@tangle-network/agent-runtime 0.57.0 → 0.59.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/agent.js +1 -1
- package/dist/{chunk-RLDUT4JL.js → chunk-45D64J7B.js} +263 -103
- package/dist/chunk-45D64J7B.js.map +1 -0
- package/dist/{chunk-WFEQCRQP.js → chunk-4FEUFYOY.js} +2 -2
- package/dist/{chunk-6BOIKGSU.js → chunk-IN7WHMGZ.js} +2 -2
- package/dist/{chunk-TVBXDW7C.js → chunk-MMDIORZY.js} +3 -3
- package/dist/{chunk-NJST5D2G.js → chunk-ZWGEA722.js} +2 -2
- package/dist/{coordination-BydGBQCZ.d.ts → coordination-Biw19JzN.d.ts} +3 -2
- package/dist/index.js +4 -4
- package/dist/loop-runner-bin.js +3 -3
- package/dist/loops.d.ts +2 -2
- package/dist/loops.js +23 -5
- package/dist/mcp/bin.js +3 -3
- package/dist/mcp/index.d.ts +2 -2
- package/dist/mcp/index.js +4 -4
- package/dist/runtime.d.ts +138 -53
- package/dist/runtime.js +23 -5
- package/dist/workflow.js +1 -1
- package/package.json +3 -1
- package/dist/chunk-RLDUT4JL.js.map +0 -1
- /package/dist/{chunk-WFEQCRQP.js.map → chunk-4FEUFYOY.js.map} +0 -0
- /package/dist/{chunk-6BOIKGSU.js.map → chunk-IN7WHMGZ.js.map} +0 -0
- /package/dist/{chunk-TVBXDW7C.js.map → chunk-MMDIORZY.js.map} +0 -0
- /package/dist/{chunk-NJST5D2G.js.map → chunk-ZWGEA722.js.map} +0 -0
|
@@ -5,7 +5,7 @@ import {
|
|
|
5
5
|
parseDetachedSessionRef,
|
|
6
6
|
runDetachedTurn,
|
|
7
7
|
runLoop
|
|
8
|
-
} from "./chunk-
|
|
8
|
+
} from "./chunk-45D64J7B.js";
|
|
9
9
|
import {
|
|
10
10
|
coderProfile,
|
|
11
11
|
multiHarnessCoderFanout
|
|
@@ -279,4 +279,4 @@ export {
|
|
|
279
279
|
coderTaskFromArgs,
|
|
280
280
|
settleDetachedCoderTurn
|
|
281
281
|
};
|
|
282
|
-
//# sourceMappingURL=chunk-
|
|
282
|
+
//# sourceMappingURL=chunk-4FEUFYOY.js.map
|
|
@@ -3,7 +3,7 @@ import {
|
|
|
3
3
|
} from "./chunk-FNMGYYSS.js";
|
|
4
4
|
import {
|
|
5
5
|
createDefaultCoderDelegate
|
|
6
|
-
} from "./chunk-
|
|
6
|
+
} from "./chunk-4FEUFYOY.js";
|
|
7
7
|
import {
|
|
8
8
|
runAnalystLoop
|
|
9
9
|
} from "./chunk-P5OKDSLB.js";
|
|
@@ -171,4 +171,4 @@ export {
|
|
|
171
171
|
runLoopRunnerCli,
|
|
172
172
|
parseLoopRunnerArgv
|
|
173
173
|
};
|
|
174
|
-
//# sourceMappingURL=chunk-
|
|
174
|
+
//# sourceMappingURL=chunk-IN7WHMGZ.js.map
|
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
import {
|
|
2
2
|
createFleetWorkspaceExecutor,
|
|
3
3
|
createSiblingSandboxExecutor
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-4FEUFYOY.js";
|
|
5
5
|
import {
|
|
6
6
|
runWorktreeHarness
|
|
7
|
-
} from "./chunk-
|
|
7
|
+
} from "./chunk-45D64J7B.js";
|
|
8
8
|
import {
|
|
9
9
|
buildLoopOtelSpans,
|
|
10
10
|
createOtelExporter
|
|
@@ -294,4 +294,4 @@ export {
|
|
|
294
294
|
createPropagatingTraceEmitter,
|
|
295
295
|
traceContextToEnv
|
|
296
296
|
};
|
|
297
|
-
//# sourceMappingURL=chunk-
|
|
297
|
+
//# sourceMappingURL=chunk-MMDIORZY.js.map
|
|
@@ -14,7 +14,7 @@ import {
|
|
|
14
14
|
DELEGATION_STATUS_DESCRIPTION,
|
|
15
15
|
DELEGATION_STATUS_INPUT_SCHEMA,
|
|
16
16
|
DELEGATION_STATUS_TOOL_NAME
|
|
17
|
-
} from "./chunk-
|
|
17
|
+
} from "./chunk-45D64J7B.js";
|
|
18
18
|
|
|
19
19
|
// src/mcp/openai-tools.ts
|
|
20
20
|
function buildTool(name, description, parameters) {
|
|
@@ -61,4 +61,4 @@ export {
|
|
|
61
61
|
mcpToolsForRuntimeMcp,
|
|
62
62
|
mcpToolsForRuntimeMcpSubset
|
|
63
63
|
};
|
|
64
|
-
//# sourceMappingURL=chunk-
|
|
64
|
+
//# sourceMappingURL=chunk-ZWGEA722.js.map
|
|
@@ -495,8 +495,9 @@ interface WidenGate<Out> {
|
|
|
495
495
|
* harness exits + the diff is captured, the worktree is removed.
|
|
496
496
|
*
|
|
497
497
|
* All operations spawn `git` via `child_process.spawn` synchronously
|
|
498
|
-
* (via a `runGit` helper). Stays narrow on purpose: no
|
|
499
|
-
*
|
|
498
|
+
* (via a `runGit` helper). Stays narrow on purpose: no commits, no rebases.
|
|
499
|
+
* Diff capture stages all changes (`git add -A`) into the ephemeral worktree's
|
|
500
|
+
* index so created (untracked) files appear in the `--cached` diff.
|
|
500
501
|
*/
|
|
501
502
|
/** @experimental */
|
|
502
503
|
interface WorktreeHandle {
|
package/dist/index.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import {
|
|
2
2
|
mcpToolsForRuntimeMcp,
|
|
3
3
|
mcpToolsForRuntimeMcpSubset
|
|
4
|
-
} from "./chunk-
|
|
4
|
+
} from "./chunk-ZWGEA722.js";
|
|
5
5
|
import {
|
|
6
6
|
DEFAULT_ROUTER_BASE_URL,
|
|
7
7
|
cleanModelId,
|
|
@@ -21,16 +21,16 @@ import {
|
|
|
21
21
|
runDelegatedLoop,
|
|
22
22
|
runLoopRunnerCli,
|
|
23
23
|
selfImproveLoopRunner
|
|
24
|
-
} from "./chunk-
|
|
24
|
+
} from "./chunk-IN7WHMGZ.js";
|
|
25
25
|
import "./chunk-FNMGYYSS.js";
|
|
26
|
-
import "./chunk-
|
|
26
|
+
import "./chunk-4FEUFYOY.js";
|
|
27
27
|
import "./chunk-P5OKDSLB.js";
|
|
28
28
|
import {
|
|
29
29
|
composeRuntimeHooks,
|
|
30
30
|
defineRuntimeHooks,
|
|
31
31
|
notifyRuntimeDecisionPoint,
|
|
32
32
|
notifyRuntimeHookEvent
|
|
33
|
-
} from "./chunk-
|
|
33
|
+
} from "./chunk-45D64J7B.js";
|
|
34
34
|
import "./chunk-7QYOXFCD.js";
|
|
35
35
|
import "./chunk-WIR4HOOJ.js";
|
|
36
36
|
import {
|
package/dist/loop-runner-bin.js
CHANGED
|
@@ -2,11 +2,11 @@
|
|
|
2
2
|
import {
|
|
3
3
|
parseLoopRunnerArgv,
|
|
4
4
|
runLoopRunnerCli
|
|
5
|
-
} from "./chunk-
|
|
5
|
+
} from "./chunk-IN7WHMGZ.js";
|
|
6
6
|
import "./chunk-FNMGYYSS.js";
|
|
7
|
-
import "./chunk-
|
|
7
|
+
import "./chunk-4FEUFYOY.js";
|
|
8
8
|
import "./chunk-P5OKDSLB.js";
|
|
9
|
-
import "./chunk-
|
|
9
|
+
import "./chunk-45D64J7B.js";
|
|
10
10
|
import "./chunk-7QYOXFCD.js";
|
|
11
11
|
import "./chunk-WIR4HOOJ.js";
|
|
12
12
|
import "./chunk-VLF5RHEQ.js";
|
package/dist/loops.d.ts
CHANGED
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
export { AgentProfile, CreateSandboxOptions, SandboxEvent, SandboxInstance } from '@tangle-network/sandbox';
|
|
2
|
-
export { AgenticOptions, AgenticRunResult, AgenticSurface, AgenticTask, AgenticTool, AnytimeReport, AnytimeStrategySummary, AnytimeTaskCurve, ArtifactHandle, AssertTraceDerivedFindings, AuditIntentInput, AuditIntentOptions, AuthorStrategyOptions, AuthoredCoderHarness, AuthoredProfile, AuthoredStrategy, BenchmarkCell, BenchmarkConfig, BenchmarkLift, BenchmarkReport, BenchmarkStrategySummary, BenchmarkTaskRow, BridgeSeam, BudgetPool, BudgetReadout, ChampionPick, ChampionPolicy, CheckpointCapableBox, CliSeam, CliWorktreeSeam, CoderDeliverableOptions, CoderWinnerStrategy, CombinatorShape, CompletionAnalyst, CompletionEvidence, CompletionPolicy, CompletionVerdict, CoordinationDriverOptions, CoordinationMcpHandle, Corpus, CorpusFilter, CorpusRecord, CreateScopeAnalystOptions, CriuCapableClient, DefinePersona, DefinePersonaInput, Deliverable, DeliverableSpec,
|
|
2
|
+
export { AgenticOptions, AgenticRunResult, AgenticSurface, AgenticTask, AgenticTool, AnytimeReport, AnytimeStrategySummary, AnytimeTaskCurve, ArtifactHandle, AssertTraceDerivedFindings, AuditIntentInput, AuditIntentOptions, AuthorStrategyOptions, AuthoredCoderHarness, AuthoredProfile, AuthoredStrategy, BenchmarkCell, BenchmarkConfig, BenchmarkLift, BenchmarkReport, BenchmarkStrategySummary, BenchmarkTaskRow, BridgeSeam, BudgetPool, BudgetReadout, ChampionPick, ChampionPolicy, CheckpointCapableBox, CliSeam, CliWorktreeSeam, CoderDeliverableOptions, CoderWinnerStrategy, CombinatorShape, CompletionAnalyst, CompletionEvidence, CompletionPolicy, CompletionVerdict, CoordinationDriverOptions, CoordinationMcpHandle, Corpus, CorpusFilter, CorpusRecord, CreateScopeAnalystOptions, CriuCapableClient, DefinePersona, DefinePersonaInput, Deliverable, DeliverableSpec, DriverChat, DriverMessage, DriverToolCall, DriverTurn, Environment, EqualKArm, EqualKOnCost, EqualKOnCostOptions, EqualKVerdict, EvolutionArchiveNode, EvolutionAuthor, EvolutionBandInfo, EvolutionCandidate, EvolutionGeneration, EvolutionReport, ExecutorConfig, Fanout, FanoutOptions, FanoutSynthesis, FanoutWinnerSelector, FileCorpus, FileResultBlobStore, FileSpawnJournal, FlatWidenGate, ForkCapableBox, GitWorkspaceOptions, HarvestCorpusOptions, HarvestFailure, HarvestReport, InMemoryCorpus, InMemoryResultBlobStore, InMemorySpawnJournal, Inbox, InboxMessage, IntentAudit, LoopDispatchOptions, LoopOptionsForDispatch, LoopShape, LoopUntil, LoopUntilSpec, LoopUntilState, McpEndpoint, McpEnvironmentOptions, NestedScopeSeam, Observation, ObserveInput, ObserveOptions, OpenSandboxRunOptions, Outcome, Panel, PanelJudge, PanelSpec, PanelVerdict, Persona, PersonaContext, PersonaExecutors, Pipeline, PipelineStage, PromotionGateOptions, PromotionVerdict, RegistryAnalyzeProjection, RenderCorpusToInstructions, RenderCorpusToInstructionsOptions, ReservationTicket, RouterSeam, RouterToolsSeam, RunAgenticOptions, RunPersonified, RunPersonifiedOptions, SandboxCapabilities, SandboxLineage, SandboxLineageHandle, SandboxRun, SandboxSeam, ScopeAnalyst, ScopeAnalyzeInput, ScopeWidenGate, SessionCapableBox, SessionMessageLike, SessionTraceBox, ShapeBudget, ShapeContext, ShapeRegistry, Shell, ShotPersona, ShotSpec, SteerContext, Strategy, StrategyCtx, StrategyEvolutionConfig, StrategyResult, SurfaceScore, ToolPartDecoder, ToolStepInput, TraceSource, TrajectoryAnalysis, TrajectoryNode, TrajectoryReport, TrajectoryReportFn, TrajectoryReportOptions, TurnResult, UsageSink, VerifierEnvironmentOptions, Verify, VerifySpec, WatchTraceOptions, WaterfallCollector, WaterfallReport, WaterfallSpan, Widen, WidenDecision, WidenLineage, WidenSpec, Workspace, WorkspaceCommit, WorkspaceRun, WorktreeCliExecutorOptions, WorktreeCoderFanoutOptions, WorktreeCommandResult, WorktreePatchArtifact, acquireSandbox, adaptiveRefine, analyzeTrace, anytimeReport, asAuthoredProfile, assertStrategyContract, assertTraceDerivedFindings, auditIntent, authorStrategy, authoredWorker, breadthDriver, buildSteerContext, builtinShapes, cliWorktreeExecutor, coderDeliverable, completionAuthorizes, contentAddress, coordinationDriverAgent, createBudgetPool, createExecutor, createExecutorRegistry, createInbox, createMcpEnvironment, createPartsTraceSource, createPushTraceSource, createRootHandle, createSandboxLineage, createScope, createScopeAnalyst, createShapeRegistry, createSupervisor, createVerifierEnvironment, createWaterfallCollector, createWorktreeCliExecutor, decodeAnthropicPart, decodeOpenAiPart, decodeOpencodePart, decodeToolPart, defaultAnalystInstruction, defaultAuditorInstruction, defaultToolDetectors, definePersona, defineStrategy, depthDriver, deterministicCompletion, discriminatingMeans, driverChild, driverExecutorFactory, driverRuntime, equalKOnCost, extractLlmCallEvent, fanout, flatWidenGate, gateOnDeliverable, gitWorkspace, harvestCorpus, inlineSandboxClient, isDriverSpec, jjWorkspace, localShell, loopDispatch, loopUntil, mapSandboxEvent, materializeTreeView, nestedScopeSeamKey, observe, openSandboxRun, panel, pickChampion, pipeline, printBenchmarkReport, probeSandboxCapabilities, promotionGate, refine, registerShape, registryScopeAnalyst, renderAnytimeTable, renderCorpusToInstructions, renderReport, replaySpawnTree, reportLoopUsage, routerDriverChat, runAgentic, runBenchmark, runInWorkspace, runPersonified, runStrategyEvolution, sample, sampleThenRefine, sandboxSessionTraceSource, selectChampion, sentinelCompletion, serveCoordinationMcp, settledToIteration, spendFromUsageEvents, stopSentinel, strategyAuthorContract, supervisorSkill, toToolSpan, toolPartDecoders, trajectoryReport, verify, watchTrace, widen, withDriverExecutor, worktreeCoderFanout } from './runtime.js';
|
|
3
3
|
export { a as RouterChatResult, b as RouterChatToolsResult, R as RouterConfig, c as RouterToolCall, d as RouterToolLoopResult, T as ToolSpec, r as routerChatWithTools, e as routerChatWithUsage, f as routerToolLoop } from './router-client-30Y_pca8.js';
|
|
4
4
|
export { R as RunLoopOptions, c as createSandboxForSpec, d as defaultSelectWinner, r as runLoop } from './run-loop-CcqfR_gy.js';
|
|
5
|
-
export { z as Agent, y as AgentSpec, B as Budget, Y as BusEvent, Z as BusRecord, _ as BusStats, $ as EventBus, U as Executor, a0 as ExecutorContext, P as ExecutorFactory, E as ExecutorRegistry, a1 as ExecutorResult, a2 as Handle, N as NodeId, a3 as NodeSnapshot, a4 as NodeStatus, a5 as PublishOptions, a6 as Restart, u as ResultBlobStore, H as RootHandle, a7 as RootSignal, a8 as Runtime, O as Scope, x as Settled, w as SpawnEvent, v as SpawnJournal, a9 as SpawnOpts, L as Spend, K as SupervisedResult, X as Supervisor, aa as SupervisorOpts, T as TreeView, V as UsageEvent, ab as WidenGate, ac as createEventBus } from './coordination-
|
|
5
|
+
export { z as Agent, y as AgentSpec, B as Budget, Y as BusEvent, Z as BusRecord, _ as BusStats, $ as EventBus, U as Executor, a0 as ExecutorContext, P as ExecutorFactory, E as ExecutorRegistry, a1 as ExecutorResult, a2 as Handle, N as NodeId, a3 as NodeSnapshot, a4 as NodeStatus, a5 as PublishOptions, a6 as Restart, u as ResultBlobStore, H as RootHandle, a7 as RootSignal, a8 as Runtime, O as Scope, x as Settled, w as SpawnEvent, v as SpawnJournal, a9 as SpawnOpts, L as Spend, K as SupervisedResult, X as Supervisor, aa as SupervisorOpts, T as TreeView, V as UsageEvent, ab as WidenGate, ac as createEventBus } from './coordination-Biw19JzN.js';
|
|
6
6
|
export { A as AgentRunSpec, D as Driver, E as ExecCtx, I as Iteration, F as LoopDecisionPayload, G as LoopEndedPayload, H as LoopIterationDispatchPayload, J as LoopIterationEndedPayload, M as LoopIterationStartedPayload, a as LoopLineageOptions, N as LoopPlanDescription, P as LoopPlanPayload, b as LoopResult, f as LoopSandboxPlacement, Q as LoopStartedPayload, T as LoopTeardownFailedPayload, e as LoopTokenUsage, g as LoopTraceEmitter, d as LoopTraceEvent, L as LoopWinner, O as OutputAdapter, S as SandboxClient, U as ValidationCtx, V as Validator } from './types-CUzjRFZ3.js';
|
|
7
7
|
export { DefaultVerdict } from '@tangle-network/agent-eval';
|
|
8
8
|
import './runtime-hooks-C7JwKb9E.js';
|
package/dist/loops.js
CHANGED
|
@@ -7,6 +7,7 @@ import {
|
|
|
7
7
|
InMemorySpawnJournal,
|
|
8
8
|
acquireSandbox,
|
|
9
9
|
adaptiveRefine,
|
|
10
|
+
analyzeTrace,
|
|
10
11
|
anytimeReport,
|
|
11
12
|
asAuthoredProfile,
|
|
12
13
|
assertStrategyContract,
|
|
@@ -23,12 +24,13 @@ import {
|
|
|
23
24
|
contentAddress,
|
|
24
25
|
coordinationDriverAgent,
|
|
25
26
|
createBudgetPool,
|
|
26
|
-
createDetectorMonitor,
|
|
27
27
|
createEventBus,
|
|
28
28
|
createExecutor,
|
|
29
29
|
createExecutorRegistry,
|
|
30
30
|
createInbox,
|
|
31
31
|
createMcpEnvironment,
|
|
32
|
+
createPartsTraceSource,
|
|
33
|
+
createPushTraceSource,
|
|
32
34
|
createRootHandle,
|
|
33
35
|
createSandboxForSpec,
|
|
34
36
|
createSandboxLineage,
|
|
@@ -36,10 +38,13 @@ import {
|
|
|
36
38
|
createScopeAnalyst,
|
|
37
39
|
createShapeRegistry,
|
|
38
40
|
createSupervisor,
|
|
39
|
-
createTrajectoryRecorder,
|
|
40
41
|
createVerifierEnvironment,
|
|
41
42
|
createWaterfallCollector,
|
|
42
43
|
createWorktreeCliExecutor,
|
|
44
|
+
decodeAnthropicPart,
|
|
45
|
+
decodeOpenAiPart,
|
|
46
|
+
decodeOpencodePart,
|
|
47
|
+
decodeToolPart,
|
|
43
48
|
defaultAnalystInstruction,
|
|
44
49
|
defaultAuditorInstruction,
|
|
45
50
|
defaultSelectWinner,
|
|
@@ -94,6 +99,7 @@ import {
|
|
|
94
99
|
runStrategyEvolution,
|
|
95
100
|
sample,
|
|
96
101
|
sampleThenRefine,
|
|
102
|
+
sandboxSessionTraceSource,
|
|
97
103
|
selectChampion,
|
|
98
104
|
sentinelCompletion,
|
|
99
105
|
serveCoordinationMcp,
|
|
@@ -102,12 +108,15 @@ import {
|
|
|
102
108
|
stopSentinel,
|
|
103
109
|
strategyAuthorContract,
|
|
104
110
|
supervisorSkill,
|
|
111
|
+
toToolSpan,
|
|
112
|
+
toolPartDecoders,
|
|
105
113
|
trajectoryReport,
|
|
106
114
|
verify,
|
|
115
|
+
watchTrace,
|
|
107
116
|
widen,
|
|
108
117
|
withDriverExecutor,
|
|
109
118
|
worktreeCoderFanout
|
|
110
|
-
} from "./chunk-
|
|
119
|
+
} from "./chunk-45D64J7B.js";
|
|
111
120
|
import "./chunk-7QYOXFCD.js";
|
|
112
121
|
import "./chunk-WIR4HOOJ.js";
|
|
113
122
|
import {
|
|
@@ -126,6 +135,7 @@ export {
|
|
|
126
135
|
InMemorySpawnJournal,
|
|
127
136
|
acquireSandbox,
|
|
128
137
|
adaptiveRefine,
|
|
138
|
+
analyzeTrace,
|
|
129
139
|
anytimeReport,
|
|
130
140
|
asAuthoredProfile,
|
|
131
141
|
assertStrategyContract,
|
|
@@ -142,12 +152,13 @@ export {
|
|
|
142
152
|
contentAddress,
|
|
143
153
|
coordinationDriverAgent,
|
|
144
154
|
createBudgetPool,
|
|
145
|
-
createDetectorMonitor,
|
|
146
155
|
createEventBus,
|
|
147
156
|
createExecutor,
|
|
148
157
|
createExecutorRegistry,
|
|
149
158
|
createInbox,
|
|
150
159
|
createMcpEnvironment,
|
|
160
|
+
createPartsTraceSource,
|
|
161
|
+
createPushTraceSource,
|
|
151
162
|
createRootHandle,
|
|
152
163
|
createSandboxForSpec,
|
|
153
164
|
createSandboxLineage,
|
|
@@ -155,10 +166,13 @@ export {
|
|
|
155
166
|
createScopeAnalyst,
|
|
156
167
|
createShapeRegistry,
|
|
157
168
|
createSupervisor,
|
|
158
|
-
createTrajectoryRecorder,
|
|
159
169
|
createVerifierEnvironment,
|
|
160
170
|
createWaterfallCollector,
|
|
161
171
|
createWorktreeCliExecutor,
|
|
172
|
+
decodeAnthropicPart,
|
|
173
|
+
decodeOpenAiPart,
|
|
174
|
+
decodeOpencodePart,
|
|
175
|
+
decodeToolPart,
|
|
162
176
|
defaultAnalystInstruction,
|
|
163
177
|
defaultAuditorInstruction,
|
|
164
178
|
defaultSelectWinner,
|
|
@@ -215,6 +229,7 @@ export {
|
|
|
215
229
|
runStrategyEvolution,
|
|
216
230
|
sample,
|
|
217
231
|
sampleThenRefine,
|
|
232
|
+
sandboxSessionTraceSource,
|
|
218
233
|
selectChampion,
|
|
219
234
|
sentinelCompletion,
|
|
220
235
|
serveCoordinationMcp,
|
|
@@ -223,8 +238,11 @@ export {
|
|
|
223
238
|
stopSentinel,
|
|
224
239
|
strategyAuthorContract,
|
|
225
240
|
supervisorSkill,
|
|
241
|
+
toToolSpan,
|
|
242
|
+
toolPartDecoders,
|
|
226
243
|
trajectoryReport,
|
|
227
244
|
verify,
|
|
245
|
+
watchTrace,
|
|
228
246
|
widen,
|
|
229
247
|
withDriverExecutor,
|
|
230
248
|
worktreeCoderFanout
|
package/dist/mcp/bin.js
CHANGED
|
@@ -3,7 +3,7 @@ import {
|
|
|
3
3
|
createPropagatingTraceEmitter,
|
|
4
4
|
detectExecutor,
|
|
5
5
|
readTraceContextFromEnv
|
|
6
|
-
} from "../chunk-
|
|
6
|
+
} from "../chunk-MMDIORZY.js";
|
|
7
7
|
import {
|
|
8
8
|
resolveRouterBaseUrl
|
|
9
9
|
} from "../chunk-GRAGM4MC.js";
|
|
@@ -11,7 +11,7 @@ import {
|
|
|
11
11
|
coderTaskFromArgs,
|
|
12
12
|
createDefaultCoderDelegate,
|
|
13
13
|
settleDetachedCoderTurn
|
|
14
|
-
} from "../chunk-
|
|
14
|
+
} from "../chunk-4FEUFYOY.js";
|
|
15
15
|
import {
|
|
16
16
|
DEFAULT_SANDBOX_BASE_URL
|
|
17
17
|
} from "../chunk-XRYEZPR6.js";
|
|
@@ -26,7 +26,7 @@ import {
|
|
|
26
26
|
parseDetachedSessionRef,
|
|
27
27
|
runDetachedTurn,
|
|
28
28
|
runLoop
|
|
29
|
-
} from "../chunk-
|
|
29
|
+
} from "../chunk-45D64J7B.js";
|
|
30
30
|
import {
|
|
31
31
|
coderProfile
|
|
32
32
|
} from "../chunk-7QYOXFCD.js";
|
package/dist/mcp/index.d.ts
CHANGED
|
@@ -2,8 +2,8 @@ import { S as SandboxClient, f as LoopSandboxPlacement } from '../types-CUzjRFZ3
|
|
|
2
2
|
import { F as FleetHandle, f as DelegationExecutor, d as DelegationTaskQueue, e as CoderDelegate, D as DelegateCodeArgs, g as DelegateCodeResult, h as DelegateFeedbackResult, b as DelegateFeedbackArgs, i as ResearchSource, R as ResearcherDelegate, j as DelegateResearchArgs, k as DelegateResearchResult, U as UiAuditorDelegate, l as DelegateUiAuditArgs, m as DelegateUiAuditResult, n as DelegationHistoryResult, o as DelegationHistoryArgs, p as DelegationStatusResult, q as DelegationStatusArgs } from '../delegates-C94qchkz.js';
|
|
3
3
|
export { r as CappedDelegationTrace, s as CoderReview, C as CoderReviewer, a as CoderWinnerSelection, t as CreateDefaultCoderDelegateOptions, u as DELEGATION_TRACE_MAX_BYTES, v as DELEGATION_TRACE_MAX_SPANS, w as DelegateCodeConfig, x as DelegateResearchConfig, y as DelegateRunCtx, z as DelegateUiAuditConfig, A as DelegateUiAuditRoute, B as DelegationError, c as DelegationFeedbackSnapshot, E as DelegationHistoryEntry, G as DelegationPersistenceError, H as DelegationProfile, I as DelegationProgress, J as DelegationRecord, K as DelegationResultPayload, L as DelegationResumeContext, M as DelegationResumeDriver, N as DelegationResumeTick, O as DelegationRunContext, P as DelegationStateCorruptError, Q as DelegationStatus, S as DelegationStore, V as DelegationTaskQueueOptions, W as DelegationTraceCaps, X as DelegationTraceCollector, Y as DelegationTraceSpan, Z as DetachedSessionRefParts, _ as DetachedTurn, $ as DriveTurnCapableBox, a0 as DriveTurnResumeDriverOptions, a1 as DriveTurnTick, a2 as FeedbackRating, a3 as FeedbackRefersTo, a4 as FileDelegationStore, a5 as FileDelegationStoreOptions, a6 as FleetWorkspaceExecutorOptions, a7 as InMemoryDelegationStore, a8 as ResearchOutputShape, a9 as RunDetachedTurnOptions, aa as SettleDetachedCoderTurnOptions, ab as SiblingSandboxExecutorOptions, ac as SubmitInput, ad as SubmitOutput, T as TraceContext, ae as UiAuditorDelegationOutput, af as buildDelegationTraceSpans, ag as capDelegationTrace, ah as coderTaskFromArgs, ai as composeLoopTraceEmitters, aj as createDefaultCoderDelegate, ak as createDelegationTraceCollector, al as createDriveTurnResumeDriver, am as createFleetWorkspaceExecutor, an as createPropagatingTraceEmitter, ao as createSiblingSandboxExecutor, ap as detachedTurnEvents, aq as formatDetachedSessionRef, ar as hashIdempotencyInput, as as parseDetachedSessionRef, at as readTraceContextFromEnv, au as runDetachedTurn, av as settleDetachedCoderTurn, aw as traceContextToEnv } from '../delegates-C94qchkz.js';
|
|
4
4
|
export { B as BuildDelegationMcpServerOptions, C as ComposeProductionAgentProfileOptions, D as DELEGATION_MCP_SERVER_KEY, b as buildDelegationMcpServer, c as composeProductionAgentProfile } from '../delegation-profile-Bvfro2m1.js';
|
|
5
|
-
import { G as GitRunner, F as FeedbackStore } from '../coordination-
|
|
6
|
-
export { A as AnalystRegistry, C as CoordinationEvent, a as CoordinationTools, b as CoordinationToolsOptions, c as CreateWorktreeOptions, D as DiffOptions, d as DiffResult, e as FeedbackEvent, I as InMemoryFeedbackStore, J as JsonRpcMessage, f as JsonRpcResponse, M as MakeWorkerAgent, g as McpServer, h as McpServerOptions, i as McpToolDescriptor, j as McpTransport, Q as Question, k as QuestionDecision, l as QuestionPolicy, m as QuestionRecord, R as RemoveWorktreeOptions, S as SettledWorker, W as WorktreeHandle, n as captureWorktreeDiff, o as createCoordinationTools, p as createInProcessTransport, q as createMcpServer, r as createWorktree, s as eventToSnapshot, t as removeWorktree } from '../coordination-
|
|
5
|
+
import { G as GitRunner, F as FeedbackStore } from '../coordination-Biw19JzN.js';
|
|
6
|
+
export { A as AnalystRegistry, C as CoordinationEvent, a as CoordinationTools, b as CoordinationToolsOptions, c as CreateWorktreeOptions, D as DiffOptions, d as DiffResult, e as FeedbackEvent, I as InMemoryFeedbackStore, J as JsonRpcMessage, f as JsonRpcResponse, M as MakeWorkerAgent, g as McpServer, h as McpServerOptions, i as McpToolDescriptor, j as McpTransport, Q as Question, k as QuestionDecision, l as QuestionPolicy, m as QuestionRecord, R as RemoveWorktreeOptions, S as SettledWorker, W as WorktreeHandle, n as captureWorktreeDiff, o as createCoordinationTools, p as createInProcessTransport, q as createMcpServer, r as createWorktree, s as eventToSnapshot, t as removeWorktree } from '../coordination-Biw19JzN.js';
|
|
7
7
|
import { L as LocalHarness, r as runLocalHarness } from '../local-harness-BE_h8szs.js';
|
|
8
8
|
export { a as LocalHarnessResult, R as RunLocalHarnessOptions } from '../local-harness-BE_h8szs.js';
|
|
9
9
|
export { C as CreateKbGateOptions, F as FactCandidate, a as FactJudge, b as FactJudgeVerdict, K as KbGateResult, c as createKbGate } from '../kb-gate-CuzMYGYM.js';
|
package/dist/mcp/index.js
CHANGED
|
@@ -4,11 +4,11 @@ import {
|
|
|
4
4
|
detectExecutor,
|
|
5
5
|
readTraceContextFromEnv,
|
|
6
6
|
traceContextToEnv
|
|
7
|
-
} from "../chunk-
|
|
7
|
+
} from "../chunk-MMDIORZY.js";
|
|
8
8
|
import {
|
|
9
9
|
mcpToolsForRuntimeMcp,
|
|
10
10
|
mcpToolsForRuntimeMcpSubset
|
|
11
|
-
} from "../chunk-
|
|
11
|
+
} from "../chunk-ZWGEA722.js";
|
|
12
12
|
import {
|
|
13
13
|
createKbGate
|
|
14
14
|
} from "../chunk-FNMGYYSS.js";
|
|
@@ -18,7 +18,7 @@ import {
|
|
|
18
18
|
createFleetWorkspaceExecutor,
|
|
19
19
|
createSiblingSandboxExecutor,
|
|
20
20
|
settleDetachedCoderTurn
|
|
21
|
-
} from "../chunk-
|
|
21
|
+
} from "../chunk-4FEUFYOY.js";
|
|
22
22
|
import {
|
|
23
23
|
DELEGATION_MCP_SERVER_KEY,
|
|
24
24
|
buildDelegationMcpServer,
|
|
@@ -81,7 +81,7 @@ import {
|
|
|
81
81
|
validateDelegateUiAuditArgs,
|
|
82
82
|
validateDelegationHistoryArgs,
|
|
83
83
|
validateDelegationStatusArgs
|
|
84
|
-
} from "../chunk-
|
|
84
|
+
} from "../chunk-45D64J7B.js";
|
|
85
85
|
import "../chunk-7QYOXFCD.js";
|
|
86
86
|
import "../chunk-WIR4HOOJ.js";
|
|
87
87
|
import "../chunk-VLF5RHEQ.js";
|
package/dist/runtime.d.ts
CHANGED
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
import { AgentProfile, BackendType, CreateSandboxOptions, SandboxInstance, SandboxEvent } from '@tangle-network/sandbox';
|
|
2
2
|
export { AgentProfile, CreateSandboxOptions, SandboxEvent, SandboxInstance } from '@tangle-network/sandbox';
|
|
3
|
-
import { u as ResultBlobStore, v as SpawnJournal, N as NodeId, w as SpawnEvent, T as TreeView, x as Settled, y as AgentSpec, E as ExecutorRegistry, B as Budget, z as Agent, H as RootHandle, K as SupervisedResult, L as Spend, O as Scope, P as ExecutorFactory, U as Executor, V as UsageEvent, G as GitRunner, M as MakeWorkerAgent, a as CoordinationTools, A as AnalystRegistry, C as CoordinationEvent, l as QuestionPolicy, X as Supervisor } from './coordination-
|
|
4
|
-
export { Y as BusEvent, Z as BusRecord, _ as BusStats, $ as EventBus, a0 as ExecutorContext, a1 as ExecutorResult, a2 as Handle, a3 as NodeSnapshot, a4 as NodeStatus, a5 as PublishOptions, a6 as Restart, a7 as RootSignal, a8 as Runtime, a9 as SpawnOpts, aa as SupervisorOpts, ab as WidenGate, ac as createEventBus } from './coordination-
|
|
3
|
+
import { u as ResultBlobStore, v as SpawnJournal, N as NodeId, w as SpawnEvent, T as TreeView, x as Settled, y as AgentSpec, E as ExecutorRegistry, B as Budget, z as Agent, H as RootHandle, K as SupervisedResult, L as Spend, O as Scope, P as ExecutorFactory, U as Executor, V as UsageEvent, G as GitRunner, M as MakeWorkerAgent, a as CoordinationTools, A as AnalystRegistry, C as CoordinationEvent, l as QuestionPolicy, X as Supervisor } from './coordination-Biw19JzN.js';
|
|
4
|
+
export { Y as BusEvent, Z as BusRecord, _ as BusStats, $ as EventBus, a0 as ExecutorContext, a1 as ExecutorResult, a2 as Handle, a3 as NodeSnapshot, a4 as NodeStatus, a5 as PublishOptions, a6 as Restart, a7 as RootSignal, a8 as Runtime, a9 as SpawnOpts, aa as SupervisorOpts, ab as WidenGate, ac as createEventBus } from './coordination-Biw19JzN.js';
|
|
5
5
|
import { R as RuntimeHooks } from './runtime-hooks-C7JwKb9E.js';
|
|
6
|
-
import { ChatClient, AnalystFinding, DefaultVerdict, AgentProfile as AgentProfile$1, AnalystRunInputs,
|
|
6
|
+
import { ChatClient, AnalystFinding, DefaultVerdict, AgentProfile as AgentProfile$1, AnalystRunInputs, ToolSpan, StreamingDetector, DetectorSignal, buildTrajectory } from '@tangle-network/agent-eval';
|
|
7
7
|
export { DefaultVerdict } from '@tangle-network/agent-eval';
|
|
8
8
|
import { I as Iteration, S as SandboxClient, b as LoopResult, e as LoopTokenUsage, R as RuntimeStreamEvent, A as AgentRunSpec, E as ExecCtx } from './types-CUzjRFZ3.js';
|
|
9
9
|
export { D as Driver, F as LoopDecisionPayload, G as LoopEndedPayload, H as LoopIterationDispatchPayload, J as LoopIterationEndedPayload, M as LoopIterationStartedPayload, a as LoopLineageOptions, N as LoopPlanDescription, P as LoopPlanPayload, f as LoopSandboxPlacement, Q as LoopStartedPayload, T as LoopTeardownFailedPayload, g as LoopTraceEmitter, d as LoopTraceEvent, L as LoopWinner, O as OutputAdapter, U as ValidationCtx, V as Validator } from './types-CUzjRFZ3.js';
|
|
@@ -3383,38 +3383,141 @@ declare function serveCoordinationMcp(opts: {
|
|
|
3383
3383
|
/**
|
|
3384
3384
|
* @experimental
|
|
3385
3385
|
*
|
|
3386
|
-
*
|
|
3387
|
-
*
|
|
3388
|
-
*
|
|
3389
|
-
*
|
|
3390
|
-
*
|
|
3386
|
+
* `TraceSource` — the ONE substrate-agnostic source of a worker's tool-call trace. The online
|
|
3387
|
+
* detectors and the settle-time analyzers consume agent-eval `ToolSpan`s from here, regardless of
|
|
3388
|
+
* whether the worker is:
|
|
3389
|
+
* - an OWNED tool loop (router-tools, cli-bridge tool dispatch) → push spans as we dispatch them;
|
|
3390
|
+
* - a SANDBOX / fleet box → read the harness's tool calls off the session (`streamPrompt` parts
|
|
3391
|
+
* live, `session.messages()` / `findCompletedTurn` at settle).
|
|
3391
3392
|
*
|
|
3392
|
-
*
|
|
3393
|
-
*
|
|
3394
|
-
*
|
|
3393
|
+
* The common currency is agent-eval's `ToolSpan` (so the same detectors + `buildTrajectory`/
|
|
3394
|
+
* `stuckLoopView`/`toolWasteView` run over any source). A source exposes two lanes:
|
|
3395
|
+
* - `onSpan` — live spans for ONLINE detection (best-effort; a black-box box may only collect).
|
|
3396
|
+
* - `collect` — the full span set at settle for the BATCH analyzers (always available).
|
|
3397
|
+
*
|
|
3398
|
+
* This module imports NO substrate SDK — it decodes generic message parts / OpenAI tool-call shapes.
|
|
3399
|
+
* The sandbox wiring (`sandboxSessionTraceSource`) is the thin adapter that feeds box session parts in.
|
|
3395
3400
|
*/
|
|
3396
3401
|
|
|
3397
|
-
interface
|
|
3402
|
+
interface ToolStepInput {
|
|
3398
3403
|
readonly toolName: string;
|
|
3399
3404
|
readonly args: unknown;
|
|
3400
|
-
/** Whether the tool call errored (drives error-streak detection). Omit/`'ok'` when unknown. */
|
|
3401
3405
|
readonly status?: 'ok' | 'error';
|
|
3402
|
-
|
|
3403
|
-
|
|
3404
|
-
|
|
3405
|
-
|
|
3406
|
-
|
|
3407
|
-
|
|
3408
|
-
|
|
3406
|
+
readonly result?: unknown;
|
|
3407
|
+
/** Stable id of the tool call — used to de-duplicate the repeated state transitions a harness
|
|
3408
|
+
* streams for one call (opencode emits pending→running→completed, plus a `raw`-wrapped copy). */
|
|
3409
|
+
readonly callId?: string;
|
|
3410
|
+
}
|
|
3411
|
+
interface TraceSource {
|
|
3412
|
+
/** Subscribe to tool spans as they are produced (ONLINE). Returns an unsubscribe. A source that
|
|
3413
|
+
* only exposes its trace at the end registers nothing and returns a no-op. */
|
|
3414
|
+
onSpan(handler: (span: ToolSpan) => void): () => void;
|
|
3415
|
+
/** The full set of tool spans for the run (SETTLE / batch). Always available. */
|
|
3416
|
+
collect(): Promise<ToolSpan[]>;
|
|
3417
|
+
}
|
|
3418
|
+
/** Project a normalized tool step into the canonical agent-eval `ToolSpan`. */
|
|
3419
|
+
declare function toToolSpan(input: ToolStepInput, runId: string, seq: number, at: number): ToolSpan;
|
|
3420
|
+
/** Decode one harness message part into a tool step, or `undefined` if it is not a (completed) tool
|
|
3421
|
+
* call. ONE adapter per harness family — each owns its real wire shape; the flow downstream is
|
|
3422
|
+
* identical. Add a harness = add a decoder + register it; no other code changes. */
|
|
3423
|
+
type ToolPartDecoder = (part: Record<string, unknown>) => ToolStepInput | undefined;
|
|
3424
|
+
/** opencode parts ARE agent-interface's canonical `ToolPart` (`{ type:'tool', tool, callID?,
|
|
3425
|
+
* state: ToolState }`) — the shape every adc sdk-provider normalizes its harness output into. We
|
|
3426
|
+
* decode against that published type (single source of truth) rather than a re-derived shape; the
|
|
3427
|
+
* `ToolState` union drives the status mapping, so a status that adc adds/renames is a compile error
|
|
3428
|
+
* here, not a silent miss. The same call streams pending→running→terminal; only a terminal state
|
|
3429
|
+
* (`completed` / `error` / `failed`) is a finished call. */
|
|
3430
|
+
declare const decodeOpencodePart: ToolPartDecoder;
|
|
3431
|
+
/** Anthropic / claude-code (also kimi's tool_use variant): a `{ type:'tool_use', id|tool_use_id,
|
|
3432
|
+
* name|tool, input }` content block. CONFIRMED against the cli-bridge's authoritative parsers
|
|
3433
|
+
* (claude.ts reads `block.type==='tool_use', block.id, block.name, block.input`; kimi.ts adds the
|
|
3434
|
+
* `tool_use_id`/`tool` fallbacks) — the canonical readers of these harnesses' real native output. */
|
|
3435
|
+
declare const decodeAnthropicPart: ToolPartDecoder;
|
|
3436
|
+
/** OpenAI-compatible (router / kimi's top-level form / glm): `{ type:'function'|'tool_call', id,
|
|
3437
|
+
* function:{ name, arguments:<JSON string> } }`. CONFIRMED against the cli-bridge (kimi.ts surfaces
|
|
3438
|
+
* kimi's top-level `tool_calls:[{type:'function', id, function:{name,arguments}}]` exactly this way).
|
|
3439
|
+
* NOTE: codex does NOT emit structured tool calls (the bridge's codex.ts never yields `tool_calls` —
|
|
3440
|
+
* it runs shell internally and surfaces only text), so per-tool detection is unavailable for codex
|
|
3441
|
+
* from any path — a harness property, not a decoder gap. */
|
|
3442
|
+
declare const decodeOpenAiPart: ToolPartDecoder;
|
|
3443
|
+
/** The harness → decoder registry. Add a harness by adding one entry. */
|
|
3444
|
+
declare const toolPartDecoders: Record<string, ToolPartDecoder>;
|
|
3445
|
+
/** Decode a part with a specific harness's adapter when known, else try every registered adapter
|
|
3446
|
+
* (the composite — robust to mixed/unknown streams). Never throws. */
|
|
3447
|
+
declare function decodeToolPart(part: unknown, harness?: string): ToolStepInput | undefined;
|
|
3448
|
+
/** A push source for OWNED tool loops (router-tools / cli-bridge tool dispatch): the loop calls
|
|
3449
|
+
* `record(step)` for each tool call; it becomes a span, fan-out to live subscribers + buffered for
|
|
3450
|
+
* `collect`. */
|
|
3451
|
+
declare function createPushTraceSource(opts?: {
|
|
3452
|
+
runId?: string;
|
|
3453
|
+
now?: () => number;
|
|
3454
|
+
}): {
|
|
3455
|
+
source: TraceSource;
|
|
3456
|
+
record: (input: ToolStepInput) => ToolSpan;
|
|
3457
|
+
};
|
|
3458
|
+
/** A source backed by harness message PARTS (sandbox session, cli-bridge). `collect` reads the full
|
|
3459
|
+
* part list and decodes the tool calls; `subscribe`, when given, streams parts live for online
|
|
3460
|
+
* detection. The caller supplies how to get parts (e.g. `box.session(id).messages()` flat-mapped to
|
|
3461
|
+
* parts) — keeping this module free of any substrate SDK. */
|
|
3462
|
+
declare function createPartsTraceSource(opts: {
|
|
3463
|
+
collectParts: () => Promise<ReadonlyArray<unknown>>;
|
|
3464
|
+
subscribeParts?: (onPart: (part: unknown) => void) => () => void;
|
|
3465
|
+
/** The harness whose decoder to use (e.g. 'opencode'); omit to try every registered adapter. */
|
|
3466
|
+
harness?: string;
|
|
3467
|
+
runId?: string;
|
|
3468
|
+
now?: () => number;
|
|
3469
|
+
}): TraceSource;
|
|
3470
|
+
/** A harness session message carrying parts (the shape `box.messages()` returns). Structurally typed
|
|
3471
|
+
* so this works with the real `@tangle-network/sandbox` box AND a test double, no SDK import. */
|
|
3472
|
+
interface SessionMessageLike {
|
|
3473
|
+
readonly parts?: ReadonlyArray<unknown>;
|
|
3474
|
+
}
|
|
3475
|
+
/** The minimal box surface this needs: list a session's messages (incl. mid-turn partials). */
|
|
3476
|
+
interface SessionTraceBox {
|
|
3477
|
+
messages(opts: {
|
|
3478
|
+
sessionId: string;
|
|
3479
|
+
}): Promise<ReadonlyArray<SessionMessageLike>>;
|
|
3480
|
+
}
|
|
3481
|
+
/** The SANDBOX / fleet trace source: read a box session's message parts and decode the harness's tool
|
|
3482
|
+
* calls into spans. `collect` (settle) is the solid path — `box.messages({sessionId})` → parts → spans;
|
|
3483
|
+
* black-box harnesses aren't mid-step interruptible, so online steering is the owned-loop's job and a
|
|
3484
|
+
* live `subscribe` is opt-in (pass `subscribeParts` from `streamPrompt` when the harness streams parts). */
|
|
3485
|
+
declare function sandboxSessionTraceSource(box: SessionTraceBox, sessionId: string, opts?: {
|
|
3486
|
+
/** The box's harness (e.g. 'opencode', 'claude-code') → selects its decoder adapter. */
|
|
3487
|
+
harness?: string;
|
|
3488
|
+
subscribeParts?: (onPart: (part: unknown) => void) => () => void;
|
|
3489
|
+
runId?: string;
|
|
3490
|
+
now?: () => number;
|
|
3491
|
+
}): TraceSource;
|
|
3492
|
+
|
|
3493
|
+
/**
|
|
3494
|
+
* @experimental
|
|
3495
|
+
*
|
|
3496
|
+
* The ONLINE analyst: watch a `TraceSource` and fold each tool span through agent-eval's published
|
|
3497
|
+
* streaming detector kernel (`repeatedActionDetector`/`errorStreakDetector` — the SAME kernel the
|
|
3498
|
+
* control loop folds), firing `onSignal` the moment a worker loops or error-storms. Substrate-
|
|
3499
|
+
* agnostic: it consumes spans from any source (owned router/bridge loop OR a sandbox box session),
|
|
3500
|
+
* never the raw tool seam. Detection logic + the failure taxonomy live in agent-eval; not reimplemented.
|
|
3501
|
+
*/
|
|
3502
|
+
|
|
3503
|
+
interface WatchTraceOptions {
|
|
3409
3504
|
/** The detectors to run online. Defaults to a stuck-loop + error-streak panel. */
|
|
3410
3505
|
readonly detectors?: ReadonlyArray<StreamingDetector>;
|
|
3411
3506
|
/** Fired for each signal a detector raises — the seam that raises a `finding` on the bus. */
|
|
3412
|
-
readonly onSignal?: (signal: DetectorSignal) => void | Promise<void>;
|
|
3507
|
+
readonly onSignal?: (signal: DetectorSignal, span: ToolSpan) => void | Promise<void>;
|
|
3413
3508
|
}
|
|
3414
3509
|
/** The default online panel for a tool-call pipe: a worker repeating the same call, or hammering
|
|
3415
|
-
* consecutive errors. (No-progress needs a domain progress-probe, so it is opt-in, not default.)
|
|
3510
|
+
* consecutive errors. (No-progress needs a domain progress-probe, so it is opt-in, not default.)
|
|
3511
|
+
*
|
|
3512
|
+
* Coverage note: `repeated-action` works for EVERY harness (it needs only tool name + args, which
|
|
3513
|
+
* every adapter provides). `error-streak` needs per-call status — opencode carries it inline
|
|
3514
|
+
* (`state.status`, VALIDATED live), but claude-code/codex tool-call parts do NOT (their errors live
|
|
3515
|
+
* in separate result blocks not yet decoded), so error-streak is silent for those until result-block
|
|
3516
|
+
* decoding is added + live-validated. It is in the panel because it is correct where status exists. */
|
|
3416
3517
|
declare function defaultToolDetectors(): StreamingDetector[];
|
|
3417
|
-
|
|
3518
|
+
/** Subscribe to a `TraceSource` and run the streaming detectors over its live spans. Returns an
|
|
3519
|
+
* unsubscribe. A defensive `argHash` failure (circular args) never throws out of the side-channel. */
|
|
3520
|
+
declare function watchTrace(source: TraceSource, opts?: WatchTraceOptions): () => void;
|
|
3418
3521
|
|
|
3419
3522
|
/**
|
|
3420
3523
|
* @experimental
|
|
@@ -3852,43 +3955,25 @@ declare function createRootHandle<Out>(): RootHandle<Out>;
|
|
|
3852
3955
|
/**
|
|
3853
3956
|
* @experimental
|
|
3854
3957
|
*
|
|
3855
|
-
* The
|
|
3856
|
-
*
|
|
3857
|
-
*
|
|
3858
|
-
*
|
|
3859
|
-
*
|
|
3860
|
-
* no analysis logic is reimplemented here; this is the thin adapter that bridges live tool steps into
|
|
3861
|
-
* the substrate's trace model.
|
|
3862
|
-
*
|
|
3863
|
-
* Pairs with `createDetectorMonitor` (the ONLINE pipe watcher): feed both from the worker's `onToolStep`
|
|
3864
|
-
* seam — the monitor raises findings mid-run, the recorder produces the post-hoc trajectory at settle.
|
|
3958
|
+
* The SETTLE-time analyst: when a worker finishes, collect its tool spans from a `TraceSource` and run
|
|
3959
|
+
* agent-eval's PUBLISHED batch analyzers over them — `buildTrajectory` (structured run summary),
|
|
3960
|
+
* `stuckLoopView` (full-run repeated-call view, complementing the online consecutive detector), and
|
|
3961
|
+
* `toolWasteView`. Substrate-agnostic: the spans come from any source (an owned loop's buffer OR a
|
|
3962
|
+
* sandbox box session). No analysis reimplemented — this is the thin bridge into agent-eval's analyzers.
|
|
3865
3963
|
*/
|
|
3866
3964
|
|
|
3867
|
-
interface RecordedToolStep {
|
|
3868
|
-
readonly toolName: string;
|
|
3869
|
-
readonly args: unknown;
|
|
3870
|
-
readonly status?: 'ok' | 'error';
|
|
3871
|
-
readonly result?: unknown;
|
|
3872
|
-
}
|
|
3873
3965
|
interface TrajectoryAnalysis {
|
|
3874
|
-
/** Structured run summary (tool-call count, step order). Steps
|
|
3875
|
-
*
|
|
3876
|
-
* call PATTERNS + cross-span windows, not individual durations. */
|
|
3966
|
+
/** Structured run summary (tool-call count, step order). Steps carry a single timestamp, so per-span
|
|
3967
|
+
* duration is 0; loop/waste detection keys on call PATTERNS + cross-span windows, not durations. */
|
|
3877
3968
|
readonly trajectory: Awaited<ReturnType<typeof buildTrajectory>>;
|
|
3878
|
-
/** Full-run repeated-call view (total occurrences + window)
|
|
3879
|
-
* detector
|
|
3969
|
+
/** Full-run repeated-call view (total occurrences + window) — catches a loop the online consecutive
|
|
3970
|
+
* detector interleaves past. */
|
|
3880
3971
|
readonly stuckLoop: Awaited<ReturnType<typeof stuckLoopView>>;
|
|
3881
3972
|
/** Wasted-vs-total tool-call ratio for the run. */
|
|
3882
3973
|
readonly toolWaste: Awaited<ReturnType<typeof toolWasteView>>;
|
|
3883
3974
|
}
|
|
3884
|
-
|
|
3885
|
-
|
|
3886
|
-
observeToolStep(step: RecordedToolStep): void;
|
|
3887
|
-
/** Replay the captured steps as agent-eval spans and run the batch analyzers over them. */
|
|
3888
|
-
analyze(): Promise<TrajectoryAnalysis>;
|
|
3889
|
-
reset(): void;
|
|
3890
|
-
}
|
|
3891
|
-
declare function createTrajectoryRecorder(runId: string, now?: () => number): TrajectoryRecorder;
|
|
3975
|
+
/** Collect the source's spans and run the agent-eval batch analyzers over them under one `runId`. */
|
|
3976
|
+
declare function analyzeTrace(source: TraceSource, runId?: string): Promise<TrajectoryAnalysis>;
|
|
3892
3977
|
|
|
3893
3978
|
/**
|
|
3894
3979
|
* createVerifierEnvironment — ANY checkable task as an `Environment`, no tool surface
|
|
@@ -3985,4 +4070,4 @@ declare function runInWorkspace<T>(ws: Workspace, body: (cwd: string) => Promise
|
|
|
3985
4070
|
commitOnInvalid?: boolean;
|
|
3986
4071
|
}): Promise<WorkspaceRun<T>>;
|
|
3987
4072
|
|
|
3988
|
-
export { Agent, AgentRunSpec, AgentSpec, type AgenticOptions, type AgenticRunResult, type AgenticSurface, type AgenticTask, type AgenticTool, type AnytimeReport, type AnytimeStrategySummary, type AnytimeTaskCurve, type ArtifactHandle, type AssertTraceDerivedFindings, type AuditIntentInput, type AuditIntentOptions, type AuthorStrategyOptions, type AuthoredCoderHarness, type AuthoredProfile, type AuthoredStrategy, type BenchmarkCell, type BenchmarkConfig, type BenchmarkLift, type BenchmarkReport, type BenchmarkStrategySummary, type BenchmarkTaskRow, type BridgeSeam, Budget, type BudgetPool, type BudgetReadout, type ChampionPick, type ChampionPolicy, type CheckpointCapableBox, type CliSeam, type CliWorktreeSeam, type CoderDeliverableOptions, type CoderWinnerStrategy, type CombinatorShape, type CompletionAnalyst, type CompletionEvidence, type CompletionPolicy, type CompletionVerdict, type CoordinationDriverOptions, type CoordinationMcpHandle, type Corpus, type CorpusFilter, type CorpusRecord, type CreateScopeAnalystOptions, type CriuCapableClient, type DefinePersona, type DefinePersonaInput, type Deliverable, type DeliverableSpec, type
|
|
4073
|
+
export { Agent, AgentRunSpec, AgentSpec, type AgenticOptions, type AgenticRunResult, type AgenticSurface, type AgenticTask, type AgenticTool, type AnytimeReport, type AnytimeStrategySummary, type AnytimeTaskCurve, type ArtifactHandle, type AssertTraceDerivedFindings, type AuditIntentInput, type AuditIntentOptions, type AuthorStrategyOptions, type AuthoredCoderHarness, type AuthoredProfile, type AuthoredStrategy, type BenchmarkCell, type BenchmarkConfig, type BenchmarkLift, type BenchmarkReport, type BenchmarkStrategySummary, type BenchmarkTaskRow, type BridgeSeam, Budget, type BudgetPool, type BudgetReadout, type ChampionPick, type ChampionPolicy, type CheckpointCapableBox, type CliSeam, type CliWorktreeSeam, type CoderDeliverableOptions, type CoderWinnerStrategy, type CombinatorShape, type CompletionAnalyst, type CompletionEvidence, type CompletionPolicy, type CompletionVerdict, type CoordinationDriverOptions, type CoordinationMcpHandle, type Corpus, type CorpusFilter, type CorpusRecord, type CreateScopeAnalystOptions, type CriuCapableClient, type DefinePersona, type DefinePersonaInput, type Deliverable, type DeliverableSpec, type DriverChat, type DriverMessage, type DriverToolCall, type DriverTurn, type Environment, type EqualKArm, type EqualKOnCost, type EqualKOnCostOptions, type EqualKVerdict, type EvolutionArchiveNode, type EvolutionAuthor, type EvolutionBandInfo, type EvolutionCandidate, type EvolutionGeneration, type EvolutionReport, ExecCtx, Executor, type ExecutorConfig, ExecutorFactory, ExecutorRegistry, type Fanout, type FanoutOptions, type FanoutSynthesis, type FanoutWinnerSelector, FileCorpus, FileResultBlobStore, FileSpawnJournal, type FlatWidenGate, type ForkCapableBox, type GitWorkspaceOptions, type HarvestCorpusOptions, type HarvestFailure, type HarvestReport, InMemoryCorpus, InMemoryResultBlobStore, InMemorySpawnJournal, type Inbox, type InboxMessage, type IntentAudit, Iteration, type LoopDispatchOptions, type LoopOptionsForDispatch, LoopResult, type LoopShape, LoopTokenUsage, type LoopUntil, type LoopUntilSpec, type LoopUntilState, type McpEndpoint, type McpEnvironmentOptions, type NestedScopeSeam, NodeId, type Observation, type ObserveInput, type ObserveOptions, type OpenSandboxRunOptions, type Outcome, type Panel, type PanelJudge, type PanelSpec, type PanelVerdict, type Persona, type PersonaContext, type PersonaExecutors, type Pipeline, type PipelineStage, type PromotionGateOptions, type PromotionVerdict, type RegistryAnalyzeProjection, type RenderCorpusToInstructions, type RenderCorpusToInstructionsOptions, type ReservationTicket, ResultBlobStore, RootHandle, RouterConfig, type RouterSeam, type RouterToolsSeam, type RunAgenticOptions, RunLoopOptions, type RunPersonified, type RunPersonifiedOptions, type SandboxCapabilities, SandboxClient, type SandboxLineage, type SandboxLineageHandle, type SandboxRun, type SandboxSeam, Scope, type ScopeAnalyst, type ScopeAnalyzeInput, type ScopeWidenGate, type SessionCapableBox, type SessionMessageLike, type SessionTraceBox, Settled, type ShapeBudget, type ShapeContext, type ShapeRegistry, type Shell, type ShotPersona, type ShotSpec, SpawnEvent, SpawnJournal, Spend, type SteerContext, type Strategy, type StrategyCtx, type StrategyEvolutionConfig, type StrategyResult, SupervisedResult, Supervisor, type SurfaceScore, type ToolPartDecoder, ToolSpec, type ToolStepInput, type TraceSource, type TrajectoryAnalysis, type TrajectoryNode, type TrajectoryReport, type TrajectoryReportFn, type TrajectoryReportOptions, TreeView, type TurnResult, UsageEvent, type UsageSink, type VerifierEnvironmentOptions, type Verify, type VerifySpec, type WatchTraceOptions, type WaterfallCollector, type WaterfallReport, type WaterfallSpan, type Widen, type WidenDecision, type WidenLineage, type WidenSpec, type Workspace, type WorkspaceCommit, type WorkspaceRun, type WorktreeCliExecutorOptions, type WorktreeCoderFanoutOptions, type WorktreeCommandResult, type WorktreePatchArtifact, acquireSandbox, adaptiveRefine, analyzeTrace, anytimeReport, asAuthoredProfile, assertStrategyContract, assertTraceDerivedFindings, auditIntent, authorStrategy, authoredWorker, breadthDriver, buildSteerContext, builtinShapes, cliWorktreeExecutor, coderDeliverable, completionAuthorizes, contentAddress, coordinationDriverAgent, createBudgetPool, createExecutor, createExecutorRegistry, createInbox, createMcpEnvironment, createPartsTraceSource, createPushTraceSource, createRootHandle, createSandboxLineage, createScope, createScopeAnalyst, createShapeRegistry, createSupervisor, createVerifierEnvironment, createWaterfallCollector, createWorktreeCliExecutor, decodeAnthropicPart, decodeOpenAiPart, decodeOpencodePart, decodeToolPart, defaultAnalystInstruction, defaultAuditorInstruction, defaultToolDetectors, definePersona, defineStrategy, depthDriver, deterministicCompletion, discriminatingMeans, driverChild, driverExecutorFactory, driverRuntime, equalKOnCost, extractLlmCallEvent, fanout, flatWidenGate, gateOnDeliverable, gitWorkspace, harvestCorpus, inlineSandboxClient, isDriverSpec, jjWorkspace, localShell, loopDispatch, loopUntil, mapSandboxEvent, materializeTreeView, nestedScopeSeamKey, observe, openSandboxRun, panel, pickChampion, pipeline, printBenchmarkReport, probeSandboxCapabilities, promotionGate, refine, registerShape, registryScopeAnalyst, renderAnytimeTable, renderCorpusToInstructions, renderReport, replaySpawnTree, reportLoopUsage, routerDriverChat, runAgentic, runBenchmark, runInWorkspace, runPersonified, runStrategyEvolution, sample, sampleThenRefine, sandboxSessionTraceSource, selectChampion, sentinelCompletion, serveCoordinationMcp, settledToIteration, spendFromUsageEvents, stopSentinel, strategyAuthorContract, supervisorSkill, toToolSpan, toolPartDecoders, trajectoryReport, verify, watchTrace, widen, withDriverExecutor, worktreeCoderFanout };
|