios`, `judge`, `agent`, `budget`, `llm`) passes
  * straight through to `selfImprove`.
@@ -43,9 +45,10 @@ type ImproveOptions<TScenario extends Scenario, TArtifact> = Omit<SelfImproveOpt
     /** Which profile lever to optimize. Default `'prompt'`. Selects the default
      *  generator + the baseline-surface extraction shape. */
     surface?: ImproveSurface;
-    /** The `SurfaceProposer` that mutates the surface. When unset, the facade
+    /** The `SurfaceProposer` that mutates a profile surface. When unset, the facade
      *  picks the default for prompt, skills, and memory; surfaces
-     *  with no default REQUIRE this (fail-loud otherwise). */
+     *  with no default REQUIRE this (fail-loud otherwise). Forbidden for code;
+     *  use `code.generator` so the runtime owns candidate cleanup. */
     generator?: SurfaceProposer;
     /** Gate mode. `'holdout'` (default) runs the held-out promotion gate;
      *  `'none'` is a baseline-only run (`budget.generations = 0`). */
@@ -116,6 +119,9 @@ interface ImproveCodeOptions {
     baseRef?: string;
     /** Directory worktrees are created under. Default `<repoRoot>/.worktrees`. */
     worktreeDir?: string;
+    /** Git-compatible adapter override, primarily for tests. Candidate advancement
+     *  still requires normal Git worktree and commit semantics. */
+    worktree?: WorktreeAdapter;
     /** Coding harness the agentic generator runs in each worktree. Default `claude`. */
     harness?: LocalHarness;
     /** Verify a candidate worktree before it becomes a measurable surface; failures
@@ -137,8 +143,13 @@ interface ImproveResult<TScenario extends Scenario, TArtifact> {
     lift: number;
     /** The five-valued gate verdict from `selfImprove`. */
     gateDecision: SelfImproveResult<TScenario, TArtifact>['gateDecision'];
-    /** Full `selfImprove` result for advanced inspection. */
+    /** Full `selfImprove` result for advanced inspection. For code runs,
+     *  `raw.winner.surface.worktreeRef` remains live after return whether the
+     *  candidate shipped or held; call `dispose()` after consuming it. */
     raw: SelfImproveResult<TScenario, TArtifact>;
+    /** Release resources owned by this result. Idempotent; currently disposes
+     *  the returned code worktree and is a no-op for profile-only surfaces. */
+    dispose(): Promise<void>;
 }
 /** Apply a promoted winner surface back into the profile field for `surface`.
  *  Returns a shallow copy; never mutates the input profile. */

package/dist/index.d.ts

@@ -1,33 +1,33 @@
-import { AgentProfile, AgentEvalError, AnalystFinding, KnowledgeReadinessReport, RunRecord, ControlEvalResult } from '@tangle-network/agent-eval';
+import { AgentProfile, MaximumCharge, AgentEvalError, AnalystFinding, KnowledgeReadinessReport, RunRecord, ControlEvalResult } from '@tangle-network/agent-eval';
 export { AgentEvalError, AgentEvalErrorCode, ConfigError, ControlBudget, ControlDecision, ControlEvalResult, ControlRunResult, ControlStep, DataAcquisitionPlan, JudgeError, KnowledgeReadinessReport, KnowledgeRequirement, NotFoundError, RunRecord, ValidationError } from '@tangle-network/agent-eval';
-import { h as AgentBackendInput, O as OpenAIChatTool, i as OpenAIChatToolChoice, j as OpenAIChatResponseFormat, k as AgentExecutionBackend, l as AgentBackendContext, a as RuntimeStreamEvent, K as KnowledgeReadinessDecision, m as RunAgentTaskOptions, n as AgentTaskRunResult, o as RunAgentTaskStreamOptions, p as RuntimeSessionStore, q as RuntimeSession, R as RuntimeHooks } from './types-Kjr5W6OF.js';
-export { r as AgentAdapter, s as AgentKnowledgeProvider, A as AgentRuntimeEvent, t as AgentRuntimeEventSink, u as AgentTaskContext, v as AgentTaskSpec, b as AgentTaskStatus, B as BackendErrorDetail, w as RuntimeDecisionEvidenceRef, x as RuntimeDecisionKind, y as RuntimeDecisionPoint, z as RuntimeHookContext, C as RuntimeHookErrorContext, F as RuntimeHookEvent, G as RuntimeHookPhase, H as RuntimeHookTarget, J as RuntimeRunHandle, M as RuntimeRunPersistenceAdapter, N as RuntimeRunRow, P as composeRuntimeHooks, Q as defineRuntimeHooks, T as notifyRuntimeDecisionPoint, U as notifyRuntimeHookEvent, W as startRuntimeRun } from './types-Kjr5W6OF.js';
+import { h as AgentBackendInput, O as OpenAIChatTool, i as OpenAIChatToolChoice, j as OpenAIChatResponseFormat, k as AgentExecutionBackend, l as AgentBackendContext, a as RuntimeStreamEvent, K as KnowledgeReadinessDecision, m as RunAgentTaskOptions, n as AgentTaskRunResult, o as RunAgentTaskStreamOptions, p as RuntimeSessionStore, q as RuntimeSession, R as RuntimeHooks } from './types-CKmyZ9TP.js';
+export { r as AgentAdapter, s as AgentKnowledgeProvider, A as AgentRuntimeEvent, t as AgentRuntimeEventSink, u as AgentTaskContext, v as AgentTaskSpec, b as AgentTaskStatus, B as BackendErrorDetail, w as RuntimeDecisionEvidenceRef, x as RuntimeDecisionKind, y as RuntimeDecisionPoint, z as RuntimeHookContext, C as RuntimeHookErrorContext, F as RuntimeHookEvent, G as RuntimeHookPhase, H as RuntimeHookTarget, J as RuntimeRunHandle, M as RuntimeRunPersistenceAdapter, N as RuntimeRunRow, P as composeRuntimeHooks, Q as defineRuntimeHooks, T as notifyRuntimeDecisionPoint, U as notifyRuntimeHookEvent, W as startRuntimeRun } from './types-CKmyZ9TP.js';
 export { AgentCandidateCodeSource, AgentCandidateCodeSurfaceSource, AgentCandidateModelGrantActivateInput, AgentCandidateModelGrantClient, AgentCandidateModelGrantReservation, AgentCandidateModelGrantReserveInput, AgentCandidateModelGrantSettleInput, AgentCandidateProfileSource, AgentCandidateWorkspaceArchiveLimits, BuildAgentCandidateBundleInput, CaptureAgentCandidateWorkspaceOptions, CapturedAgentCandidateWorkspace, CreateAgentCandidateWorkspacePortOptions, CreateProtectedAgentCandidateModelPortOptions, DisposePreparedAgentCandidateOptions, FileAgentCandidateExecutionClaimStore, FileAgentCandidateExecutionClaimStoreOptions, RecoverExpiredAgentCandidateOptions, applyExactAgentProfileDiff, buildAgentCandidateBundle, candidateExecutionClaim, captureAgentCandidateWorkspace, captureAgentCandidateWorkspaceFiles, createAgentCandidateWorkspacePort, createProtectedAgentCandidateModelPort, disposePreparedAgentCandidateExecution, parseExactAgentProfile, parseExactAgentProfileDiff, persistCandidateOutputArtifact, recoverExpiredAgentCandidateExecution, verifyAgentCandidateBundle } from './candidate-execution/index.js';
 export { d as AgentCandidateArtifactPort, e as AgentCandidateBenchmarkGraderIdentity, f as AgentCandidateBenchmarkGraderPort, A as AgentCandidateBundleInput, g as AgentCandidateContainerPort, h as AgentCandidateExecutionAttemptRecord, i as AgentCandidateExecutionAttemptRef, j as AgentCandidateExecutionClaim, k as AgentCandidateExecutionClaimResult, l as AgentCandidateExecutionClaimStore, m as AgentCandidateExecutionCleanupHandles, n as AgentCandidateExecutionFailureClass, o as AgentCandidateExecutionFinishResult, p as AgentCandidateExecutionLease, q as AgentCandidateExecutionPhase, r as AgentCandidateExecutionPhaseResult, b as AgentCandidateExecutionPorts, s as AgentCandidateExecutionRecoveryEvidence, t as AgentCandidateExecutionStageResult, u as AgentCandidateExecutionTerminalRecord, v as AgentCandidateExecutionTerminalResult, w as AgentCandidateExecutionUsage, x as AgentCandidateExecutorFinalCapture, y as AgentCandidateExecutorMemoryCapture, z as AgentCandidateExecutorPort, B as AgentCandidateExecutorProfileFile, C as AgentCandidateExecutorRequest, D as AgentCandidateExecutorStopRequest, F as AgentCandidateExecutorTaskOutcomeCapture, G as AgentCandidateExecutorWorkspaceFile, H as AgentCandidateExecutorWorkspaceInput, I as AgentCandidateMemoryPort, J as AgentCandidateMemoryResetResult, K as AgentCandidateModelLimits, L as AgentCandidateModelPort, M as AgentCandidateOutputArtifactPort, N as AgentCandidateOutputPurpose, O as AgentCandidateProtectedModelActivation, Q as AgentCandidateProtectedModelCall, R as AgentCandidateProtectedModelReservation, S as AgentCandidateProtectedModelSettlement, T as AgentCandidateProtectedRunCapture, U as AgentCandidateRepositoryPort, V as AgentCandidateRetryRejection, c as AgentCandidateRunFinalization, a as AgentCandidateTaskExecution, W as AgentCandidateVerificationPorts, X as AgentCandidateWorkspacePort, Y as CANDIDATE_TRACE_ENV, Z as CANDIDATE_TRACE_TAGS, _ as CanonicalCandidateDocument, E as ExecutePreparedAgentCandidateOptions, $ as InMemoryAgentCandidateExecutionClaimStore, P as PrepareAgentCandidateExecutionOptions, a0 as PreparedAgentCandidateExecution, a1 as PreparedAgentCandidateInstruction, a2 as PreparedAgentCandidateLaunch, a3 as PreparedAgentCandidateTrace, a4 as ResolvedAgentCandidateContainer, a5 as VerifiedAgentCandidate, a6 as VerifiedAgentCandidateTaskOutcome, a7 as executePreparedAgentCandidate, a8 as prepareAgentCandidateExecution, a9 as sealAgentCandidateBundle } from './prepare-DIeVgQG3.js';
 import { Scenario, ProfileDispatchFn, ProposeContext, SurfaceProposer } from '@tangle-network/agent-eval/campaign';
-import { C as CandidateGenerator } from './agentic-generator-CL1q0H91.js';
-export { A as AgenticGeneratorOptions, I as ImprovementDriverOptions, M as ManagedImprovementDriver, V as Verifier, a as VerifyResult, b as agenticGenerator, c as commandVerifier, i as improvementDriver } from './agentic-generator-CL1q0H91.js';
-export { c as ImproveCodeOptions, d as ImproveMemoryOptions, a as ImproveOptions, b as ImproveResult, e as ImproveSkillsOptions, I as ImproveSurface, f as applyImprovementWinnerToProfile, i as improve } from './improve-Ciuvu2jh.js';
-export { M as McpServeSpec, m as mcpServeVerifier } from './mcp-serve-verifier-D7hCZIkT.js';
+import { C as CandidateGenerator } from './agentic-generator-1g7GDba7.js';
+export { A as AGENTIC_PROFILE_RESOURCE_ROOT, a as AgenticGeneratorOptions, b as AgenticGeneratorShotReceipt, I as ImprovementDriverOptions, M as ManagedImprovementDriver, V as Verifier, c as VerifyResult, d as agenticGenerator, e as commandVerifier, i as improvementDriver } from './agentic-generator-1g7GDba7.js';
+export { c as ImproveCodeOptions, d as ImproveMemoryOptions, a as ImproveOptions, b as ImproveResult, e as ImproveSkillsOptions, I as ImproveSurface, f as applyImprovementWinnerToProfile, i as improve } from './improve-DZeatcs7.js';
+export { M as McpServeSpec, m as mcpServeVerifier } from './mcp-serve-verifier-BO4MEsR3.js';
 import { AgentProfileDiff, AgentProfile as AgentProfile$1 } from '@tangle-network/agent-interface';
 import { Scenario as Scenario$1, SelfImproveOptions } from '@tangle-network/agent-eval/contract';
 import { S as SurfaceImprovementEdit } from './improvement-adapter-BieWeK5J.js';
 import { I as ImprovementAdapter } from './types-BC3bZpH0.js';
 export { AgentKnowledgeReadinessCheckOptions, KnowledgeImprovementJobMeasurement, KnowledgeImprovementJobResult, KnowledgeReadinessCheck, KnowledgeReadinessCheckInput, KnowledgeReadinessCheckResult, RESEARCH_SUPERVISOR_SYSTEM_PROMPT, RunKnowledgeImprovementJobOptions, SupervisedKnowledgeUpdateInput, SupervisedKnowledgeUpdateOptions, SupervisedKnowledgeUpdateResult, SupervisedKnowledgeUpdater, createAgentKnowledgeReadinessCheck, createSupervisedKnowledgeUpdater, formatSupervisedKnowledgeTask, knowledgeReadinessDeliverable, runKnowledgeImprovementJob, runSupervisedKnowledgeUpdate } from './knowledge.js';
-export { D as DELEGATED_LOOP_MODES, a as DelegatedLoopMode, b as DelegatedLoopRegistry, c as DelegatedLoopResult, d as DelegatedLoopRunner, L as LoopRunnerCliArgs, e as LoopRunnerCliResult, R as ResearchLoopResult, f as ResearchLoopRunnerOptions, g as RunDelegatedLoopOptions, V as VetoedFact, W as WorktreeLoopRunnerOptions, h as auditLoopRunner, i as isDelegatedLoopMode, p as parseLoopRunnerArgv, r as researchLoopRunner, j as runDelegatedLoop, k as runLoopRunnerCli, s as selfImproveLoopRunner, w as worktreeLoopRunner } from './loop-runner-bin-B4PwjVaF.js';
-export { m as mcpToolsForRuntimeMcp, a as mcpToolsForRuntimeMcpSubset } from './openai-tools-BMOu1t3A.js';
-export { b0 as EvalRunEvent, b1 as EvalRunGeneration, b2 as EvalRunsExportConfig, b3 as EvalRunsExportResult, b4 as INTELLIGENCE_WIRE_VERSION, b5 as LoopSpanNode, b6 as OtelAttribute, b7 as OtelExportConfig, b8 as OtelExporter, b9 as OtelSpan, ba as RuntimeEventOtelOptions, bb as buildLoopOtelSpans, bc as buildLoopSpanNodes, bd as buildRuntimeEventOtelSpans, be as createOtelExporter, bf as exportEvalRuns, bg as loopEventToOtelSpan } from './coordination-pOGZuYS7.js';
-import { c as RuntimeTelemetryOptions } from './sanitize-Dr3SliRC.js';
-export { d as RuntimeEventCollector, e as RuntimeStreamEventCollector, S as SanitizedKnowledgeReadinessReport, f as createRuntimeEventCollector, g as createRuntimeStreamEventCollector, s as sanitizeAgentRuntimeEvent, h as sanitizeKnowledgeReadinessReport, i as sanitizeRuntimeStreamEvent } from './sanitize-Dr3SliRC.js';
+export { D as DELEGATED_LOOP_MODES, a as DelegatedLoopMode, b as DelegatedLoopRegistry, c as DelegatedLoopResult, d as DelegatedLoopRunner, L as LoopRunnerCliArgs, e as LoopRunnerCliResult, R as ResearchLoopResult, f as ResearchLoopRunnerOptions, g as RunDelegatedLoopOptions, V as VetoedFact, W as WorktreeLoopRunnerOptions, h as auditLoopRunner, i as isDelegatedLoopMode, p as parseLoopRunnerArgv, r as researchLoopRunner, j as runDelegatedLoop, k as runLoopRunnerCli, s as selfImproveLoopRunner, w as worktreeLoopRunner } from './loop-runner-bin-iTSxSmpv.js';
+export { m as mcpToolsForRuntimeMcp, a as mcpToolsForRuntimeMcpSubset } from './openai-tools-0wACz8f8.js';
+export { b0 as EvalRunEvent, b1 as EvalRunGeneration, b2 as EvalRunsExportConfig, b3 as EvalRunsExportResult, b4 as INTELLIGENCE_WIRE_VERSION, b5 as LoopSpanNode, b6 as OtelAttribute, b7 as OtelExportConfig, b8 as OtelExporter, b9 as OtelSpan, ba as RuntimeEventOtelOptions, bb as buildLoopOtelSpans, bc as buildLoopSpanNodes, bd as buildRuntimeEventOtelSpans, be as createOtelExporter, bf as exportEvalRuns, bg as loopEventToOtelSpan } from './coordination-o0TzS7Ms.js';
+import { c as RuntimeTelemetryOptions } from './sanitize-Dcbjp0tU.js';
+export { d as RuntimeEventCollector, e as RuntimeStreamEventCollector, S as SanitizedKnowledgeReadinessReport, f as createRuntimeEventCollector, g as createRuntimeStreamEventCollector, s as sanitizeAgentRuntimeEvent, h as sanitizeKnowledgeReadinessReport, i as sanitizeRuntimeStreamEvent } from './sanitize-Dcbjp0tU.js';
 import '@tangle-network/sandbox';
-import './local-harness-CePtFrwu.js';
+import './local-harness-ZLMkKdZc.js';
 import 'node:child_process';
 import '@tangle-network/agent-knowledge';
-import './supervise-Y3w878fN.js';
-import './types-DHLXdLVN.js';
-import './completion-gate-CfX2thvL.js';
+import './supervise-CRwaJIkd.js';
+import './types-CogNCaw7.js';
+import './completion-gate-tzwyyD-E.js';
 import './kb-gate-CwHO0vz6.js';
-import './worktree-fanout-DmBvR-_O.js';
+import './worktree-fanout-BJy24LYe.js';
 import './substrate-DO2GHNg2.js';
 import './environment-provider.js';
 import '@tangle-network/agent-interface/environment-provider';
@@ -917,6 +917,9 @@ interface RunPersonaConfig<TScenario extends Scenario, TArtifact> {
     maxTurns?: (scenario: TScenario) => number;
     seed?: (scenario: TScenario) => string;
     workerName?: string;
+    /** Provider- or executor-enforced maximum for the whole worker conversation.
+     * Required before execution when the enclosing campaign is cost-capped. */
+    maximumCharge?: MaximumCharge | ((worker: AgentProfile, scenario: TScenario) => MaximumCharge | undefined);
 }
 /**
  * Wrap {@link runPersonaConversation} as a `ProfileDispatchFn` for

package/dist/index.js

@@ -11,7 +11,7 @@ import {
 import {
   mcpToolsForRuntimeMcp,
   mcpToolsForRuntimeMcpSubset
-} from "./chunk-4VMRCEI6.js";
+} from "./chunk-MSUWXCHD.js";
 import {
   DEFAULT_ROUTER_BASE_URL,
   cleanModelId,
@@ -30,14 +30,14 @@ import {
   runLoopRunnerCli,
   selfImproveLoopRunner,
   worktreeLoopRunner
-} from "./chunk-63TIZHMI.js";
+} from "./chunk-KYPVUEJ4.js";
 import "./chunk-SGKPNBXE.js";
 import {
   applyImprovementWinnerToProfile,
   improve,
   improvementDriver,
   rawTraceDistiller
-} from "./chunk-JMT5VYY3.js";
+} from "./chunk-7NCP74SL.js";
 import {
   CANDIDATE_TRACE_ENV,
   CANDIDATE_TRACE_TAGS,
@@ -62,17 +62,18 @@ import {
   normalizeBackendStreamEvent,
   nowIso,
   touchSession
-} from "./chunk-GKZ6DFDN.js";
-import "./chunk-ZQZX77MM.js";
+} from "./chunk-7HH22XN4.js";
+import "./chunk-QDSOD7RC.js";
 import {
   mcpBuildPrompt,
   mcpServeVerifier,
   toolBuildPrompt
 } from "./chunk-6O73TRHW.js";
 import {
+  AGENTIC_PROFILE_RESOURCE_ROOT,
   agenticGenerator,
   commandVerifier
-} from "./chunk-4WWSICFY.js";
+} from "./chunk-GHXW56PO.js";
 import {
   RESEARCH_SUPERVISOR_SYSTEM_PROMPT,
   createAgentKnowledgeReadinessCheck,
@@ -81,14 +82,14 @@ import {
   knowledgeReadinessDeliverable,
   runKnowledgeImprovementJob,
   runSupervisedKnowledgeUpdate
-} from "./chunk-RNWDH5Z3.js";
+} from "./chunk-TNUBPPZA.js";
 import "./chunk-5QOB7H74.js";
 import {
   composeRuntimeHooks,
   defineRuntimeHooks,
   notifyRuntimeDecisionPoint,
   notifyRuntimeHookEvent
-} from "./chunk-LWNMCOEA.js";
+} from "./chunk-FRXDOBFP.js";
 import "./chunk-DPEUKJRO.js";
 import {
   INTELLIGENCE_WIRE_VERSION,
@@ -107,8 +108,8 @@ import {
 import "./chunk-WIPGQ4GT.js";
 import "./chunk-VSWBYWFK.js";
 import "./chunk-3MDZX7YU.js";
-import "./chunk-FVJ7M3DA.js";
-import "./chunk-WQA5Y6PD.js";
+import "./chunk-M22PL4GH.js";
+import "./chunk-EP6RVHMX.js";
 import {
   AgentEvalError,
   BackendTransportError,
@@ -1219,18 +1220,33 @@ async function runPersonaConversation(opts) {
 }
 function runPersonaDispatch(config) {
   return async (worker, scenario, ctx) => {
-    const result = await runPersonaConversation({
-      worker,
-      persona: config.personaOf(scenario),
-      backendFor: config.backendFor,
-      systemPromptOf: config.systemPromptOf,
-      maxTurns: config.maxTurns?.(scenario),
-      seed: config.seed?.(scenario),
+    const model = worker.model?.default ?? "unknown";
+    const maximumCharge = typeof config.maximumCharge === "function" ? config.maximumCharge(worker, scenario) : config.maximumCharge;
+    const paid = await ctx.cost.runPaidCall({
+      channel: "agent",
+      actor: "persona-conversation",
+      model,
       signal: ctx.signal,
-      workerName: config.workerName
+      ...maximumCharge ? { maximumCharge } : {},
+      execute: (executionSignal) => runPersonaConversation({
+        worker,
+        persona: config.personaOf(scenario),
+        backendFor: config.backendFor,
+        systemPromptOf: config.systemPromptOf,
+        maxTurns: config.maxTurns?.(scenario),
+        seed: config.seed?.(scenario),
+        signal: executionSignal,
+        workerName: config.workerName
+      }),
+      receipt: (result2) => ({
+        model,
+        inputTokens: result2.tokensIn,
+        outputTokens: result2.tokensOut,
+        ...result2.costUsd > 0 ? { actualCostUsd: result2.costUsd } : {}
+      })
     });
-    ctx.cost.observe(result.costUsd, "persona-conversation");
-    ctx.cost.observeTokens({ input: result.tokensIn, output: result.tokensOut });
+    if (!paid.succeeded) throw paid.error;
+    const result = paid.value;
     return config.artifactOf(result.transcript, scenario);
   };
 }
@@ -2478,6 +2494,7 @@ function randomSuffix2(len = 8) {
   return Math.random().toString(36).slice(2, 2 + len);
 }
 export {
+  AGENTIC_PROFILE_RESOURCE_ROOT,
   AgentEvalError,
   BackendTransportError,
   CANDIDATE_TRACE_ENV,