@stigmer/runner 3.0.8-dev.20260613085218 → 3.0.9-dev.20260615145121

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 (106) hide show
  1. package/dist/.build-fingerprint +1 -1
  2. package/dist/activities/execute-cursor/hook-script.d.ts +23 -12
  3. package/dist/activities/execute-cursor/hook-script.js +85 -51
  4. package/dist/activities/execute-cursor/hook-script.js.map +1 -1
  5. package/dist/activities/execute-cursor/index.js +220 -79
  6. package/dist/activities/execute-cursor/index.js.map +1 -1
  7. package/dist/activities/execute-cursor/message-translator.d.ts +51 -9
  8. package/dist/activities/execute-cursor/message-translator.js +146 -20
  9. package/dist/activities/execute-cursor/message-translator.js.map +1 -1
  10. package/dist/activities/execute-cursor/persist-decision.d.ts +42 -0
  11. package/dist/activities/execute-cursor/persist-decision.js +30 -0
  12. package/dist/activities/execute-cursor/persist-decision.js.map +1 -0
  13. package/dist/activities/execute-cursor/prompt-builder.d.ts +25 -0
  14. package/dist/activities/execute-cursor/prompt-builder.js +54 -0
  15. package/dist/activities/execute-cursor/prompt-builder.js.map +1 -1
  16. package/dist/activities/execute-cursor/workspace-setup.d.ts +8 -2
  17. package/dist/activities/execute-cursor/workspace-setup.js +62 -30
  18. package/dist/activities/execute-cursor/workspace-setup.js.map +1 -1
  19. package/dist/activities/execute-deep-agent/index.js +15 -5
  20. package/dist/activities/execute-deep-agent/index.js.map +1 -1
  21. package/dist/activities/execute-deep-agent/status-builder-shared.d.ts +0 -1
  22. package/dist/activities/execute-deep-agent/status-builder-shared.js +32 -8
  23. package/dist/activities/execute-deep-agent/status-builder-shared.js.map +1 -1
  24. package/dist/activities/execute-deep-agent/status-builder.js +4 -5
  25. package/dist/activities/execute-deep-agent/status-builder.js.map +1 -1
  26. package/dist/activities/execute-deep-agent/streaming-v3.js +4 -5
  27. package/dist/activities/execute-deep-agent/streaming-v3.js.map +1 -1
  28. package/dist/activities/execute-deep-agent/streaming.d.ts +9 -1
  29. package/dist/activities/execute-deep-agent/streaming.js +4 -5
  30. package/dist/activities/execute-deep-agent/streaming.js.map +1 -1
  31. package/dist/activities/execute-deep-agent/subagent-tracker.js +4 -5
  32. package/dist/activities/execute-deep-agent/subagent-tracker.js.map +1 -1
  33. package/dist/activities/execute-deep-agent/v3-status-builder.js +6 -5
  34. package/dist/activities/execute-deep-agent/v3-status-builder.js.map +1 -1
  35. package/dist/config.d.ts +21 -0
  36. package/dist/config.js +12 -0
  37. package/dist/config.js.map +1 -1
  38. package/dist/in-flight.d.ts +35 -0
  39. package/dist/in-flight.js +61 -0
  40. package/dist/in-flight.js.map +1 -0
  41. package/dist/runner-manager.d.ts +2 -0
  42. package/dist/runner-manager.js +90 -29
  43. package/dist/runner-manager.js.map +1 -1
  44. package/dist/runner.d.ts +2 -0
  45. package/dist/runner.js +2 -0
  46. package/dist/runner.js.map +1 -1
  47. package/dist/shared/grpc-retry.d.ts +9 -20
  48. package/dist/shared/grpc-retry.js +9 -52
  49. package/dist/shared/grpc-retry.js.map +1 -1
  50. package/dist/shared/stall-watchdog.d.ts +68 -0
  51. package/dist/shared/stall-watchdog.js +102 -0
  52. package/dist/shared/stall-watchdog.js.map +1 -0
  53. package/dist/shared/status-offload.d.ts +84 -0
  54. package/dist/shared/status-offload.js +292 -0
  55. package/dist/shared/status-offload.js.map +1 -0
  56. package/dist/shared/status.d.ts +34 -3
  57. package/dist/shared/status.js +102 -9
  58. package/dist/shared/status.js.map +1 -1
  59. package/dist/{activities/execute-deep-agent → shared}/streaming-scheduler.d.ts +4 -0
  60. package/dist/{activities/execute-deep-agent → shared}/streaming-scheduler.js +4 -0
  61. package/dist/shared/streaming-scheduler.js.map +1 -0
  62. package/package.json +2 -2
  63. package/src/__tests__/config.test.ts +8 -0
  64. package/src/__tests__/in-flight.test.ts +84 -0
  65. package/src/activities/__tests__/classify-tool-approvals.test.ts +1 -0
  66. package/src/activities/__tests__/discover-mcp-server.test.ts +1 -0
  67. package/src/activities/execute-cursor/__tests__/build-prompt.test.ts +74 -0
  68. package/src/activities/execute-cursor/__tests__/hook-script.test.ts +90 -15
  69. package/src/activities/execute-cursor/__tests__/message-translator.test.ts +124 -12
  70. package/src/activities/execute-cursor/__tests__/persist-decision.test.ts +99 -0
  71. package/src/activities/execute-cursor/__tests__/tool-result-image.test.ts +244 -0
  72. package/src/activities/execute-cursor/__tests__/workspace-setup.test.ts +53 -4
  73. package/src/activities/execute-cursor/hook-script.ts +85 -51
  74. package/src/activities/execute-cursor/index.ts +187 -38
  75. package/src/activities/execute-cursor/message-translator.ts +146 -20
  76. package/src/activities/execute-cursor/persist-decision.ts +54 -0
  77. package/src/activities/execute-cursor/prompt-builder.ts +59 -0
  78. package/src/activities/execute-cursor/workspace-setup.ts +76 -44
  79. package/src/activities/execute-deep-agent/__tests__/index.test.ts +1 -0
  80. package/src/activities/execute-deep-agent/__tests__/status-builder-shared.test.ts +66 -0
  81. package/src/activities/execute-deep-agent/__tests__/status-builder.test.ts +6 -3
  82. package/src/activities/execute-deep-agent/__tests__/streaming-v3.test.ts +70 -0
  83. package/src/activities/execute-deep-agent/index.ts +17 -5
  84. package/src/activities/execute-deep-agent/status-builder-shared.ts +27 -5
  85. package/src/activities/execute-deep-agent/status-builder.ts +3 -5
  86. package/src/activities/execute-deep-agent/streaming-v3.ts +5 -5
  87. package/src/activities/execute-deep-agent/streaming.ts +14 -5
  88. package/src/activities/execute-deep-agent/subagent-tracker.ts +4 -5
  89. package/src/activities/execute-deep-agent/v3-status-builder.ts +5 -5
  90. package/src/config.ts +27 -0
  91. package/src/in-flight.ts +71 -0
  92. package/src/runner-manager.ts +127 -33
  93. package/src/runner.ts +6 -0
  94. package/src/shared/__tests__/artifact-storage.test.ts +1 -0
  95. package/src/shared/__tests__/grpc-retry-extended.test.ts +6 -144
  96. package/src/shared/__tests__/grpc-retry.test.ts +5 -134
  97. package/src/shared/__tests__/stall-watchdog.test.ts +193 -0
  98. package/src/shared/__tests__/status-offload.test.ts +256 -0
  99. package/src/shared/__tests__/status.test.ts +199 -0
  100. package/src/shared/grpc-retry.ts +9 -72
  101. package/src/shared/stall-watchdog.ts +122 -0
  102. package/src/shared/status-offload.ts +342 -0
  103. package/src/shared/status.ts +142 -8
  104. package/src/{activities/execute-deep-agent → shared}/streaming-scheduler.ts +4 -0
  105. package/dist/activities/execute-deep-agent/streaming-scheduler.js.map +0 -1
  106. /package/src/{activities/execute-deep-agent → shared}/__tests__/streaming-scheduler.test.ts +0 -0
@@ -19,28 +19,162 @@ import {
19
19
  ExecutionPhase,
20
20
  } from "@stigmer/protos/ai/stigmer/agentic/agentexecution/v1/enum_pb";
21
21
  import type { StigmerClient } from "../client/stigmer-client.js";
22
+ import {
23
+ offloadOversizedToolOutputs,
24
+ enforceStatusSizeLimit,
25
+ STATUS_PAYLOAD_HARD_LIMIT_BYTES,
26
+ type ToolOutputOffloadContext,
27
+ } from "./status-offload.js";
28
+ import {
29
+ isRetryableError,
30
+ isTerminalError,
31
+ type RetryOptions,
32
+ } from "./grpc-retry.js";
22
33
 
23
34
  export function utcTimestamp(): string {
24
35
  return new Date().toISOString().replace("+00:00", "Z");
25
36
  }
26
37
 
38
+ /**
39
+ * A connect-rpc `resource_exhausted` (gRPC code 8) — the payload exceeded the
40
+ * server's receive cap. Matched by code and, defensively, by the message buf's
41
+ * size check emits ("exceeds maximum size").
42
+ */
43
+ function isPayloadTooLarge(err: unknown): boolean {
44
+ if (typeof err === "object" && err !== null && "code" in err) {
45
+ if ((err as { code?: unknown }).code === 8) return true;
46
+ }
47
+ const msg = err instanceof Error ? err.message : String(err);
48
+ return /resource_exhausted|exceeds maximum size/i.test(msg);
49
+ }
50
+
51
+ /** Options controlling how a status payload is bounded and persisted. */
52
+ export interface PersistStatusOptions {
53
+ /**
54
+ * When supplied, oversized tool outputs (a screenshot's base64, a multi-MB
55
+ * dump) are spilled to artifact storage before persisting — the only path
56
+ * that populates a renderable `ToolCallOutputRef`. When omitted, the
57
+ * aggregate size backstop still runs, so the payload is bounded regardless.
58
+ */
59
+ readonly offload?: ToolOutputOffloadContext;
60
+ /**
61
+ * Transient-error backoff policy. Omitted fields fall back to
62
+ * {@link DEFAULT_PERSIST_RETRY}; an omitted `retry` uses it entirely.
63
+ */
64
+ readonly retry?: RetryOptions;
65
+ }
66
+
67
+ /**
68
+ * Default transient-retry policy: three attempts with exponential backoff.
69
+ * Applied to every persist so a transient drop (server restart, brief
70
+ * unavailability) recovers silently instead of dropping a status update.
71
+ */
72
+ const DEFAULT_PERSIST_RETRY = {
73
+ baseDelayMs: 100,
74
+ backoffFactor: 2,
75
+ maxRetries: 3,
76
+ } as const;
77
+
78
+ function defaultDelay(ms: number): Promise<void> {
79
+ return new Promise((resolve) => setTimeout(resolve, ms));
80
+ }
81
+
27
82
  /**
28
83
  * Persist execution status via gRPC. Returns the control signal from the
29
- * backend (e.g. STOP to abort the execution). Errors are logged and
30
- * swallowed — a failed status update should not crash the activity.
84
+ * backend (e.g. STOP to abort the execution).
85
+ *
86
+ * This is the single, unforgeable chokepoint for status persistence across
87
+ * every harness (streaming and terminal, native and Cursor). Routing all
88
+ * persistence through here guarantees three invariants that must never be
89
+ * skipped by an individual call site:
90
+ *
91
+ * 1. Size bounding — when `offload` is supplied, oversized tool outputs are
92
+ * spilled to artifact storage first (the path that makes screenshots
93
+ * render); an aggregate backstop then runs unconditionally so callers
94
+ * without storage are still protected from the 4 MiB gRPC cap.
95
+ * 2. Transient resilience — `Unavailable`/`DeadlineExceeded` back off and
96
+ * retry (default three attempts); terminal errors never retry.
97
+ * 3. Oversize recovery — a `resource_exhausted` rejection is never silently
98
+ * swallowed: we hard-elide and retry once, so an oversized payload can
99
+ * never freeze persistence invisibly (the bug that stalled the UI).
31
100
  */
32
101
  export async function persistStatus(
33
102
  client: StigmerClient,
34
103
  executionId: string,
35
104
  status: AgentExecutionStatus,
105
+ options: PersistStatusOptions = {},
36
106
  ): Promise<ExecutionControlSignal> {
37
- try {
38
- const response = await client.updateStatus(executionId, status);
39
- return response.signal;
40
- } catch (err) {
41
- console.error(`Failed to persist status for ${executionId}:`, err);
42
- return ExecutionControlSignal.UNSPECIFIED;
107
+ const { offload, retry } = options;
108
+
109
+ if (offload) {
110
+ try {
111
+ await offloadOversizedToolOutputs(status, offload);
112
+ } catch (err) {
113
+ console.warn(`[persistStatus] ${executionId}: tool-output offload failed (non-fatal): ${err}`);
114
+ }
43
115
  }
116
+ if (enforceStatusSizeLimit(status)) {
117
+ console.warn(
118
+ `[persistStatus] ${executionId}: status exceeded the soft size limit; ` +
119
+ `elided oversized inline fields to fit under the gRPC cap`,
120
+ );
121
+ }
122
+
123
+ const baseDelayMs = retry?.baseDelayMs ?? DEFAULT_PERSIST_RETRY.baseDelayMs;
124
+ const backoffFactor = retry?.backoffFactor ?? DEFAULT_PERSIST_RETRY.backoffFactor;
125
+ const maxRetries = retry?.maxRetries ?? DEFAULT_PERSIST_RETRY.maxRetries;
126
+ const delay = retry?.delayFn ?? defaultDelay;
127
+
128
+ for (let attempt = 0; attempt <= maxRetries; attempt++) {
129
+ try {
130
+ const response = await client.updateStatus(executionId, status);
131
+ return response.signal;
132
+ } catch (err) {
133
+ // Oversize rejection is recoverable exactly once and out-of-band from the
134
+ // transient budget: hard-elide the largest inline fields and retry now.
135
+ if (isPayloadTooLarge(err)) {
136
+ console.error(
137
+ `[persistStatus] ${executionId}: payload rejected as too large despite ` +
138
+ `the size guard; hard-eliding and retrying once`,
139
+ );
140
+ enforceStatusSizeLimit(status, STATUS_PAYLOAD_HARD_LIMIT_BYTES);
141
+ try {
142
+ const response = await client.updateStatus(executionId, status);
143
+ return response.signal;
144
+ } catch (retryErr) {
145
+ console.error(`[persistStatus] ${executionId}: still failing after hard elide:`, retryErr);
146
+ return ExecutionControlSignal.UNSPECIFIED;
147
+ }
148
+ }
149
+
150
+ // Terminal errors (invalid argument, not found, permission denied) are
151
+ // deterministic — retrying cannot help.
152
+ if (isTerminalError(err)) {
153
+ console.error(
154
+ `[persistStatus] ${executionId}: terminal error persisting status ` +
155
+ `(attempt ${attempt + 1}/${maxRetries + 1}): ${err}`,
156
+ );
157
+ return ExecutionControlSignal.UNSPECIFIED;
158
+ }
159
+
160
+ // Transient errors back off and retry until the budget is exhausted.
161
+ if (isRetryableError(err) && attempt < maxRetries) {
162
+ const delayMs = baseDelayMs * Math.pow(backoffFactor, attempt);
163
+ console.warn(
164
+ `[persistStatus] ${executionId}: retryable error ` +
165
+ `(attempt ${attempt + 1}/${maxRetries + 1}, retry in ${delayMs}ms): ${err}`,
166
+ );
167
+ await delay(delayMs);
168
+ continue;
169
+ }
170
+
171
+ // Non-retryable, non-terminal (e.g. a bare Error), or retries exhausted.
172
+ console.error(`Failed to persist status for ${executionId}:`, err);
173
+ return ExecutionControlSignal.UNSPECIFIED;
174
+ }
175
+ }
176
+
177
+ return ExecutionControlSignal.UNSPECIFIED;
44
178
  }
45
179
 
46
180
  /**
@@ -13,6 +13,10 @@
13
13
  * OR firstUpdate
14
14
  *
15
15
  * Ported from Python StreamingUpdateScheduler.
16
+ *
17
+ * Shared infrastructure: both the native deep-agent harness and the Cursor
18
+ * harness drive their incremental status persistence through this scheduler so
19
+ * the live stream cadence is identical across harnesses.
16
20
  */
17
21
 
18
22
  export enum UpdateReason {
@@ -1 +0,0 @@
1
- {"version":3,"file":"streaming-scheduler.js","sourceRoot":"","sources":["../../../src/activities/execute-deep-agent/streaming-scheduler.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,MAAM,CAAN,IAAY,YAMX;AAND,WAAY,YAAY;IACtB,iDAAiC,CAAA;IACjC,qDAAqC,CAAA;IACrC,uCAAuB,CAAA;IACvB,6CAA6B,CAAA;IAC7B,6BAAa,CAAA;AACf,CAAC,EANW,YAAY,KAAZ,YAAY,QAMvB;AAWD,MAAM,cAAc,GAAoB;IACtC,aAAa,EAAE,GAAG;IAClB,aAAa,EAAE,IAAI;IACnB,cAAc,EAAE,EAAE;CACnB,CAAC;AAEF;;;GAGG;AACH,MAAM,UAAU,mBAAmB;IACjC,MAAM,aAAa,GAAG,gBAAgB,CACpC,OAAO,CAAC,GAAG,CAAC,yBAAyB,EACrC,cAAc,CAAC,aAAa,EAC5B,2BAA2B,CAC5B,CAAC;IACF,IAAI,aAAa,GAAG,gBAAgB,CAClC,OAAO,CAAC,GAAG,CAAC,yBAAyB,EACrC,cAAc,CAAC,aAAa,EAC5B,2BAA2B,CAC5B,CAAC;IACF,MAAM,cAAc,GAAG,gBAAgB,CACrC,OAAO,CAAC,GAAG,CAAC,yBAAyB,EACrC,cAAc,CAAC,cAAc,EAC7B,2BAA2B,CAC5B,CAAC;IAEF,IAAI,aAAa,GAAG,aAAa,EAAE,CAAC;QAClC,OAAO,CAAC,IAAI,CACV,8BAA8B,aAAa,MAAM;YACjD,8BAA8B,aAAa,8BAA8B,CAC1E,CAAC;QACF,aAAa,GAAG,aAAa,CAAC;IAChC,CAAC;IAED,OAAO,EAAE,aAAa,EAAE,aAAa,EAAE,cAAc,EAAE,CAAC;AAC1D,CAAC;AAED,SAAS,gBAAgB,CACvB,GAAuB,EACvB,QAAgB,EAChB,OAAe;IAEf,IAAI,CAAC,GAAG;QAAE,OAAO,QAAQ,CAAC;IAC1B,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;IAC3B,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,MAAM,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,MAAM,CAAC,EAAE,CAAC;QACzE,OAAO,CAAC,IAAI,CACV,WAAW,OAAO,KAAK,GAAG,qBAAqB,QAAQ,EAAE,CAC1D,CAAC;QACF,OAAO,QAAQ,CAAC;IAClB,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,OAAO,wBAAwB;IAC1B,MAAM,CAAkB;IAEzB,cAAc,CAAS;IACvB,gBAAgB,GAAG,CAAC,CAAC;IACrB,UAAU,GAAiB,YAAY,CAAC,IAAI,CAAC;IAC7C,UAAU,GAAG,IAAI,CAAC;IAE1B;;;OAGG;IACH,YAAY,MAAwB,EAAE,KAAc;QAClD,IAAI,CAAC,MAAM,GAAG,MAAM,IAAI,cAAc,CAAC;QACvC,IAAI,CAAC,cAAc,GAAG,KAAK,IAAI,WAAW,CAAC,GAAG,EAAE,CAAC;IACnD,CAAC;IAED;;;;;OAKG;IACH,gBAAgB,CAAC,eAAuB,EAAE,KAAc;QACtD,MAAM,GAAG,GAAG,KAAK,IAAI,WAAW,CAAC,GAAG,EAAE,CAAC;QACvC,MAAM,eAAe,GAAG,GAAG,GAAG,IAAI,CAAC,cAAc,CAAC;QAClD,MAAM,eAAe,GAAG,eAAe,GAAG,IAAI,CAAC,gBAAgB,CAAC;QAEhE,IAAI,IAAI,CAAC,UAAU,IAAI,eAAe,IAAI,CAAC,EAAE,CAAC;YAC5C,IAAI,CAAC,UAAU,GAAG,YAAY,CAAC,YAAY,CAAC;YAC5C,OAAO,IAAI,CAAC;QACd,CAAC;QAED,IAAI,eAAe,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,IAAI,eAAe,IAAI,CAAC,EAAE,CAAC;YACzE,IAAI,CAAC,UAAU,GAAG,YAAY,CAAC,cAAc,CAAC;YAC9C,OAAO,IAAI,CAAC;QACd,CAAC;QAED,IAAI,eAAe,IAAI,IAAI,CAAC,MAAM,CAAC,cAAc,EAAE,CAAC;YAClD,IAAI,CAAC,UAAU,GAAG,YAAY,CAAC,gBAAgB,CAAC;YAChD,OAAO,IAAI,CAAC;QACd,CAAC;QAED,IAAI,eAAe,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,EAAE,CAAC;YACjD,IAAI,CAAC,UAAU,GAAG,YAAY,CAAC,SAAS,CAAC;YACzC,OAAO,IAAI,CAAC;QACd,CAAC;QAED,IAAI,CAAC,UAAU,GAAG,YAAY,CAAC,IAAI,CAAC;QACpC,OAAO,KAAK,CAAC;IACf,CAAC;IAED,qDAAqD;IACrD,cAAc,CAAC,eAAuB,EAAE,KAAc;QACpD,IAAI,CAAC,cAAc,GAAG,KAAK,IAAI,WAAW,CAAC,GAAG,EAAE,CAAC;QACjD,IAAI,CAAC,gBAAgB,GAAG,eAAe,CAAC;QACxC,IAAI,CAAC,UAAU,GAAG,KAAK,CAAC;IAC1B,CAAC;IAED,IAAI,YAAY;QACd,OAAO,IAAI,CAAC,UAAU,CAAC;IACzB,CAAC;IAED,qBAAqB,CAAC,KAAc;QAClC,OAAO,CAAC,KAAK,IAAI,WAAW,CAAC,GAAG,EAAE,CAAC,GAAG,IAAI,CAAC,cAAc,CAAC;IAC5D,CAAC;IAED,qBAAqB,CAAC,eAAuB;QAC3C,OAAO,eAAe,GAAG,IAAI,CAAC,gBAAgB,CAAC;IACjD,CAAC;CACF"}