@evolvingmachines/evolve 0.0.67 → 0.0.69

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.
Files changed (64) hide show
  1. package/dist/chunk-FHL4GIAZ.js +1 -0
  2. package/dist/chunk-ORD7JDZB.js +1 -0
  3. package/dist/chunk-SQDR2TAW.js +427 -0
  4. package/dist/cli/index.cjs +38 -38
  5. package/dist/cli/index.d.cts +1 -1
  6. package/dist/cli/index.d.ts +1 -1
  7. package/dist/cli/index.js +30 -30
  8. package/dist/index.cjs +65 -60
  9. package/dist/index.d.cts +332 -29
  10. package/dist/index.d.ts +332 -29
  11. package/dist/index.js +48 -43
  12. package/dist/managed-modal-IXZBT4D5.js +4 -0
  13. package/dist/resumable-XNW5SKRW.js +1 -0
  14. package/dist/{types-84M4J_Vm.d.cts → types-DGx_qo9M.d.cts} +118 -77
  15. package/dist/{types-84M4J_Vm.d.ts → types-DGx_qo9M.d.ts} +118 -77
  16. package/harness-capabilities.json +325 -77
  17. package/hosted-error-codes.json +1 -0
  18. package/package.json +9 -7
  19. package/skills/create-adapter/SKILL.md +1 -1
  20. package/skills/create-task/SKILL.md +1 -1
  21. package/skills/evolve/SKILL.md +1 -1
  22. package/skills/evolve-evals/SKILL.md +1 -1
  23. package/skills/evolve-evals/references/cli-reference/agent.md +1 -1
  24. package/skills/evolve-evals/references/cli-reference/analyze.md +7 -4
  25. package/skills/evolve-evals/references/cli-reference/check.md +5 -4
  26. package/skills/evolve-evals/references/cli-reference/index.md +1 -1
  27. package/skills/evolve-evals/references/cli-reference/run.md +6 -5
  28. package/skills/evolve-evals/references/core-concepts/agents.md +19 -5
  29. package/skills/evolve-evals/references/core-concepts/analyze.md +39 -5
  30. package/skills/evolve-evals/references/core-concepts/check.md +62 -6
  31. package/skills/evolve-evals/references/core-concepts/datasets.md +1 -1
  32. package/skills/evolve-evals/references/core-concepts/index.md +1 -1
  33. package/skills/evolve-evals/references/core-concepts/jobs.md +3 -3
  34. package/skills/evolve-evals/references/core-concepts/models.md +393 -41
  35. package/skills/evolve-evals/references/core-concepts/sandboxes.md +1 -1
  36. package/skills/evolve-evals/references/core-concepts/secrets.md +1 -1
  37. package/skills/evolve-evals/references/core-concepts/skills.md +1 -1
  38. package/skills/evolve-evals/references/core-concepts/task-verifiers.md +1 -1
  39. package/skills/evolve-evals/references/core-concepts/teams.md +1 -1
  40. package/skills/evolve-evals/references/core-concepts/upload.md +1 -1
  41. package/skills/evolve-evals/references/dashboard/checks.md +5 -3
  42. package/skills/evolve-evals/references/dashboard/jobs.md +1 -1
  43. package/skills/evolve-evals/references/getting-started/first-task.md +1 -1
  44. package/skills/evolve-evals/references/getting-started/quick-start.md +1 -1
  45. package/skills/evolve-evals/references/index.md +1 -1
  46. package/skills/evolve-evals/references/sdk/http.md +1 -1
  47. package/skills/evolve-evals/references/sdk/python.md +1 -1
  48. package/skills/evolve-evals/references/sdk/typescript.md +1 -1
  49. package/skills/evolve-evals/references/sdk-reference/analyses.md +16 -5
  50. package/skills/evolve-evals/references/sdk-reference/checks.md +14 -5
  51. package/skills/evolve-evals/references/sdk-reference/error-codes.md +3 -2
  52. package/skills/evolve-evals/references/sdk-reference/jobs.md +3 -3
  53. package/skills/evolve-evals/references/sdk-reference/meta.md +4 -3
  54. package/skills/evolve-evals/references/sdk-reference/methods/analyses.md +6 -4
  55. package/skills/evolve-evals/references/sdk-reference/methods/checks.md +8 -4
  56. package/skills/evolve-evals/references/sdk-reference/methods/jobs.md +5 -3
  57. package/skills/evolve-evals/references/sdk-reference/methods/meta.md +15 -4
  58. package/skills/evolve-evals/references/sdk-reference/skills.md +2 -2
  59. package/skills/evolve-evals/references/sdk-reference/types.md +11 -5
  60. package/spec/openapi.yaml +290 -137
  61. package/dist/chunk-HQJI3HHU.js +0 -1
  62. package/dist/chunk-RFLMRPGK.js +0 -427
  63. package/dist/managed-modal-IJW62SSK.js +0 -4
  64. package/dist/resumable-HFSICSH3.js +0 -1
@@ -0,0 +1,4 @@
1
+ import {a}from'./chunk-ORD7JDZB.js';var p=1024*1024,h=class extends Error{bytes;constructor(e,t){super(`Managed Modal ${e} body is ${t} bytes; the managed door caps every request at ${p} bytes (1 MiB) of wire bytes, base64 inflation included. Split the payload into smaller writes.`),this.name="ManagedModalWriteLimitError",this.bytes=t;}},u=class extends Error{status;retryAfterSec;constructor(e,t,r,n){super(`Managed Modal ${e} failed (${t})${r?`: ${r}`:""}`),this.name="ManagedModalDoorError",this.status=t,n!==void 0&&(this.retryAfterSec=n);}};function C(a){if(typeof a=="string")return new TextEncoder().encode(a);if(a instanceof Uint8Array)return new Uint8Array(a);if(a instanceof ArrayBuffer)return new Uint8Array(a);throw new Error(`Unsupported data type: ${typeof a}`)}var f=class{constructor(e){this.config=e;}async request(e,t,r){let n=await fetch(`${this.config.baseUrl}${t}`,{...r,headers:{authorization:`Bearer ${this.config.apiKey}`,...r.headers??{}}});if(!n.ok){let o=await n.text().catch(()=>""),i="";try{let c=JSON.parse(o);typeof c?.error=="string"&&(i=c.error);}catch{}throw new u(e,n.status,i,a(o,n))}return n}json(e){return {method:"POST",headers:{"content-type":"application/json"},body:JSON.stringify(e)}}},b=class{constructor(e,t){this.door=e;this.sandboxId=t;}async run(e,t){let r=await this.door.request("exec",`/sandboxes/${encodeURIComponent(this.sandboxId)}/exec`,this.door.json({command:e,...t?.cwd?{cwd:t.cwd}:{},...t?.envs?{envs:t.envs}:{},...t?.timeoutMs!==void 0?{timeoutMs:t.timeoutMs}:{}})),n="",o="",i,c=d=>{if(!d.trim())return;let s;try{s=JSON.parse(d);}catch{throw new Error(`Managed Modal exec stream carried a garbled record: ${d}`)}s.stream==="stdout"&&typeof s.data=="string"?(n+=s.data,t?.onStdout?.(s.data)):s.stream==="stderr"&&typeof s.data=="string"?(o+=s.data,t?.onStderr?.(s.data)):i=s;};if(r.body){let d=r.body.getReader(),s=new TextDecoder,l="";for(;;){let{done:S,value:P}=await d.read();if(S)break;l+=s.decode(P,{stream:true});let g;for(;(g=l.indexOf(`
2
+ `))!==-1;)c(l.slice(0,g)),l=l.slice(g+1);}l+=s.decode(),c(l);}else for(let d of (await r.text()).split(`
3
+ `))c(d);if(i?.error!==void 0)throw new Error(`Managed Modal exec failed: ${i.error}`);if(typeof i?.exitCode!="number")throw new Error("Managed Modal exec stream ended without a terminal record \u2014 the connection died mid-run.");return {exitCode:i.exitCode,stdout:n,stderr:o}}async spawn(e,t){let r=this.run(e,t);return r.catch(()=>{}),{processId:`managed-modal-${Date.now()}-${Math.random().toString(36).slice(2)}`,wait:()=>r,kill:async()=>false}}async list(){return (await this.run("ps -eo pid,comm,args")).stdout.trim().split(`
4
+ `).slice(1).filter(r=>r.trim()).map(r=>{let n=r.trim().split(/\s+/);return {processId:n[0],cmd:n[1]||"",args:n.slice(2),envs:{}}})}async kill(e){return /^\d+$/.test(e)?(await this.run(`kill -9 ${e}`)).exitCode===0:false}},y=class{constructor(e,t){this.door=e;this.sandboxId=t;}filesPath(e){return `/sandboxes/${encodeURIComponent(this.sandboxId)}/files/${e}`}encodeContent(e){return typeof e=="string"?{content:e}:{content:Buffer.from(C(e)).toString("base64"),encoding:"base64"}}boundedJson(e,t){let r=this.door.json(t),n=Buffer.byteLength(r.body,"utf8");if(n>p)throw new h(e,n);return r}async read(e){let r=await(await this.door.request("file read",this.filesPath("read"),this.door.json({path:e}))).json();return r.encoding==="base64"?new Uint8Array(Buffer.from(r.content??"","base64")):r.content??""}async write(e,t){await this.door.request("file write",this.filesPath("write"),this.boundedJson("file write",{path:e,...this.encodeContent(t)}));}async writeBatch(e){e.length!==0&&await this.door.request("file writeBatch",this.filesPath("writeBatch"),this.boundedJson("file writeBatch",{files:e.map(t=>({path:t.path,...this.encodeContent(t.data)}))}));}async makeDir(e){await this.door.request("makeDir",this.filesPath("makeDir"),this.door.json({path:e}));}};function M(a){return {sandboxId:a.sandboxId,image:a.image??"",metadata:a.metadata??{},startedAt:a.startedAt??""}}var m=class{constructor(e,t){this.door=e;this.sandboxId=t;this.commands=new b(e,t),this.files=new y(e,t);}commands;files;async getHost(e){throw new Error("Managed Modal sandboxes expose no tunnels, so getHost() is unavailable through the managed door.")}async kill(){try{await this.door.request("kill",`/sandboxes/${encodeURIComponent(this.sandboxId)}`,{method:"DELETE"});}catch(e){if(e instanceof u&&e.status===404)return;throw e}}async pause(){throw new Error("Modal does not support pause/resume. Persist progress with Evolve checkpoints and resume in a fresh sandbox, or use kill() to terminate.")}async isRunning(){try{return await this.getInfo(),!0}catch(e){if(e instanceof u&&e.status===404)return false;throw e}}async getInfo(){let e=await this.door.request("get",`/sandboxes/${encodeURIComponent(this.sandboxId)}`,{method:"GET"});return M(await e.json())}},w=class{providerType="modal";name="Managed Modal";door;retryCreateRequest;constructor(e){this.door=new f(e),this.retryCreateRequest=e.retryCreateRequest;}async create(e){if(e.resources!==void 0)throw new Error("Managed Modal sandboxes are platform-sized; `resources` cannot be enforced through the managed door.");if(e.network!==void 0)throw new Error("Managed Modal sandboxes run the platform's network policy; `network` cannot be enforced through the managed door.");if(e.user!==void 0)throw new Error("Managed Modal sandboxes run as the platform's sandbox user; `user` cannot be enforced through the managed door.");if(e.idleTimeoutMs!==void 0)throw new Error("The managed Modal door takes no idle timeout; bound the sandbox with `timeoutMs` instead.");if(e.bootCommand!==void 0)throw new Error("Managed Modal sandboxes boot the platform image as configured; `bootCommand` cannot be enforced through the managed door.");let t=this.door.json({...e.image?{image:e.image}:{},...e.timeoutMs!==void 0?{timeoutMs:e.timeoutMs}:{},...e.workingDirectory?{workingDirectory:e.workingDirectory}:{},...e.envs?{envs:e.envs}:{},...e.metadata?{metadata:e.metadata}:{}}),r=()=>this.door.request("create","/sandboxes",t),o=await(await(this.retryCreateRequest?this.retryCreateRequest(r):r())).json();return new m(this.door,o.sandboxId)}async connect(e,t){let r=new m(this.door,e);return await r.getInfo(),r}async list(e){if(e?.state&&!e.state.includes("running"))return [];let n=((await(await this.door.request("list","/sandboxes",{method:"GET"})).json()).sandboxes??[]).map(M);if(e?.metadata){let o=Object.entries(e.metadata);n=n.filter(i=>o.every(([c,d])=>i.metadata[c]===d));}return e?.limit!==void 0&&(n=n.slice(0,e.limit)),n}async listAll(e){if(e?.state&&!e.state.includes("running"))return {sandboxes:[],complete:true,pagesFetched:0};let t;try{t=await this.list({...e,limit:void 0});}catch(o){return {sandboxes:[],complete:false,pagesFetched:0,error:`sandbox list failed: ${o instanceof Error?o.message:String(o)}`}}let r=e?.limit!==void 0?t.slice(0,e.limit):t,n=r.length<t.length;return {sandboxes:r,complete:!n,pagesFetched:1,...n?{error:`stopped at the requested limit of ${e?.limit} with more sandboxes available`}:{}}}},v=(a,e)=>new m(new f(a),e);export{u as ManagedModalDoorError,w as ManagedModalProvider,h as ManagedModalWriteLimitError,v as _testManagedModalSandbox};
@@ -0,0 +1 @@
1
+ export{c as RESUMABLE_UPLOAD_CHUNK_BYTES,d as RESUMABLE_UPLOAD_MAX_ATTEMPTS,e as RESUMABLE_UPLOAD_MAX_RATE_LIMIT_WAITS,f as RESUMABLE_UPLOAD_MAX_RETRY_AFTER_SEC,a as RESUMABLE_UPLOAD_THRESHOLD_BYTES,g as retryAfterWaitMs,b as setResumableUploadThresholdBytes,h as uploadArchiveResumable}from'./chunk-FHL4GIAZ.js';import'./chunk-ORD7JDZB.js';
@@ -450,8 +450,8 @@ interface Rubric {
450
450
  * openrouter/deepseek/deepseek-v4.1-flash at its per-model effort (high)
451
451
  * over the platform's default analyze rubric and its default prompt body.
452
452
  *
453
- * The analyzer always runs the claude-code harness (Harbor's default analyze
454
- * agent) in its own sealed sandbox — on the provider `sandbox_provider`
453
+ * The analyzer runs on the `agent` harness (default claude, Harbor's claude-code)
454
+ * in its own sealed sandbox — on the provider `sandbox_provider`
455
455
  * names, or the platform's analysis default when it names none; its spend is
456
456
  * capped per analysis and metered as its own line, never blended into the
457
457
  * trial's own bill.
@@ -460,31 +460,36 @@ interface Rubric {
460
460
  * exact names — `n_concurrent` (`-n/--n-concurrent`), `passing` / `failing`,
461
461
  * `n_trials` (`-l/--n-trials`; their cli/analyze.py:278-290). All omitted
462
462
  * is every analyzable trial, as wide as the organization's
463
- * `max_concurrent_analyses` allows. Harbor's `-a/--agent`, `--job-name`,
463
+ * `max_concurrent_analyses` allows. Harbor's `--job-name`,
464
464
  * `-o/--jobs-dir`, `-k/--n-attempts` and the local-runner kwargs are not
465
465
  * on this surface; the contract (`AnalyzeConfigInput` in spec/openapi.yaml)
466
466
  * records each with its reason.
467
467
  */
468
468
  interface AnalyzeConfigInput {
469
- /**
470
- * Model the analyzer agent runs — Harbor's `--model`. The default is
471
- * openrouter/deepseek/deepseek-v4.1-flash on this platform's claude
472
- * roster (DeepSeek V4.1 Flash served through OpenRouter, at its default
473
- * effort high — the owner's ruling 2026-09-10: far more parallel capacity
469
+ /** Harbor's `-a/--agent`, spelled as the arms spell it (claude, not claude-code) so one set of names covers arms and reviewers; omitted: claude. The agents: `GET /api/meta` `analyze.agents`; a retired one (`retired_agents`) is refused `agent_retired`. */
470
+ agent?: string;
471
+ /**
472
+ * Model the analyzer agent runs — Harbor's `--model`. Omitted, the agent's
473
+ * default: the platform's pick, openrouter/deepseek/deepseek-v4.1-flash
474
+ * (`GET /api/meta` `analyze.default_model`), on every agent whose roster
475
+ * carries it, else that agent's own default (`analyze.agents[].default_model`,
476
+ * the model the SDK runs that agent on when none is named). The platform's
477
+ * pick is DeepSeek V4.1 Flash served through OpenRouter, at its default
478
+ * effort high (the owner's ruling 2026-09-10: far more parallel capacity
474
479
  * through OpenRouter's provider pool than one pinned Fireworks host) — a
475
480
  * recorded deviation from Harbor's default analyze model (their
476
481
  * cli/analyze.py `claude-haiku-4-5`): analysis is input-dominated, and
477
482
  * this is the roster's intelligence-per-input-dollar pick; `glm-5.3-flash`
478
483
  * (at max, the effort its published scores use) and `haiku` stay on the
479
- * roster as alternatives, `glm-5.3` to escalate, and the same model on
480
- * its Fireworks route, `fireworks/deepseek-v4.1-flash`, is a further
481
- * option (the OpenRouter id stays the default). The value speaks
482
- * the same vocabulary as `agents[].model_name`: either advertised
483
- * spelling is accepted and stored AS GIVEN (the default is the roster
484
- * alias), the wire id is resolved only when the analyzer runs, and every
485
- * stored analysis serves the spelling it was created under. Must be on
486
- * the claude roster (`GET /api/meta`, `agents[].models`); anything else
487
- * is refused at accept (`invalid_input`, roster in the message).
484
+ * claude roster as alternatives, `glm-5.3` to escalate, and the same model
485
+ * on its Fireworks route, `fireworks/deepseek-v4.1-flash`, is a further
486
+ * option (the OpenRouter id stays the pick). The value speaks the same
487
+ * vocabulary as `agents[].model_name`: either advertised spelling is
488
+ * accepted and stored AS GIVEN (the default is the roster alias), the wire
489
+ * id is resolved only when the analyzer runs, and every stored analysis
490
+ * serves the spelling it was created under. A named model must be on the
491
+ * agent's roster (`GET /api/meta`, `analyze.agents[].models`); anything
492
+ * else is refused at accept (`invalid_input`, roster in the message).
488
493
  */
489
494
  model_name?: string;
490
495
  rubric?: Rubric;
@@ -511,15 +516,14 @@ interface AnalyzeConfigInput {
511
516
  prompt?: string;
512
517
  /**
513
518
  * Reasoning effort the analyzer runs at — the platform's `agents[].
514
- * reasoning_effort` vocabulary applied to the analyzer, which IS the
515
- * claude harness: the accepted values are `GET /api/meta`'s
516
- * `analyze.reasoning_efforts`, an unknown value is refused
519
+ * reasoning_effort` vocabulary applied to the analyzer's agent: the accepted
520
+ * values are `GET /api/meta`'s `analyze.agents[].reasoning_efforts`, an unknown value is refused
517
521
  * `invalid_input` exactly as an arm's is. Omitted, the PER-MODEL default
518
- * applies (`analyze.models[].default_reasoning_effort`: high on
522
+ * applies (`analyze.agents[].models[].default_reasoning_effort`: high on
519
523
  * openrouter/deepseek/deepseek-v4.1-flash, the default model — DeepSeek's
520
524
  * own documented default, the owner's ruling 2026-09-10; max on
521
525
  * glm-5.3-flash — the platform's ruling 2026-09-08, the effort its
522
- * published scores use; the claude harness default elsewhere). The
526
+ * published scores use; the agent's own default elsewhere, none where it takes none). The
523
527
  * effort is always passed to the analyzer explicitly and
524
528
  * recorded on the analysis (`TrialAnalysis.reasoning_effort`). A hosted
525
529
  * extension: Harbor's analyze has no effort option; this is the run
@@ -596,10 +600,12 @@ interface AnalyzeConfigInput {
596
600
  * the day, resolved at accept and stored, so the record always states the
597
601
  * policy it executes (same law as RetryConfig). Echoed on the job body when
598
602
  * the job was created with `analyze`; each analysis additionally carries the
599
- * exact policy IT ran under (`Trial.analysis.model_name` / `.rubric` /
600
- * `.prompt`), which a later manual re-analysis may have changed.
603
+ * exact policy IT ran under (`Trial.analysis.agent` / `.model_name` /
604
+ * `.rubric` / `.prompt`), which a later manual re-analysis may have changed.
601
605
  */
602
606
  interface AnalyzeConfig {
607
+ /** The agent this policy's analyses run on — as named, or the default (claude). */
608
+ agent: string;
603
609
  model_name: string;
604
610
  rubric: Rubric;
605
611
  /** The caller's prompt template as stored; null = the platform's default analyze body. */
@@ -607,10 +613,10 @@ interface AnalyzeConfig {
607
613
  /**
608
614
  * The effort this policy's analyses run at. Named at create it is served
609
615
  * as stored; when the create named none, this echoes the per-model
610
- * default of the day for `model_name` — the value the next enqueue under
611
- * this policy stamps (the same nuance as `sandbox_provider` below).
616
+ * default of the day for `model_name` on `agent` — the value the next enqueue under
617
+ * this policy stamps (the same nuance as `sandbox_provider` below). Null when the agent takes no effort.
612
618
  */
613
- reasoning_effort: string;
619
+ reasoning_effort: string | null;
614
620
  /**
615
621
  * The provider this policy's analyses run on. Named at create it is served
616
622
  * as stored, forever. When the create named none, this echoes the
@@ -1485,9 +1491,12 @@ interface AnalysisFailure {
1485
1491
  * `timeout` (the analyzer's run budget ran out with no valid analysis.json
1486
1492
  * — the file missing, or a partial one that failed validation, its reasons
1487
1493
  * in the message — never re-run: a timeout is deterministic; the message
1488
- * names the budget, the seconds used and the exit code), or an
1489
- * infrastructure stage of the analyzer run (`mint_key`, `boot`,
1490
- * `harness_install`, `agent`, `artifact_read`, `lease_expired`, ...).
1494
+ * names the budget, the seconds used and the exit code), `agent_retired`
1495
+ * (the job's stored analysis agent was retired after the job was created:
1496
+ * recorded when the trial settles, nothing run or charged, never re-run; the
1497
+ * message names the replacement), or an infrastructure stage of the analyzer
1498
+ * run (`mint_key`, `boot`, `harness_install`, `agent`, `artifact_read`,
1499
+ * `lease_expired`, ...).
1491
1500
  */
1492
1501
  phase: string;
1493
1502
  message: string;
@@ -1523,11 +1532,13 @@ interface TrialAnalysis {
1523
1532
  * `failed`, never left `running` forever.
1524
1533
  */
1525
1534
  status: AnalysisStatus;
1535
+ /** The agent THIS analysis ran on; every analysis from before the choice existed ran on claude. */
1536
+ agent: string;
1526
1537
  model_name: string;
1527
1538
  /**
1528
1539
  * The reasoning effort THIS analysis ran at — passed to the analyzer
1529
- * explicitly, so it is what the model was asked for. Null only on
1530
- * analyses recorded before the effort was stamped.
1540
+ * explicitly, so it is what the model was asked for. Null when its agent
1541
+ * takes no effort, or on analyses recorded before the effort was stamped.
1531
1542
  */
1532
1543
  reasoning_effort: string | null;
1533
1544
  rubric: Rubric;
@@ -2268,10 +2279,11 @@ interface CheckRow {
2268
2279
  name: string;
2269
2280
  status: CheckStatus;
2270
2281
  source: CheckSource;
2282
+ agent: string;
2271
2283
  /** The checker's model. */
2272
2284
  model_name: string;
2273
- /** The effort every task's checker ran at. */
2274
- reasoning_effort: string;
2285
+ /** The effort every task's checker ran at; null when its agent takes none. */
2286
+ reasoning_effort: string | null;
2275
2287
  sandbox_provider: EvalSandboxProvider;
2276
2288
  /** The owning organization's slug (every check has one). */
2277
2289
  org: string;
@@ -4548,8 +4560,10 @@ interface AnalysesClient {
4548
4560
  * feed's species-blind events door, which answers `trial_not_found`.
4549
4561
  */
4550
4562
  list(options?: ListAnalysesOptions): AnalysisList;
4551
- /** The defaults an analysis runs under when its config names nothing (GET /api/analyses/defaults): model, effort, provider, rubric and the unrendered prompt template. */
4552
- defaults(): Promise<AnalyzeDefaults>;
4563
+ /** The defaults an analysis runs under when its config names nothing (GET /api/analyses/defaults): agent, model, effort, provider, rubric and the unrendered prompt template. `{ agent }` reads what that agent runs under — its default model and the effort it takes; an unknown name is refused `invalid_input`. */
4564
+ defaults(options?: {
4565
+ agent?: string;
4566
+ }): Promise<AnalyzeDefaults>;
4553
4567
  /**
4554
4568
  * The result document — the wire's TrialAnalysis, statuses and typed
4555
4569
  * failure included, for EVERY analysis (not only completed ones). The same
@@ -4583,10 +4597,10 @@ interface AnalysesClient {
4583
4597
  * one directory named as Harbor names the wrapper trial
4584
4598
  * (`analyze-<analyzed trial dir>__<7 chars>/`): config.json, lock.json,
4585
4599
  * result.json, trial.log, exception.txt (an infrastructure failure only),
4586
- * agent/claude-code.txt (Harbor's tee name for claude-code),
4600
+ * agent/<Harbor's tee name for its agent> (claude-code.txt, codex.txt, …),
4587
4601
  * agent/stderr.log, agent/trace-parsed.jsonl, the captured home at its
4588
- * real names with agent/agent-home.json beside it and Harbor's copy at
4589
- * agent/sessions/, verifier/{test-stdout.txt,reward.txt,reward.json}
4602
+ * real names with agent/agent-home.json beside it and Harbor's copy at its
4603
+ * per-agent slot (agent/sessions/ for claude and codex), verifier/{test-stdout.txt,reward.txt,reward.json}
4590
4604
  * when the validator ruled (reward 1 = a valid analysis.json, 0 = it was
4591
4605
  * refused), and artifacts/manifest.json with artifacts/analysis.json (the
4592
4606
  * validated {summary, checks}) on a completed run — absent artifacts are
@@ -4609,8 +4623,8 @@ interface AnalysesClient {
4609
4623
  /**
4610
4624
  * Task-check configuration — Harbor's `harbor check` vocabulary (their
4611
4625
  * cli/analyze.py:84-148 check_command), the spec's `CheckConfigInput`. The
4612
- * rubric-agent trio is the analyze door's, under the same rules
4613
- * (`AnalyzeConfigInput` states them; refusals name `check.*`): `model_name`
4626
+ * rubric-agent knobs are the analyze door's, under the same rules
4627
+ * (`AnalyzeConfigInput` states them; refusals name `check.*`): `agent` (`-a/--agent`), `model_name`
4614
4628
  * (Harbor's check default is `claude-sonnet-4-6`; this platform's is the
4615
4629
  * analyzer's `openrouter/deepseek/deepseek-v4.1-flash` — one roster, one
4616
4630
  * default for both rubric
@@ -4628,14 +4642,15 @@ interface AnalysesClient {
4628
4642
  * (`-x/--exclude-task-name`), `n_tasks` (`-l/--n-tasks`) — applied in
4629
4643
  * checker.py's order (:132-138: include, exclude, then the cap) over the
4630
4644
  * sorted task directory names, Python fnmatch globs against the NAME.
4631
- * Harbor's `-a/--agent`, `--job-name`, `-o/--jobs-dir`, `-k/--n-attempts`
4632
- * and the local-runner kwargs are not on this surface; the contract records
4633
- * each with its reason.
4645
+ * Harbor's `-o/--jobs-dir`, `-k/--n-attempts` and the local-runner kwargs
4646
+ * are not on this surface; the contract records each with its reason.
4634
4647
  */
4635
4648
  interface CheckConfigInput {
4636
4649
  /** A name for the check (Harbor's `--job-name`); omitted, the accept timestamp `YYYY-MM-DD__HH-MM-SS`. 1-120 characters. */
4637
4650
  name?: string;
4638
- /** Model the checker agent runs (Harbor's `-m/--model`); must be on the claude roster (`GET /api/meta`). */
4651
+ /** The agent the checker runs on (Harbor's `-a/--agent`) — the analyze door's `agent`, same rule: a retired one is refused `agent_retired`. Omitted: claude. */
4652
+ agent?: string;
4653
+ /** Model the checker agent runs (Harbor's `-m/--model`) — the analyze door's `model_name`, same rule: omitted, the agent's default (`GET /api/meta` `analyze.agents[].default_model`); named, on the agent's roster. */
4639
4654
  model_name?: string;
4640
4655
  /** The rubric (Harbor's `-r/--rubric` file as its `{criteria}` object); default: the platform's check rubric (`checks().defaults()`). */
4641
4656
  rubric?: Rubric;
@@ -4740,9 +4755,11 @@ interface Check {
4740
4755
  name: string;
4741
4756
  status: CheckStatus;
4742
4757
  source: CheckSource;
4758
+ /** The agent every task's checker ran on — named at create, or the default (claude). */
4759
+ agent: string;
4743
4760
  model_name: string;
4744
- /** The effort every task's checker ran at — named at create, or the model's default, resolved at accept. */
4745
- reasoning_effort: string;
4761
+ /** The effort every task's checker ran at — named at create, or the model's default on the agent, resolved at accept; null when the agent takes none. */
4762
+ reasoning_effort: string | null;
4746
4763
  rubric: Rubric;
4747
4764
  /** The prompt template as stored; null = the platform's default check body. */
4748
4765
  prompt: string | null;
@@ -4770,33 +4787,38 @@ interface Check {
4770
4787
  finished_at: string | null;
4771
4788
  }
4772
4789
  /**
4773
- * The policy an empty analyze config resolves to (GET /api/analyses/defaults):
4774
- * each key the value `AnalyzeConfig` echoes for a job created with
4775
- * `analyze: {}`, except `prompt`, which `AnalyzeConfig` serves as null and
4776
- * this serves as the template text.
4790
+ * The policy an empty analyze config resolves to (GET /api/analyses/defaults),
4791
+ * or one naming only `agent` when `defaults({ agent })` names it: each key the
4792
+ * value `AnalyzeConfig` echoes for a job created with that config, except
4793
+ * `prompt`, which `AnalyzeConfig` serves as null and this serves as the
4794
+ * template text.
4777
4795
  */
4778
4796
  interface AnalyzeDefaults {
4797
+ agent: string;
4798
+ /** The model an omitted `model_name` takes on `agent` (`GET /api/meta` `analyze.agents[].default_model`). */
4779
4799
  model_name: string;
4780
4800
  rubric: Rubric;
4781
4801
  /** The built-in analyze prompt template, unrendered — pass it as `prompt` to run the default body explicitly, or edit it from here. */
4782
4802
  prompt: string;
4783
- /** The effort the default model runs at when the config names none. */
4784
- reasoning_effort: string;
4803
+ /** The effort `model_name` runs at on `agent` when the config names none; null if that agent takes none. */
4804
+ reasoning_effort: string | null;
4785
4805
  sandbox_provider: EvalSandboxProvider;
4786
4806
  }
4787
4807
  /**
4788
- * The policy an empty check config resolves to (GET /api/checks/defaults):
4789
- * each key the value `Check` echoes for a check created with no config,
4790
- * except `prompt`, which `Check` serves as null and this serves as the
4791
- * template text.
4808
+ * The policy an empty check config resolves to (GET /api/checks/defaults), or
4809
+ * one naming only `agent` when `defaults({ agent })` names it: each key the
4810
+ * value `Check` echoes for a check created with that config, except `prompt`,
4811
+ * which `Check` serves as null and this serves as the template text.
4792
4812
  */
4793
4813
  interface CheckDefaults {
4814
+ agent: string;
4815
+ /** The model an omitted `model_name` takes on `agent` (`GET /api/meta` `analyze.agents[].default_model`). */
4794
4816
  model_name: string;
4795
4817
  rubric: Rubric;
4796
4818
  /** The built-in check prompt template, unrendered — pass it as `prompt` to run the default body explicitly, or edit it from here. */
4797
4819
  prompt: string;
4798
- /** The effort the default model runs at when the config names none. */
4799
- reasoning_effort: string;
4820
+ /** The effort `model_name` runs at on `agent` when the config names none; null if that agent takes none. */
4821
+ reasoning_effort: string | null;
4800
4822
  sandbox_provider: EvalSandboxProvider;
4801
4823
  }
4802
4824
  /** Options for checks().watch() */
@@ -4867,8 +4889,10 @@ interface ChecksClient {
4867
4889
  get(checkId: string): Promise<Check>;
4868
4890
  /** Every check you may read, newest first (cursor-paged); `{ scope, status, dataset }` narrow it. */
4869
4891
  list(options?: ListChecksOptions): CheckList;
4870
- /** The defaults a check runs under when its config names nothing (GET /api/checks/defaults): model, effort, provider, rubric and the unrendered prompt template. */
4871
- defaults(): Promise<CheckDefaults>;
4892
+ /** The defaults a check runs under when its config names nothing (GET /api/checks/defaults): agent, model, effort, provider, rubric and the unrendered prompt template. `{ agent }` reads what that agent runs under — its default model and the effort it takes; an unknown name is refused `invalid_input`. */
4893
+ defaults(options?: {
4894
+ agent?: string;
4895
+ }): Promise<CheckDefaults>;
4872
4896
  /** Poll a check until every task settled; resolves with the final Check. */
4873
4897
  watch(checkId: string, options?: WatchCheckOptions): Promise<Check>;
4874
4898
  /**
@@ -4901,9 +4925,9 @@ interface ChecksClient {
4901
4925
  * the TASK CHECK id (`Check.results[].id`). Each folder is Harbor's
4902
4926
  * TrialPaths for the checker's run: config.json, lock.json, result.json,
4903
4927
  * trial.log, exception.txt (an infrastructure failure only),
4904
- * agent/claude-code.txt, agent/stderr.log, agent/trace-parsed.jsonl, the
4905
- * captured home at its real names with agent/agent-home.json and Harbor's
4906
- * copy at agent/sessions/, verifier/{test-stdout.txt,reward.txt,
4928
+ * agent/<its agent's tee name>, agent/stderr.log, agent/trace-parsed.jsonl, the
4929
+ * captured home at its real names with agent/agent-home.json and Harbor's copy at its
4930
+ * per-agent slot (agent/sessions/ for claude and codex), verifier/{test-stdout.txt,reward.txt,
4907
4931
  * reward.json} when the validator ruled (reward 1 = a valid
4908
4932
  * check-result.json, 0 = it was refused), artifacts/manifest.json and
4909
4933
  * artifacts/check-result.json (the validated flat checks) on a completed
@@ -5092,7 +5116,7 @@ interface OrgsClient {
5092
5116
  * that file existed. Adding a code means editing the spec, that file, this
5093
5117
  * list, and the Python pair.
5094
5118
  */
5095
- declare const HOSTED_ERROR_CODES: readonly ["missing_authorization", "invalid_api_key", "read_only_key", "credential_service_unavailable", "rate_limited", "insufficient_credits", "quota_exceeded", "invalid_json", "invalid_input", "invalid_limit", "invalid_status", "invalid_visibility", "invalid_cursor", "invalid_after", "invalid_format", "invalid_ids", "invalid_multipart", "idempotency_key_reused", "dataset_not_found", "dataset_version_not_found", "dataset_name_taken", "dataset_in_use", "dataset_not_owned", "dataset_import_in_progress", "upstream_not_watchable", "no_active_version", "version_not_ready", "version_not_activatable", "unknown_task_names", "no_tasks", "task_not_found", "task_failed_to_build", "upload_session_not_found", "upload_offset_mismatch", "upload_chunk_digest_mismatch", "upload_incomplete", "upload_archive_digest_mismatch", "upload_session_failed", "too_many_concurrent_upload_chunks", "agent_not_found", "agent_name_taken", "agent_name_reserved", "agent_invalid_name", "agent_source_required", "agent_source_conflict", "agent_invalid_env", "agent_too_large", "agent_limit_reached", "skill_not_found", "skill_name_not_found", "skill_ref_invalid", "skill_unresolvable", "skill_invalid", "skill_in_use", "skill_too_large", "skill_limit_reached", "too_many_concurrent_skill_uploads", "secret_not_found", "secret_ambiguous", "secret_brokered_unsupported", "secret_exists", "secret_not_attached", "agent_version_not_found", "agent_version_unresolvable", "agent_kwarg_unsupported", "agent_config_unsupported", "agent_config_key_refused", "agent_preset_unsupported", "provider_unsupported", "job_not_found", "job_not_terminal", "no_failed_trials", "trial_not_found", "trial_not_settled", "concurrent_update", "regrade_source_ineligible", "no_regradable_trials", "invalid_rubric", "analysis_already_running", "analysis_not_found", "analysis_not_terminal", "check_not_found", "check_not_terminal", "no_checkable_tasks", "too_many_concurrent_check_uploads", "no_analyzable_trials", "session_not_found", "not_a_job_dir", "invalid_trial", "trial_too_large", "upload_too_large", "job_uploaded", "job_already_uploaded", "too_many_concurrent_job_uploads", "job_import_not_found", "import_not_found", "import_too_large", "too_many_concurrent_imports", "invalid_archive", "unpinned_git_ref", "hub_package_not_found", "hub_unreachable", "package_not_retained", "package_corrupt", "package_missing", "too_many_concurrent_package_downloads", "org_not_found", "org_slug_taken", "org_forbidden", "org_personal_immutable", "org_last_owner", "org_in_use", "org_member_not_found", "invite_not_found", "invite_invalid", "not_found", "not_captured", "filesystem_state", "feature_unsupported", "provider_unreachable", "task_package_not_retained", "internal_error"];
5119
+ declare const HOSTED_ERROR_CODES: readonly ["missing_authorization", "invalid_api_key", "read_only_key", "credential_service_unavailable", "rate_limited", "insufficient_credits", "quota_exceeded", "invalid_json", "invalid_input", "invalid_limit", "invalid_status", "invalid_visibility", "invalid_cursor", "invalid_after", "invalid_format", "invalid_ids", "invalid_multipart", "idempotency_key_reused", "dataset_not_found", "dataset_version_not_found", "dataset_name_taken", "dataset_in_use", "dataset_not_owned", "dataset_import_in_progress", "upstream_not_watchable", "no_active_version", "version_not_ready", "version_not_activatable", "unknown_task_names", "no_tasks", "task_not_found", "task_failed_to_build", "upload_session_not_found", "upload_offset_mismatch", "upload_chunk_digest_mismatch", "upload_incomplete", "upload_archive_digest_mismatch", "upload_session_failed", "too_many_concurrent_upload_chunks", "agent_not_found", "agent_name_taken", "agent_name_reserved", "agent_invalid_name", "agent_source_required", "agent_source_conflict", "agent_invalid_env", "agent_too_large", "agent_limit_reached", "skill_not_found", "skill_name_not_found", "skill_ref_invalid", "skill_unresolvable", "skill_invalid", "skill_in_use", "skill_too_large", "skill_limit_reached", "too_many_concurrent_skill_uploads", "secret_not_found", "secret_ambiguous", "secret_brokered_unsupported", "secret_exists", "secret_not_attached", "agent_version_not_found", "agent_version_unresolvable", "agent_kwarg_unsupported", "agent_config_unsupported", "agent_config_key_refused", "agent_preset_unsupported", "agent_retired", "provider_unsupported", "job_not_found", "job_not_terminal", "no_failed_trials", "trial_not_found", "trial_not_settled", "concurrent_update", "regrade_source_ineligible", "no_regradable_trials", "invalid_rubric", "analysis_already_running", "analysis_not_found", "analysis_not_terminal", "check_not_found", "check_not_terminal", "no_checkable_tasks", "too_many_concurrent_check_uploads", "no_analyzable_trials", "session_not_found", "not_a_job_dir", "invalid_trial", "trial_too_large", "upload_too_large", "job_uploaded", "job_already_uploaded", "too_many_concurrent_job_uploads", "job_import_not_found", "import_not_found", "import_too_large", "too_many_concurrent_imports", "invalid_archive", "unpinned_git_ref", "hub_package_not_found", "hub_unreachable", "package_not_retained", "package_corrupt", "package_missing", "too_many_concurrent_package_downloads", "org_not_found", "org_slug_taken", "org_forbidden", "org_personal_immutable", "org_last_owner", "org_in_use", "org_member_not_found", "invite_not_found", "invite_invalid", "not_found", "not_captured", "filesystem_state", "feature_unsupported", "provider_unreachable", "task_package_not_retained", "internal_error"];
5096
5120
  /** One of the API's stable error codes. */
5097
5121
  type HostedErrorCode = (typeof HOSTED_ERROR_CODES)[number];
5098
5122
  /** True when `value` is a code this SDK version knows about (narrowing guard). */
@@ -5117,6 +5141,11 @@ interface AgentModelOption {
5117
5141
  model_id: string;
5118
5142
  description: string | null;
5119
5143
  }
5144
+ /** One retired built-in agent and the agent to use instead — the row `agent_retired` details carry too. */
5145
+ interface RetiredAgent {
5146
+ agent: string;
5147
+ replaced_by: string;
5148
+ }
5120
5149
  /** One built-in agent's declared capabilities. */
5121
5150
  interface AgentCapability {
5122
5151
  name: string;
@@ -5228,8 +5257,15 @@ interface ManagedProviderCapability {
5228
5257
  */
5229
5258
  interface CapabilityDocument {
5230
5259
  schema_version: number;
5231
- /** Built-in agents and their declared capabilities. */
5260
+ /** Built-in agents a new job may name, and their declared capabilities. */
5232
5261
  agents: AgentCapability[];
5262
+ /**
5263
+ * Retired built-in agents, left out of `agents`: a new job, resume, retry,
5264
+ * analysis or check naming one is refused `agent_retired`, while its records
5265
+ * stay readable.
5266
+ * Absent on servers predating the field.
5267
+ */
5268
+ retired_agents?: RetiredAgent[];
5233
5269
  /** Rules a bring-your-own agent registration must satisfy. */
5234
5270
  agent_registration: {
5235
5271
  name_pattern: string;
@@ -5267,19 +5303,24 @@ interface CapabilityDocument {
5267
5303
  dataset_version: StatusVocabulary;
5268
5304
  };
5269
5305
  /**
5270
- * The trace analyzer's roster and defaults: the model an omitted
5271
- * `analyze.model_name` takes, the efforts `analyze.reasoning_effort`
5272
- * accepts (the claude harness's — the analyzer IS that harness), and
5273
- * every roster model with the effort an omitted `reasoning_effort` takes
5274
- * for it. Absent on servers predating the field.
5306
+ * The rubric agents' roster and defaults, the analyzer's and the checker's alike, so a client
5307
+ * knows what "omitted" meant. Absent on servers predating the field.
5275
5308
  */
5276
5309
  analyze?: {
5310
+ default_agent: string;
5311
+ /** The platform's pick — what an omitted `model_name` takes on every agent whose roster carries it. */
5277
5312
  default_model: string;
5278
- reasoning_efforts: string[];
5279
- models: {
5280
- alias: string;
5281
- model_id: string;
5282
- default_reasoning_effort: string;
5313
+ agents: {
5314
+ name: string;
5315
+ /** What an omitted `model_name` takes on this agent; one of its `models[].alias`. */
5316
+ default_model: string;
5317
+ /** Empty for an agent that takes no effort. */
5318
+ reasoning_efforts: string[];
5319
+ models: {
5320
+ alias: string;
5321
+ model_id: string;
5322
+ default_reasoning_effort: string | null;
5323
+ }[];
5283
5324
  }[];
5284
5325
  };
5285
5326
  limits: {
@@ -5375,4 +5416,4 @@ interface CapabilityDocument {
5375
5416
  error_codes: string[];
5376
5417
  }
5377
5418
 
5378
- export { type AnalysisEvidence as $, type AgentInput as A, type AgentArm as B, type Check as C, type DatasetVersion as D, type AgentArmInput as E, type AgentCapability as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentDatasetStats as I, type JobCreate as J, type AgentEffortSupport as K, type AgentInfo as L, type AgentList as M, type AgentModelOption as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentPage as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentResult as V, type AgentSource as W, type AgentSourceInput as X, type AgentUpsertInput as Y, type AnalysisArtifactStream as Z, type AnalysisCheck as _, type TrialAnalysis as a, type FilesystemSource as a$, type AnalysisFailure as a0, type AnalysisList as a1, type AnalysisPage as a2, type AnalysisStatus as a3, type AnalysisTranscript as a4, type AnalysisTranscriptOptions as a5, type AnalyzeConfig as a6, type AnalyzeConfigInput as a7, type AnalyzeDefaults as a8, type ApiKey as a9, type DatasetVersionArchiveSource as aA, type DatasetVersionArchiveUrlSource as aB, type DatasetVersionGitSource as aC, type DatasetVersionHubSource as aD, type DatasetVersionSource as aE, type DatasetVersionState as aF, type DownloadDatasetOptions as aG, type DownloadJobOptions as aH, EVAL_SANDBOX_PROVIDERS as aI, type Agent as aJ, type ModelInfo as aK, type EvalSandboxProvider as aL, type StepResult as aM, type ExceptionInfo as aN, type FilesystemArchiveOptions as aO, type FilesystemBox as aP, type FilesystemCapture as aQ, type FilesystemChange as aR, type FilesystemChanges as aS, type FilesystemChangesOptions as aT, type FilesystemEntry as aU, type FilesystemListOptions as aV, type FilesystemListing as aW, type FilesystemReadOptions as aX, type FilesystemSearchHit as aY, type FilesystemSearchOptions as aZ, type FilesystemSearchResult as a_, type AttemptPhase as aa, type AuthStatus as ab, type Awaitable as ac, CHECK_STATUSES as ad, type CheckDefaults as ae, type CheckRow as af, type CheckRowList as ag, type CheckRowPage as ah, type CheckTaskTally as ai, type CompareCell as aj, type CompareCoverage as ak, type CompareJobAggregate as al, type CompareResponse as am, type CompareTaskRow as an, type Dataset as ao, type DatasetImportFailure as ap, type DatasetImportList as aq, type DatasetImportPage as ar, type DatasetImportStatus as as, type DatasetList as at, type DatasetPage as au, type DatasetPatch as av, type DatasetPreflight as aw, type DatasetRef as ax, type DatasetSelector as ay, type DatasetSource as az, type DatasetImport as b, type OrgQuota as b$, type FilesystemState as b0, type FilesystemStatus as b1, type FilesystemStreamEvent as b2, type FilesystemStreamOptions as b3, type FilesystemWatchResult as b4, GATEWAY_TRACE_SEQ_BASE as b5, type GatewayUsage as b6, type GetDatasetOptions as b7, type GrepJobOptions as b8, HOSTED_ERROR_CODES as b9, type JobPage as bA, type JobShareEmail as bB, type JobShareLink as bC, type JobStats as bD, type JobStatus as bE, type JobTaskLink as bF, type JobTaskRollup as bG, type JobTaskRollupList as bH, type JobTaskRollupPage as bI, type JobViewer as bJ, type JobWatch as bK, type JudgeResult as bL, type ListAgentsOptions as bM, type ListAnalysesOptions as bN, type ListDatasetsOptions as bO, type ListImportsOptions as bP, type ListJobImportsOptions as bQ, type ListJobTasksOptions as bR, type ListJobsOptions as bS, type ListSkillsOptions as bT, type ListTrialFilesOptions as bU, type ListTrialsOptions as bV, type ManagedProviderCapability as bW, type OrgInvite as bX, type OrgInviteCreated as bY, type OrgJoined as bZ, type OrgMember as b_, type ImportPhase as ba, type ImportPhaseProgress as bb, type ImportWarning as bc, type InfraFailureSignature as bd, JOB_LIST_KINDS as be, JOB_LIST_SCOPES as bf, type JobAnalysisStats as bg, type JobDeleteResult as bh, type JobFailure as bi, type JobGrepGroup as bj, type JobGrepPage as bk, type JobImport as bl, type JobImportFailure as bm, type JobImportList as bn, type JobImportPage as bo, type JobImportPhaseName as bp, type JobImportPhaseProgress as bq, type JobImportProgress as br, type JobImportSkippedTrial as bs, type JobImportSource as bt, type JobList as bu, type JobListItem as bv, type JobListItemList as bw, type JobListItemPage as bx, type JobListKind as by, type JobListScope as bz, type JobEvent as c, type UploadProvenance as c$, type OrgRole as c0, type OrgUsage as c1, type Organization as c2, type OrganizationDetail as c3, type Page as c4, type PageOptions as c5, type PassAtKGroup as c6, type PassAtKPoint as c7, type PreflightDatasetInput as c8, type PreflightDeferredCheck as c9, TASK_LINK_REASONS as cA, TRIAL_ARTIFACT_STREAMS as cB, TRIAL_STATUSES as cC, type Task as cD, type TaskCheckTranscript as cE, type TaskLinkReason as cF, type TaskLinkedBy as cG, type TaskNote as cH, type TaskPackageFiles as cI, type TaskPackageFilesystemStatus as cJ, type TaskProviderVerdict as cK, type TimingInfo as cL, type TraceEventPage as cM, type TraceOptions as cN, type TrialArtifactStream as cO, type TrialCounts as cP, type TrialFile as cQ, type TrialFilePage as cR, type TrialFileRange as cS, type TrialGpuCost as cT, type TrialList as cU, type TrialPage as cV, type TrialStatus as cW, type TrialStatusTally as cX, type TrialTaskLink as cY, type TrialUploadProvenance as cZ, type UploadJobOptions as c_, type PreflightManifestVerdict as ca, type PreflightTaskVerdict as cb, type ProviderCapability as cc, type PublishDatasetOptions as cd, type RegradeRequest as ce, type ResumeRequest as cf, type RetryRequest as cg, type RubricCriterion as ch, type RunFilesystem as ci, SANDBOX_LOG_STREAMS as cj, type SandboxLogEvent as ck, type SandboxLogLine as cl, type SandboxLogLines as cm, type SandboxLogOptions as cn, type SandboxLogStream as co, type SandboxProcs as cp, type SkillLock as cq, type SkillUpload as cr, type SkillUploadList as cs, type SkillUploadPage as ct, type SourceJob as cu, type SpendSource as cv, type StartJobOptions as cw, type StatusVocabulary as cx, type StopResponse as cy, TASK_LINKED_BY as cz, type DatasetFailedTask as d, type UploadSkillOptions as d0, type UpstreamStatus as d1, type VerifierEnvironmentMode as d2, type VerifierResult as d3, type WatchAnalysisOptions as d4, type WatchImportOptions as d5, type WatchJobImportOptions as d6, type WatchJobOptions as d7, gatewayUsageOf as d8, isHostedErrorCode as d9, passAtK as da, type DatasetImportProgress as e, type JobSecretInline as f, type JobSecretRef as g, type TraceEvent as h, type JobVisibility as i, type JobShareRequest as j, type JobShares as k, type Job as l, type TaskCheck as m, type DatasetsClient as n, type AgentsClient as o, type JobsClient as p, type TrialsClient as q, type AnalysesClient as r, type ChecksClient as s, type CapabilityDocument as t, type HostedClientConfig as u, type AuthClient as v, AGENT_EFFORT_SUPPORT_VALUES as w, ANALYSIS_ARTIFACT_STREAMS as x, ANALYSIS_STATUSES as y, type ActiveDataset as z };
5419
+ export { type AnalysisEvidence as $, type AgentInput as A, type AgentArm as B, type Check as C, type DatasetVersion as D, type AgentArmInput as E, type AgentCapability as F, type GatewayUsageEvent as G, type HostedErrorCode as H, type AgentDatasetStats as I, type JobCreate as J, type AgentEffortSupport as K, type AgentInfo as L, type AgentList as M, type AgentModelOption as N, type OrgsClient as O, type PublishDatasetInput as P, type AgentPage as Q, type Rubric as R, type SkillsClient as S, type Trial as T, type UsageReading as U, type AgentResult as V, type AgentSource as W, type AgentSourceInput as X, type AgentUpsertInput as Y, type AnalysisArtifactStream as Z, type AnalysisCheck as _, type TrialAnalysis as a, type FilesystemSource as a$, type AnalysisFailure as a0, type AnalysisList as a1, type AnalysisPage as a2, type AnalysisStatus as a3, type AnalysisTranscript as a4, type AnalysisTranscriptOptions as a5, type AnalyzeConfig as a6, type AnalyzeConfigInput as a7, type AnalyzeDefaults as a8, type ApiKey as a9, type DatasetVersionArchiveSource as aA, type DatasetVersionArchiveUrlSource as aB, type DatasetVersionGitSource as aC, type DatasetVersionHubSource as aD, type DatasetVersionSource as aE, type DatasetVersionState as aF, type DownloadDatasetOptions as aG, type DownloadJobOptions as aH, EVAL_SANDBOX_PROVIDERS as aI, type Agent as aJ, type ModelInfo as aK, type EvalSandboxProvider as aL, type StepResult as aM, type ExceptionInfo as aN, type FilesystemArchiveOptions as aO, type FilesystemBox as aP, type FilesystemCapture as aQ, type FilesystemChange as aR, type FilesystemChanges as aS, type FilesystemChangesOptions as aT, type FilesystemEntry as aU, type FilesystemListOptions as aV, type FilesystemListing as aW, type FilesystemReadOptions as aX, type FilesystemSearchHit as aY, type FilesystemSearchOptions as aZ, type FilesystemSearchResult as a_, type AttemptPhase as aa, type AuthStatus as ab, type Awaitable as ac, CHECK_STATUSES as ad, type CheckDefaults as ae, type CheckRow as af, type CheckRowList as ag, type CheckRowPage as ah, type CheckTaskTally as ai, type CompareCell as aj, type CompareCoverage as ak, type CompareJobAggregate as al, type CompareResponse as am, type CompareTaskRow as an, type Dataset as ao, type DatasetImportFailure as ap, type DatasetImportList as aq, type DatasetImportPage as ar, type DatasetImportStatus as as, type DatasetList as at, type DatasetPage as au, type DatasetPatch as av, type DatasetPreflight as aw, type DatasetRef as ax, type DatasetSelector as ay, type DatasetSource as az, type DatasetImport as b, type OrgQuota as b$, type FilesystemState as b0, type FilesystemStatus as b1, type FilesystemStreamEvent as b2, type FilesystemStreamOptions as b3, type FilesystemWatchResult as b4, GATEWAY_TRACE_SEQ_BASE as b5, type GatewayUsage as b6, type GetDatasetOptions as b7, type GrepJobOptions as b8, HOSTED_ERROR_CODES as b9, type JobPage as bA, type JobShareEmail as bB, type JobShareLink as bC, type JobStats as bD, type JobStatus as bE, type JobTaskLink as bF, type JobTaskRollup as bG, type JobTaskRollupList as bH, type JobTaskRollupPage as bI, type JobViewer as bJ, type JobWatch as bK, type JudgeResult as bL, type ListAgentsOptions as bM, type ListAnalysesOptions as bN, type ListDatasetsOptions as bO, type ListImportsOptions as bP, type ListJobImportsOptions as bQ, type ListJobTasksOptions as bR, type ListJobsOptions as bS, type ListSkillsOptions as bT, type ListTrialFilesOptions as bU, type ListTrialsOptions as bV, type ManagedProviderCapability as bW, type OrgInvite as bX, type OrgInviteCreated as bY, type OrgJoined as bZ, type OrgMember as b_, type ImportPhase as ba, type ImportPhaseProgress as bb, type ImportWarning as bc, type InfraFailureSignature as bd, JOB_LIST_KINDS as be, JOB_LIST_SCOPES as bf, type JobAnalysisStats as bg, type JobDeleteResult as bh, type JobFailure as bi, type JobGrepGroup as bj, type JobGrepPage as bk, type JobImport as bl, type JobImportFailure as bm, type JobImportList as bn, type JobImportPage as bo, type JobImportPhaseName as bp, type JobImportPhaseProgress as bq, type JobImportProgress as br, type JobImportSkippedTrial as bs, type JobImportSource as bt, type JobList as bu, type JobListItem as bv, type JobListItemList as bw, type JobListItemPage as bx, type JobListKind as by, type JobListScope as bz, type JobEvent as c, type UploadJobOptions as c$, type OrgRole as c0, type OrgUsage as c1, type Organization as c2, type OrganizationDetail as c3, type Page as c4, type PageOptions as c5, type PassAtKGroup as c6, type PassAtKPoint as c7, type PreflightDatasetInput as c8, type PreflightDeferredCheck as c9, TASK_LINKED_BY as cA, TASK_LINK_REASONS as cB, TRIAL_ARTIFACT_STREAMS as cC, TRIAL_STATUSES as cD, type Task as cE, type TaskCheckTranscript as cF, type TaskLinkReason as cG, type TaskLinkedBy as cH, type TaskNote as cI, type TaskPackageFiles as cJ, type TaskPackageFilesystemStatus as cK, type TaskProviderVerdict as cL, type TimingInfo as cM, type TraceEventPage as cN, type TraceOptions as cO, type TrialArtifactStream as cP, type TrialCounts as cQ, type TrialFile as cR, type TrialFilePage as cS, type TrialFileRange as cT, type TrialGpuCost as cU, type TrialList as cV, type TrialPage as cW, type TrialStatus as cX, type TrialStatusTally as cY, type TrialTaskLink as cZ, type TrialUploadProvenance as c_, type PreflightManifestVerdict as ca, type PreflightTaskVerdict as cb, type ProviderCapability as cc, type PublishDatasetOptions as cd, type RegradeRequest as ce, type ResumeRequest as cf, type RetiredAgent as cg, type RetryRequest as ch, type RubricCriterion as ci, type RunFilesystem as cj, SANDBOX_LOG_STREAMS as ck, type SandboxLogEvent as cl, type SandboxLogLine as cm, type SandboxLogLines as cn, type SandboxLogOptions as co, type SandboxLogStream as cp, type SandboxProcs as cq, type SkillLock as cr, type SkillUpload as cs, type SkillUploadList as ct, type SkillUploadPage as cu, type SourceJob as cv, type SpendSource as cw, type StartJobOptions as cx, type StatusVocabulary as cy, type StopResponse as cz, type DatasetFailedTask as d, type UploadProvenance as d0, type UploadSkillOptions as d1, type UpstreamStatus as d2, type VerifierEnvironmentMode as d3, type VerifierResult as d4, type WatchAnalysisOptions as d5, type WatchImportOptions as d6, type WatchJobImportOptions as d7, type WatchJobOptions as d8, gatewayUsageOf as d9, isHostedErrorCode as da, passAtK as db, type DatasetImportProgress as e, type JobSecretInline as f, type JobSecretRef as g, type TraceEvent as h, type JobVisibility as i, type JobShareRequest as j, type JobShares as k, type Job as l, type TaskCheck as m, type DatasetsClient as n, type AgentsClient as o, type JobsClient as p, type TrialsClient as q, type AnalysesClient as r, type ChecksClient as s, type CapabilityDocument as t, type HostedClientConfig as u, type AuthClient as v, AGENT_EFFORT_SUPPORT_VALUES as w, ANALYSIS_ARTIFACT_STREAMS as x, ANALYSIS_STATUSES as y, type ActiveDataset as z };