@odla-ai/harness 0.9.0 → 0.9.1

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.
package/dist/node.d.cts CHANGED
@@ -1,4 +1,4 @@
1
- import { t as HarnessTaskSpec, g as HarnessAgentOutput, f as HarnessAgentInput, H as HarnessControlPlane, w as HarnessToolBroker, a as HarnessLease, d as HarnessInferenceRequest, e as HarnessInferenceResponse, i as CodeSessionEventData, x as HarnessToolName, y as HarnessToolRequest } from './types-BNJikP5h.cjs';
1
+ import { t as HarnessTaskSpec, g as HarnessAgentOutput, f as HarnessAgentInput, H as HarnessControlPlane, w as HarnessToolBroker, a as HarnessLease, d as HarnessInferenceRequest, e as HarnessInferenceResponse, i as CodeSessionEventData, x as HarnessToolName, y as HarnessToolRequest } from './types-BazqxWK8.cjs';
2
2
  import { CodePortableCheckpoint, CodeSourceFile, CodeVerificationReceipt, CodeCheckpointState } from '@odla-ai/camel/code';
3
3
  import { TaintLabel, ToolOutput, Skill, Inference, AgentRunBudget, AgentRun, CompactionPolicy } from '@odla-ai/ai';
4
4
  import { PolicyOutcome } from '@odla-ai/camel/policy';
@@ -370,8 +370,12 @@ interface CodeRuntimeCommandEngine {
370
370
  declare class CodeRuntimeReconciler {
371
371
  private readonly control;
372
372
  private readonly engine;
373
+ /** Reports post-acknowledgement failures that must not end the host. */
374
+ private readonly onDiagnostic?;
373
375
  private readonly results;
374
- constructor(control: CodeRuntimeControlPlane, engine: CodeRuntimeCommandEngine);
376
+ constructor(control: CodeRuntimeControlPlane, engine: CodeRuntimeCommandEngine,
377
+ /** Reports post-acknowledgement failures that must not end the host. */
378
+ onDiagnostic?: ((message: string) => void) | undefined);
375
379
  reconcile(snapshot: CodeRuntimeSnapshot): Promise<void>;
376
380
  }
377
381
 
package/dist/node.d.ts CHANGED
@@ -1,4 +1,4 @@
1
- import { t as HarnessTaskSpec, g as HarnessAgentOutput, f as HarnessAgentInput, H as HarnessControlPlane, w as HarnessToolBroker, a as HarnessLease, d as HarnessInferenceRequest, e as HarnessInferenceResponse, i as CodeSessionEventData, x as HarnessToolName, y as HarnessToolRequest } from './types-BNJikP5h.js';
1
+ import { t as HarnessTaskSpec, g as HarnessAgentOutput, f as HarnessAgentInput, H as HarnessControlPlane, w as HarnessToolBroker, a as HarnessLease, d as HarnessInferenceRequest, e as HarnessInferenceResponse, i as CodeSessionEventData, x as HarnessToolName, y as HarnessToolRequest } from './types-BazqxWK8.js';
2
2
  import { CodePortableCheckpoint, CodeSourceFile, CodeVerificationReceipt, CodeCheckpointState } from '@odla-ai/camel/code';
3
3
  import { TaintLabel, ToolOutput, Skill, Inference, AgentRunBudget, AgentRun, CompactionPolicy } from '@odla-ai/ai';
4
4
  import { PolicyOutcome } from '@odla-ai/camel/policy';
@@ -370,8 +370,12 @@ interface CodeRuntimeCommandEngine {
370
370
  declare class CodeRuntimeReconciler {
371
371
  private readonly control;
372
372
  private readonly engine;
373
+ /** Reports post-acknowledgement failures that must not end the host. */
374
+ private readonly onDiagnostic?;
373
375
  private readonly results;
374
- constructor(control: CodeRuntimeControlPlane, engine: CodeRuntimeCommandEngine);
376
+ constructor(control: CodeRuntimeControlPlane, engine: CodeRuntimeCommandEngine,
377
+ /** Reports post-acknowledgement failures that must not end the host. */
378
+ onDiagnostic?: ((message: string) => void) | undefined);
375
379
  reconcile(snapshot: CodeRuntimeSnapshot): Promise<void>;
376
380
  }
377
381
 
package/dist/node.js CHANGED
@@ -1,7 +1,7 @@
1
1
  import {
2
2
  runHarnessRunner,
3
3
  runLeasedAttempt
4
- } from "./chunk-NWNBSX56.js";
4
+ } from "./chunk-CIKYMC67.js";
5
5
  import {
6
6
  CODE_RUNTIME_PROTOCOL_VERSION,
7
7
  CodeRuntimeCheckpointManager,
@@ -46,7 +46,7 @@ import {
46
46
  validateMemory,
47
47
  validateRelativePath,
48
48
  verifyCodeCandidate
49
- } from "./chunk-NG7AYYH3.js";
49
+ } from "./chunk-4EPJJMFG.js";
50
50
  import {
51
51
  assertPinnedImage,
52
52
  buildContainerRunArgs,
@@ -57,9 +57,9 @@ import {
57
57
  stageWorkspace,
58
58
  stageWorkspacePair,
59
59
  verifyContainerEngineBoundary
60
- } from "./chunk-CR6RE3A2.js";
61
- import "./chunk-Q7CKOT7T.js";
62
- import "./chunk-RXNHCGWE.js";
60
+ } from "./chunk-HDIR4MM5.js";
61
+ import "./chunk-VGNIDRKM.js";
62
+ import "./chunk-I43KTCJ2.js";
63
63
 
64
64
  // src/code-runtime-memory.ts
65
65
  async function mutationId(memory) {
@@ -1 +1 @@
1
- {"version":3,"sources":["../src/testing.ts","../src/types.ts"],"sourcesContent":["import type { OracleResponse } from \"@odla-ai/ai\";\nimport {\n HARNESS_PROTOCOL_VERSION,\n type HarnessCompletion,\n type HarnessControlPlane,\n type HarnessEventInput,\n type HarnessInferenceRequest,\n type HarnessInferenceResponse,\n type HarnessLease,\n} from \"./types\";\n\n/** Create a deterministic, valid lease for harness unit and integration tests. */\nexport function fixtureLease(overrides: Partial<HarnessLease> = {}): HarnessLease {\n return {\n protocolVersion: HARNESS_PROTOCOL_VERSION,\n leaseId: \"lease_fixture\",\n generation: 1,\n expiresAt: Date.now() + 60_000,\n task: {\n taskId: \"task_fixture\",\n attemptId: \"attempt_fixture\",\n title: \"Fixture task\",\n prompt: \"Create HARNESS_RESULT.md\",\n workspace: \"fixture\",\n aiRoute: \"coding\",\n policy: {\n network: \"none\",\n timeoutMs: 30_000,\n maxOutputBytes: 1_000_000,\n maxPatchBytes: 1_000_000,\n },\n },\n ...overrides,\n };\n}\n\n/** In-memory control plane that records runner interactions without network access. */\nexport class FakeHarnessControlPlane implements HarnessControlPlane {\n readonly leases: HarnessLease[] = [];\n readonly events: HarnessEventInput[] = [];\n readonly completions: HarnessCompletion[] = [];\n readonly inferenceRequests: HarnessInferenceRequest[] = [];\n cancelRequested = false;\n response: OracleResponse = {\n id: \"oracle_fixture\",\n model: \"fixture-model\",\n provider: \"openai\",\n role: \"assistant\",\n content: [{ type: \"text\", text: \"deterministic fixture response\" }],\n stopReason: \"end_turn\",\n usage: { inputTokens: 5, outputTokens: 4 },\n };\n\n constructor(...leases: HarnessLease[]) { this.leases.push(...leases); }\n\n async lease(workspaces: string[]): Promise<HarnessLease | null> {\n const index = this.leases.findIndex((candidate) => workspaces.includes(candidate.task.workspace));\n return index < 0 ? null : this.leases.splice(index, 1)[0]!;\n }\n async heartbeat(): Promise<{ cancelRequested: boolean; expiresAt: number }> {\n return { cancelRequested: this.cancelRequested, expiresAt: Date.now() + 60_000 };\n }\n async appendEvents(_attemptId: string, _leaseId: string, events: HarnessEventInput[]): Promise<void> {\n this.events.push(...events);\n }\n async infer(_attemptId: string, _leaseId: string, request: HarnessInferenceRequest): Promise<HarnessInferenceResponse> {\n this.inferenceRequests.push(request);\n return {\n requestId: request.requestId,\n response: this.response,\n receipt: {\n provider: this.response.provider,\n model: this.response.model,\n policyVersion: 1,\n inputTokens: this.response.usage.inputTokens ?? 0,\n outputTokens: this.response.usage.outputTokens ?? 0,\n },\n };\n }\n async complete(_attemptId: string, _leaseId: string, completion: HarnessCompletion): Promise<void> {\n this.completions.push(completion);\n }\n}\n","import type { ChatInput, OracleResponse } from \"@odla-ai/ai\";\nimport type { CodeToolPresentation } from \"./code-session-event-types\";\nexport type { CodeToolLocationPreview, CodeToolPresentation } from \"./code-session-event-types\";\n\n/** Current JSONL protocol version exchanged between a runner and an agent container. */\nexport const HARNESS_PROTOCOL_VERSION = 1 as const;\n\n/** Default control-plane route used for model inference requested by coding agents. */\nexport const DEFAULT_AI_ROUTE = \"coding\" as const;\n\n/** Lifecycle state reported for a harness task and its active attempt. */\nexport type HarnessTaskStatus =\n | \"queued\"\n | \"running\"\n | \"cancel_requested\"\n | \"completed\"\n | \"failed\"\n | \"cancelled\";\n\n/** Lifecycle state of one execution attempt for a task. */\nexport type HarnessAttemptStatus = HarnessTaskStatus;\n\n/** Trusted or untrusted participant that emitted a harness event. */\nexport type HarnessActor = \"operator\" | \"runner\" | \"agent\" | \"model\" | \"system\";\n\n/** Resource and isolation limits enforced while an untrusted task executes. */\nexport interface HarnessPolicy {\n network: \"none\";\n timeoutMs: number;\n maxOutputBytes: number;\n maxPatchBytes: number;\n}\n\n/** Immutable task instructions and execution policy delivered with a lease. */\nexport interface HarnessTaskSpec {\n taskId: string;\n attemptId: string;\n title: string;\n prompt: string;\n workspace: string;\n aiRoute: string;\n policy: HarnessPolicy;\n parentAttemptId?: string | null;\n checkpointSeq?: number | null;\n}\n\n/** Time-bound assignment authorizing a runner to execute one task attempt. */\nexport interface HarnessLease {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n leaseId: string;\n generation: number;\n expiresAt: number;\n task: HarnessTaskSpec;\n}\n\n/** Runner-supplied event before the control plane assigns sequence and task metadata. */\nexport interface HarnessEventInput {\n eventId: string;\n kind: string;\n actor: HarnessActor;\n payload: unknown;\n createdAt: number;\n}\n\n/** Persisted, ordered event associated with a specific task attempt. */\nexport interface HarnessEvent extends HarnessEventInput {\n seq: number;\n taskId: string;\n attemptId: string;\n}\n\n/** List-view metadata for a task and its current attempt. */\nexport interface HarnessTaskSummary {\n taskId: string;\n attemptId: string;\n appId: string;\n env: string;\n title: string;\n prompt: string;\n workspace: string;\n aiRoute: string;\n status: HarnessTaskStatus;\n parentAttemptId: string | null;\n checkpointSeq: number | null;\n runnerId: string | null;\n generation: number;\n createdAt: number;\n updatedAt: number;\n}\n\n/** List-view metadata for one attempt, including retry ancestry and runner ownership. */\nexport interface HarnessAttemptSummary {\n attemptId: string;\n taskId: string;\n parentAttemptId: string | null;\n checkpointSeq: number | null;\n status: HarnessAttemptStatus;\n runnerId: string | null;\n generation: number;\n createdAt: number;\n updatedAt: number;\n}\n\n/** Complete task view including attempts, events, result, and generated patch. */\nexport interface HarnessTaskDetail extends HarnessTaskSummary {\n attempts: HarnessAttemptSummary[];\n events: HarnessEvent[];\n patch: string | null;\n result: unknown;\n}\n\n/** Public control-plane view of a registered harness runner. */\nexport interface HarnessRunnerView {\n runnerId: string;\n appId: string;\n env: string;\n name: string;\n createdAt: number;\n lastSeenAt: number | null;\n revokedAt: number | null;\n}\n\n/** Credential-free normalized model request sent from an agent through the runner. */\nexport interface HarnessInferenceRequest {\n requestId: string;\n /** Exact initial, follow-up, or resume command whose budget this call consumes. */\n interactionId?: string;\n call: ChatInput;\n}\n\n/** Normalized model response plus auditable provider, policy, and token metadata. */\nexport interface HarnessInferenceResponse {\n requestId: string;\n response: OracleResponse;\n receipt: {\n provider: string;\n model: string;\n policyVersion: number;\n inputTokens: number;\n outputTokens: number;\n /**\n * USD charged for this call, priced against the model the control plane\n * actually resolved.\n *\n * ABSENT when the live catalog has no price for that model — never zero.\n * An unpriced call is unknown spend, and reporting it as free is what let\n * a goal's `maxUsd` look enforced while nothing enforced it. The runtime\n * cannot compute this itself: it asks for `brokered` and only the control\n * plane knows which model answered.\n */\n costUsd?: number;\n };\n}\n\n/** Bounded, content-minimized session activity emitted by a Code runtime.\n * Message bodies are projected separately into the app's owner-private\n * odla-db chat. `interactionId` is optional so stored v1 events remain valid. */\nexport type CodeSessionEventData = (\n | { type: \"message\"; actor: \"agent\" | \"system\"; body: string }\n | { type: \"diagnostic\"; level: \"error\"; message: string }\n | { type: \"thinking\"; available: true; durationMs: number }\n | {\n type: \"tool\";\n phase: \"started\";\n tool: HarnessToolName;\n operationId?: string;\n presentation?: CodeToolPresentation;\n }\n | {\n type: \"tool\";\n phase: \"completed\";\n tool: HarnessToolName;\n ok: boolean;\n durationMs: number;\n operationId?: string;\n presentation?: CodeToolPresentation;\n }\n | {\n type: \"usage\"; provider: string; model: string;\n inputTokens: number; outputTokens: number; durationMs: number;\n interactionTokens?: number; interactionMaxTokens?: number;\n /** USD for this call; absent when the model is unpriced, never zero. */\n costUsd?: number;\n /** Cumulative USD for this owner interaction, when every call in it was\n * priced. Absent the moment one was not, so a partial total can never be\n * mistaken for the whole. */\n interactionCostUsd?: number;\n }\n | {\n type: \"status\"; status: \"running\" | \"idle\" | \"failed\" | \"checkpointed\";\n durationMs?: number;\n }\n) & { interactionId?: string };\n\n/** Registry-assigned cursor and timestamp for an owner-visible Code event. */\nexport type CodeSessionEvent = CodeSessionEventData & {\n eventId: string; sequence: number; createdAt: number;\n};\n\n/** Closed set of effects an agent container may request from its trusted broker. */\nexport type HarnessToolName =\n | \"sandbox.read\"\n | \"sandbox.list\"\n | \"sandbox.search\"\n | \"sandbox.overview\"\n | \"sandbox.where_is\"\n | \"sandbox.who_imports\"\n | \"sandbox.who_touches\"\n | \"sandbox.apply_patch\"\n | \"sandbox.run_recipe\";\n/** Correlated, structured tool request emitted by an untrusted agent container. */\nexport interface HarnessToolRequest {\n requestId: string;\n tool: HarnessToolName;\n input: Record<string, unknown>;\n}\n/** Bounded tool result returned to an agent after trusted policy evaluation. */\nexport interface HarnessToolResponse {\n requestId: string;\n ok: boolean;\n content: string;\n details?: Record<string, unknown>;\n}\n\n/** Validated JSONL message emitted by an untrusted agent container. */\nexport type HarnessAgentOutput =\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"event\";\n kind: string;\n payload?: unknown;\n }\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"inference.request\";\n requestId: string;\n call: ChatInput;\n }\n | ({ protocolVersion: typeof HARNESS_PROTOCOL_VERSION; type: \"tool.request\" } & HarnessToolRequest)\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"attempt.complete\";\n status: \"completed\" | \"failed\" | \"cancelled\";\n result?: unknown;\n };\n\n/** JSONL command or inference result written by the trusted runner to an agent. */\nexport type HarnessAgentInput =\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"task.start\";\n task: HarnessTaskSpec;\n }\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"inference.response\";\n requestId: string;\n response: OracleResponse;\n }\n | ({ protocolVersion: typeof HARNESS_PROTOCOL_VERSION; type: \"tool.response\" } & HarnessToolResponse)\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"attempt.cancel\";\n reason: string;\n };\n\n/** Terminal attempt report submitted by a runner to the control plane. */\nexport interface HarnessCompletion {\n status: \"completed\" | \"failed\" | \"cancelled\";\n result?: unknown;\n patch?: string;\n error?: string;\n}\n\n/** Operations a credentialed runner may perform against the harness control plane. */\nexport interface HarnessControlPlane {\n lease(workspaces: string[]): Promise<HarnessLease | null>;\n heartbeat(attemptId: string, leaseId: string): Promise<{ cancelRequested: boolean; expiresAt: number }>;\n appendEvents(attemptId: string, leaseId: string, events: HarnessEventInput[]): Promise<void>;\n infer(attemptId: string, leaseId: string, request: HarnessInferenceRequest): Promise<HarnessInferenceResponse>;\n complete(attemptId: string, leaseId: string, completion: HarnessCompletion): Promise<void>;\n}\n\n/** Trusted inference bridge used to keep model credentials outside agent containers. */\nexport interface HarnessAiConnection {\n infer(request: HarnessInferenceRequest): Promise<HarnessInferenceResponse>;\n}\n\n/** Trusted tool boundary. Implementations must evaluate CaMeL policy before effects. */\nexport interface HarnessToolBroker {\n execute(\n context: { lease: HarnessLease; workspaceDir: string; signal?: AbortSignal },\n request: HarnessToolRequest,\n ): Promise<HarnessToolResponse>;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACKO,IAAM,2BAA2B;;;ADOjC,SAAS,aAAa,YAAmC,CAAC,GAAiB;AAChF,SAAO;AAAA,IACL,iBAAiB;AAAA,IACjB,SAAS;AAAA,IACT,YAAY;AAAA,IACZ,WAAW,KAAK,IAAI,IAAI;AAAA,IACxB,MAAM;AAAA,MACJ,QAAQ;AAAA,MACR,WAAW;AAAA,MACX,OAAO;AAAA,MACP,QAAQ;AAAA,MACR,WAAW;AAAA,MACX,SAAS;AAAA,MACT,QAAQ;AAAA,QACN,SAAS;AAAA,QACT,WAAW;AAAA,QACX,gBAAgB;AAAA,QAChB,eAAe;AAAA,MACjB;AAAA,IACF;AAAA,IACA,GAAG;AAAA,EACL;AACF;AAGO,IAAM,0BAAN,MAA6D;AAAA,EACzD,SAAyB,CAAC;AAAA,EAC1B,SAA8B,CAAC;AAAA,EAC/B,cAAmC,CAAC;AAAA,EACpC,oBAA+C,CAAC;AAAA,EACzD,kBAAkB;AAAA,EAClB,WAA2B;AAAA,IACzB,IAAI;AAAA,IACJ,OAAO;AAAA,IACP,UAAU;AAAA,IACV,MAAM;AAAA,IACN,SAAS,CAAC,EAAE,MAAM,QAAQ,MAAM,iCAAiC,CAAC;AAAA,IAClE,YAAY;AAAA,IACZ,OAAO,EAAE,aAAa,GAAG,cAAc,EAAE;AAAA,EAC3C;AAAA,EAEA,eAAe,QAAwB;AAAE,SAAK,OAAO,KAAK,GAAG,MAAM;AAAA,EAAG;AAAA,EAEtE,MAAM,MAAM,YAAoD;AAC9D,UAAM,QAAQ,KAAK,OAAO,UAAU,CAAC,cAAc,WAAW,SAAS,UAAU,KAAK,SAAS,CAAC;AAChG,WAAO,QAAQ,IAAI,OAAO,KAAK,OAAO,OAAO,OAAO,CAAC,EAAE,CAAC;AAAA,EAC1D;AAAA,EACA,MAAM,YAAsE;AAC1E,WAAO,EAAE,iBAAiB,KAAK,iBAAiB,WAAW,KAAK,IAAI,IAAI,IAAO;AAAA,EACjF;AAAA,EACA,MAAM,aAAa,YAAoB,UAAkB,QAA4C;AACnG,SAAK,OAAO,KAAK,GAAG,MAAM;AAAA,EAC5B;AAAA,EACA,MAAM,MAAM,YAAoB,UAAkB,SAAqE;AACrH,SAAK,kBAAkB,KAAK,OAAO;AACnC,WAAO;AAAA,MACL,WAAW,QAAQ;AAAA,MACnB,UAAU,KAAK;AAAA,MACf,SAAS;AAAA,QACP,UAAU,KAAK,SAAS;AAAA,QACxB,OAAO,KAAK,SAAS;AAAA,QACrB,eAAe;AAAA,QACf,aAAa,KAAK,SAAS,MAAM,eAAe;AAAA,QAChD,cAAc,KAAK,SAAS,MAAM,gBAAgB;AAAA,MACpD;AAAA,IACF;AAAA,EACF;AAAA,EACA,MAAM,SAAS,YAAoB,UAAkB,YAA8C;AACjG,SAAK,YAAY,KAAK,UAAU;AAAA,EAClC;AACF;","names":[]}
1
+ {"version":3,"sources":["../src/testing.ts","../src/types.ts"],"sourcesContent":["import type { OracleResponse } from \"@odla-ai/ai\";\nimport {\n HARNESS_PROTOCOL_VERSION,\n type HarnessCompletion,\n type HarnessControlPlane,\n type HarnessEventInput,\n type HarnessInferenceRequest,\n type HarnessInferenceResponse,\n type HarnessLease,\n} from \"./types\";\n\n/** Create a deterministic, valid lease for harness unit and integration tests. */\nexport function fixtureLease(overrides: Partial<HarnessLease> = {}): HarnessLease {\n return {\n protocolVersion: HARNESS_PROTOCOL_VERSION,\n leaseId: \"lease_fixture\",\n generation: 1,\n expiresAt: Date.now() + 60_000,\n task: {\n taskId: \"task_fixture\",\n attemptId: \"attempt_fixture\",\n title: \"Fixture task\",\n prompt: \"Create HARNESS_RESULT.md\",\n workspace: \"fixture\",\n aiRoute: \"coding\",\n policy: {\n network: \"none\",\n timeoutMs: 30_000,\n maxOutputBytes: 1_000_000,\n maxPatchBytes: 1_000_000,\n },\n },\n ...overrides,\n };\n}\n\n/** In-memory control plane that records runner interactions without network access. */\nexport class FakeHarnessControlPlane implements HarnessControlPlane {\n readonly leases: HarnessLease[] = [];\n readonly events: HarnessEventInput[] = [];\n readonly completions: HarnessCompletion[] = [];\n readonly inferenceRequests: HarnessInferenceRequest[] = [];\n cancelRequested = false;\n response: OracleResponse = {\n id: \"oracle_fixture\",\n model: \"fixture-model\",\n provider: \"openai\",\n role: \"assistant\",\n content: [{ type: \"text\", text: \"deterministic fixture response\" }],\n stopReason: \"end_turn\",\n usage: { inputTokens: 5, outputTokens: 4 },\n };\n\n constructor(...leases: HarnessLease[]) { this.leases.push(...leases); }\n\n async lease(workspaces: string[]): Promise<HarnessLease | null> {\n const index = this.leases.findIndex((candidate) => workspaces.includes(candidate.task.workspace));\n return index < 0 ? null : this.leases.splice(index, 1)[0]!;\n }\n async heartbeat(): Promise<{ cancelRequested: boolean; expiresAt: number }> {\n return { cancelRequested: this.cancelRequested, expiresAt: Date.now() + 60_000 };\n }\n async appendEvents(_attemptId: string, _leaseId: string, events: HarnessEventInput[]): Promise<void> {\n this.events.push(...events);\n }\n async infer(_attemptId: string, _leaseId: string, request: HarnessInferenceRequest): Promise<HarnessInferenceResponse> {\n this.inferenceRequests.push(request);\n return {\n requestId: request.requestId,\n response: this.response,\n receipt: {\n provider: this.response.provider,\n model: this.response.model,\n policyVersion: 1,\n inputTokens: this.response.usage.inputTokens ?? 0,\n outputTokens: this.response.usage.outputTokens ?? 0,\n },\n };\n }\n async complete(_attemptId: string, _leaseId: string, completion: HarnessCompletion): Promise<void> {\n this.completions.push(completion);\n }\n}\n","import type { ChatInput, OracleResponse } from \"@odla-ai/ai\";\nimport type { CodeToolPresentation } from \"./code-session-event-types\";\nexport type { CodeToolLocationPreview, CodeToolPresentation } from \"./code-session-event-types\";\n\n/** Current JSONL protocol version exchanged between a runner and an agent container. */\nexport const HARNESS_PROTOCOL_VERSION = 1 as const;\n\n/** Default control-plane route used for model inference requested by coding agents. */\nexport const DEFAULT_AI_ROUTE = \"coding\" as const;\n\n/** Lifecycle state reported for a harness task and its active attempt. */\nexport type HarnessTaskStatus =\n | \"queued\"\n | \"running\"\n | \"cancel_requested\"\n | \"completed\"\n | \"failed\"\n | \"cancelled\";\n\n/** Lifecycle state of one execution attempt for a task. */\nexport type HarnessAttemptStatus = HarnessTaskStatus;\n\n/** Trusted or untrusted participant that emitted a harness event. */\nexport type HarnessActor = \"operator\" | \"runner\" | \"agent\" | \"model\" | \"system\";\n\n/** Resource and isolation limits enforced while an untrusted task executes. */\nexport interface HarnessPolicy {\n network: \"none\";\n timeoutMs: number;\n maxOutputBytes: number;\n maxPatchBytes: number;\n}\n\n/** Immutable task instructions and execution policy delivered with a lease. */\nexport interface HarnessTaskSpec {\n taskId: string;\n attemptId: string;\n title: string;\n prompt: string;\n workspace: string;\n aiRoute: string;\n policy: HarnessPolicy;\n parentAttemptId?: string | null;\n checkpointSeq?: number | null;\n}\n\n/** Time-bound assignment authorizing a runner to execute one task attempt. */\nexport interface HarnessLease {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n leaseId: string;\n generation: number;\n expiresAt: number;\n task: HarnessTaskSpec;\n}\n\n/** Runner-supplied event before the control plane assigns sequence and task metadata. */\nexport interface HarnessEventInput {\n eventId: string;\n kind: string;\n actor: HarnessActor;\n payload: unknown;\n createdAt: number;\n}\n\n/** Persisted, ordered event associated with a specific task attempt. */\nexport interface HarnessEvent extends HarnessEventInput {\n seq: number;\n taskId: string;\n attemptId: string;\n}\n\n/** List-view metadata for a task and its current attempt. */\nexport interface HarnessTaskSummary {\n taskId: string;\n attemptId: string;\n appId: string;\n env: string;\n title: string;\n prompt: string;\n workspace: string;\n aiRoute: string;\n status: HarnessTaskStatus;\n parentAttemptId: string | null;\n checkpointSeq: number | null;\n runnerId: string | null;\n generation: number;\n createdAt: number;\n updatedAt: number;\n}\n\n/** List-view metadata for one attempt, including retry ancestry and runner ownership. */\nexport interface HarnessAttemptSummary {\n attemptId: string;\n taskId: string;\n parentAttemptId: string | null;\n checkpointSeq: number | null;\n status: HarnessAttemptStatus;\n runnerId: string | null;\n generation: number;\n createdAt: number;\n updatedAt: number;\n}\n\n/** Complete task view including attempts, events, result, and generated patch. */\nexport interface HarnessTaskDetail extends HarnessTaskSummary {\n attempts: HarnessAttemptSummary[];\n events: HarnessEvent[];\n patch: string | null;\n result: unknown;\n}\n\n/** Public control-plane view of a registered harness runner. */\nexport interface HarnessRunnerView {\n runnerId: string;\n appId: string;\n env: string;\n name: string;\n createdAt: number;\n lastSeenAt: number | null;\n revokedAt: number | null;\n}\n\n/** Credential-free normalized model request sent from an agent through the runner. */\nexport interface HarnessInferenceRequest {\n requestId: string;\n /** Exact initial, follow-up, or resume command whose budget this call consumes. */\n interactionId?: string;\n call: ChatInput;\n}\n\n/** Normalized model response plus auditable provider, policy, and token metadata. */\nexport interface HarnessInferenceResponse {\n requestId: string;\n response: OracleResponse;\n receipt: {\n provider: string;\n model: string;\n policyVersion: number;\n inputTokens: number;\n outputTokens: number;\n /**\n * USD charged for this call, priced against the model the control plane\n * actually resolved.\n *\n * ABSENT when the live catalog has no price for that model — never zero.\n * An unpriced call is unknown spend, and reporting it as free is what let\n * a goal's `maxUsd` look enforced while nothing enforced it. The runtime\n * cannot compute this itself: it asks for `brokered` and only the control\n * plane knows which model answered.\n */\n costUsd?: number;\n };\n}\n\n/** Bounded, content-minimized session activity emitted by a Code runtime.\n * Message bodies are projected separately into the app's owner-private\n * odla-db chat. `interactionId` is optional so stored v1 events remain valid. */\nexport type CodeSessionEventData = (\n | { type: \"message\"; actor: \"agent\" | \"system\"; body: string }\n | { type: \"diagnostic\"; level: \"error\"; message: string }\n | { type: \"thinking\"; available: true; durationMs: number }\n | {\n type: \"tool\";\n phase: \"started\";\n tool: HarnessToolName;\n operationId?: string;\n presentation?: CodeToolPresentation;\n }\n | {\n type: \"tool\";\n phase: \"completed\";\n tool: HarnessToolName;\n ok: boolean;\n durationMs: number;\n operationId?: string;\n presentation?: CodeToolPresentation;\n }\n | {\n type: \"collaboration\";\n phase: \"started\";\n skill: string;\n tool: string;\n operationId: string;\n }\n | {\n type: \"collaboration\";\n phase: \"completed\";\n skill: string;\n tool: string;\n ok: boolean;\n durationMs: number;\n operationId: string;\n }\n | {\n type: \"usage\"; provider: string; model: string;\n inputTokens: number; outputTokens: number; durationMs: number;\n interactionTokens?: number; interactionMaxTokens?: number;\n /** USD for this call; absent when the model is unpriced, never zero. */\n costUsd?: number;\n /** Cumulative USD for this owner interaction, when every call in it was\n * priced. Absent the moment one was not, so a partial total can never be\n * mistaken for the whole. */\n interactionCostUsd?: number;\n }\n | {\n type: \"status\"; status: \"running\" | \"idle\" | \"failed\" | \"checkpointed\";\n durationMs?: number;\n }\n) & { interactionId?: string };\n\n/** Registry-assigned cursor and timestamp for an owner-visible Code event. */\nexport type CodeSessionEvent = CodeSessionEventData & {\n eventId: string; sequence: number; createdAt: number;\n};\n\n/** Closed set of effects an agent container may request from its trusted broker. */\nexport type HarnessToolName =\n | \"sandbox.read\"\n | \"sandbox.list\"\n | \"sandbox.search\"\n | \"sandbox.overview\"\n | \"sandbox.where_is\"\n | \"sandbox.who_imports\"\n | \"sandbox.who_touches\"\n | \"sandbox.apply_patch\"\n | \"sandbox.run_recipe\";\n/** Correlated, structured tool request emitted by an untrusted agent container. */\nexport interface HarnessToolRequest {\n requestId: string;\n tool: HarnessToolName;\n input: Record<string, unknown>;\n}\n/** Bounded tool result returned to an agent after trusted policy evaluation. */\nexport interface HarnessToolResponse {\n requestId: string;\n ok: boolean;\n content: string;\n details?: Record<string, unknown>;\n}\n\n/** Validated JSONL message emitted by an untrusted agent container. */\nexport type HarnessAgentOutput =\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"event\";\n kind: string;\n payload?: unknown;\n }\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"inference.request\";\n requestId: string;\n call: ChatInput;\n }\n | ({ protocolVersion: typeof HARNESS_PROTOCOL_VERSION; type: \"tool.request\" } & HarnessToolRequest)\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"attempt.complete\";\n status: \"completed\" | \"failed\" | \"cancelled\";\n result?: unknown;\n };\n\n/** JSONL command or inference result written by the trusted runner to an agent. */\nexport type HarnessAgentInput =\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"task.start\";\n task: HarnessTaskSpec;\n }\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"inference.response\";\n requestId: string;\n response: OracleResponse;\n }\n | ({ protocolVersion: typeof HARNESS_PROTOCOL_VERSION; type: \"tool.response\" } & HarnessToolResponse)\n | {\n protocolVersion: typeof HARNESS_PROTOCOL_VERSION;\n type: \"attempt.cancel\";\n reason: string;\n };\n\n/** Terminal attempt report submitted by a runner to the control plane. */\nexport interface HarnessCompletion {\n status: \"completed\" | \"failed\" | \"cancelled\";\n result?: unknown;\n patch?: string;\n error?: string;\n}\n\n/** Operations a credentialed runner may perform against the harness control plane. */\nexport interface HarnessControlPlane {\n lease(workspaces: string[]): Promise<HarnessLease | null>;\n heartbeat(attemptId: string, leaseId: string): Promise<{ cancelRequested: boolean; expiresAt: number }>;\n appendEvents(attemptId: string, leaseId: string, events: HarnessEventInput[]): Promise<void>;\n infer(attemptId: string, leaseId: string, request: HarnessInferenceRequest): Promise<HarnessInferenceResponse>;\n complete(attemptId: string, leaseId: string, completion: HarnessCompletion): Promise<void>;\n}\n\n/** Trusted inference bridge used to keep model credentials outside agent containers. */\nexport interface HarnessAiConnection {\n infer(request: HarnessInferenceRequest): Promise<HarnessInferenceResponse>;\n}\n\n/** Trusted tool boundary. Implementations must evaluate CaMeL policy before effects. */\nexport interface HarnessToolBroker {\n execute(\n context: { lease: HarnessLease; workspaceDir: string; signal?: AbortSignal },\n request: HarnessToolRequest,\n ): Promise<HarnessToolResponse>;\n}\n"],"mappings":";;;;;;;;;;;;;;;;;;;;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;;;ACKO,IAAM,2BAA2B;;;ADOjC,SAAS,aAAa,YAAmC,CAAC,GAAiB;AAChF,SAAO;AAAA,IACL,iBAAiB;AAAA,IACjB,SAAS;AAAA,IACT,YAAY;AAAA,IACZ,WAAW,KAAK,IAAI,IAAI;AAAA,IACxB,MAAM;AAAA,MACJ,QAAQ;AAAA,MACR,WAAW;AAAA,MACX,OAAO;AAAA,MACP,QAAQ;AAAA,MACR,WAAW;AAAA,MACX,SAAS;AAAA,MACT,QAAQ;AAAA,QACN,SAAS;AAAA,QACT,WAAW;AAAA,QACX,gBAAgB;AAAA,QAChB,eAAe;AAAA,MACjB;AAAA,IACF;AAAA,IACA,GAAG;AAAA,EACL;AACF;AAGO,IAAM,0BAAN,MAA6D;AAAA,EACzD,SAAyB,CAAC;AAAA,EAC1B,SAA8B,CAAC;AAAA,EAC/B,cAAmC,CAAC;AAAA,EACpC,oBAA+C,CAAC;AAAA,EACzD,kBAAkB;AAAA,EAClB,WAA2B;AAAA,IACzB,IAAI;AAAA,IACJ,OAAO;AAAA,IACP,UAAU;AAAA,IACV,MAAM;AAAA,IACN,SAAS,CAAC,EAAE,MAAM,QAAQ,MAAM,iCAAiC,CAAC;AAAA,IAClE,YAAY;AAAA,IACZ,OAAO,EAAE,aAAa,GAAG,cAAc,EAAE;AAAA,EAC3C;AAAA,EAEA,eAAe,QAAwB;AAAE,SAAK,OAAO,KAAK,GAAG,MAAM;AAAA,EAAG;AAAA,EAEtE,MAAM,MAAM,YAAoD;AAC9D,UAAM,QAAQ,KAAK,OAAO,UAAU,CAAC,cAAc,WAAW,SAAS,UAAU,KAAK,SAAS,CAAC;AAChG,WAAO,QAAQ,IAAI,OAAO,KAAK,OAAO,OAAO,OAAO,CAAC,EAAE,CAAC;AAAA,EAC1D;AAAA,EACA,MAAM,YAAsE;AAC1E,WAAO,EAAE,iBAAiB,KAAK,iBAAiB,WAAW,KAAK,IAAI,IAAI,IAAO;AAAA,EACjF;AAAA,EACA,MAAM,aAAa,YAAoB,UAAkB,QAA4C;AACnG,SAAK,OAAO,KAAK,GAAG,MAAM;AAAA,EAC5B;AAAA,EACA,MAAM,MAAM,YAAoB,UAAkB,SAAqE;AACrH,SAAK,kBAAkB,KAAK,OAAO;AACnC,WAAO;AAAA,MACL,WAAW,QAAQ;AAAA,MACnB,UAAU,KAAK;AAAA,MACf,SAAS;AAAA,QACP,UAAU,KAAK,SAAS;AAAA,QACxB,OAAO,KAAK,SAAS;AAAA,QACrB,eAAe;AAAA,QACf,aAAa,KAAK,SAAS,MAAM,eAAe;AAAA,QAChD,cAAc,KAAK,SAAS,MAAM,gBAAgB;AAAA,MACpD;AAAA,IACF;AAAA,EACF;AAAA,EACA,MAAM,SAAS,YAAoB,UAAkB,YAA8C;AACjG,SAAK,YAAY,KAAK,UAAU;AAAA,EAClC;AACF;","names":[]}
@@ -1,5 +1,5 @@
1
1
  import { OracleResponse } from '@odla-ai/ai';
2
- import { H as HarnessControlPlane, a as HarnessLease, b as HarnessEventInput, c as HarnessCompletion, d as HarnessInferenceRequest, e as HarnessInferenceResponse } from './types-BNJikP5h.cjs';
2
+ import { H as HarnessControlPlane, a as HarnessLease, b as HarnessEventInput, c as HarnessCompletion, d as HarnessInferenceRequest, e as HarnessInferenceResponse } from './types-BazqxWK8.cjs';
3
3
 
4
4
  /** Create a deterministic, valid lease for harness unit and integration tests. */
5
5
  declare function fixtureLease(overrides?: Partial<HarnessLease>): HarnessLease;
package/dist/testing.d.ts CHANGED
@@ -1,5 +1,5 @@
1
1
  import { OracleResponse } from '@odla-ai/ai';
2
- import { H as HarnessControlPlane, a as HarnessLease, b as HarnessEventInput, c as HarnessCompletion, d as HarnessInferenceRequest, e as HarnessInferenceResponse } from './types-BNJikP5h.js';
2
+ import { H as HarnessControlPlane, a as HarnessLease, b as HarnessEventInput, c as HarnessCompletion, d as HarnessInferenceRequest, e as HarnessInferenceResponse } from './types-BazqxWK8.js';
3
3
 
4
4
  /** Create a deterministic, valid lease for harness unit and integration tests. */
5
5
  declare function fixtureLease(overrides?: Partial<HarnessLease>): HarnessLease;
package/dist/testing.js CHANGED
@@ -1,6 +1,6 @@
1
1
  import {
2
2
  HARNESS_PROTOCOL_VERSION
3
- } from "./chunk-RXNHCGWE.js";
3
+ } from "./chunk-I43KTCJ2.js";
4
4
 
5
5
  // src/testing.ts
6
6
  function fixtureLease(overrides = {}) {
@@ -199,6 +199,20 @@ type CodeSessionEventData = ({
199
199
  durationMs: number;
200
200
  operationId?: string;
201
201
  presentation?: CodeToolPresentation;
202
+ } | {
203
+ type: "collaboration";
204
+ phase: "started";
205
+ skill: string;
206
+ tool: string;
207
+ operationId: string;
208
+ } | {
209
+ type: "collaboration";
210
+ phase: "completed";
211
+ skill: string;
212
+ tool: string;
213
+ ok: boolean;
214
+ durationMs: number;
215
+ operationId: string;
202
216
  } | {
203
217
  type: "usage";
204
218
  provider: string;
@@ -199,6 +199,20 @@ type CodeSessionEventData = ({
199
199
  durationMs: number;
200
200
  operationId?: string;
201
201
  presentation?: CodeToolPresentation;
202
+ } | {
203
+ type: "collaboration";
204
+ phase: "started";
205
+ skill: string;
206
+ tool: string;
207
+ operationId: string;
208
+ } | {
209
+ type: "collaboration";
210
+ phase: "completed";
211
+ skill: string;
212
+ tool: string;
213
+ ok: boolean;
214
+ durationMs: number;
215
+ operationId: string;
202
216
  } | {
203
217
  type: "usage";
204
218
  provider: string;
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@odla-ai/harness",
3
- "version": "0.9.0",
3
+ "version": "0.9.1",
4
4
  "description": "Safe, inspectable coding-task protocol and credentialless container runner for odla Studio.",
5
5
  "license": "MIT",
6
6
  "homepage": "https://odla.ai/docs/packages/harness",