@tangle-network/agent-runtime 0.199.0 → 0.200.0
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/{activation-c3um4SY1.d.ts → activation-BPxs-Iu2.d.ts} +2 -2
- package/dist/{activation-zQWmiiPg.js → activation-Ct-BTTzF.js} +2 -2
- package/dist/{activation-zQWmiiPg.js.map → activation-Ct-BTTzF.js.map} +1 -1
- package/dist/agent.d.ts +1 -1
- package/dist/agent.js +2 -3
- package/dist/agent.js.map +1 -1
- package/dist/analyst-loop-DZw5QWT9.js +176 -0
- package/dist/analyst-loop-DZw5QWT9.js.map +1 -0
- package/dist/analyst-loop.d.ts +1 -11
- package/dist/analyst-loop.js +2 -2
- package/dist/{coordination-driver-vWb7kViA.js → coordination-driver-qdAwriPV.js} +8 -5
- package/dist/{coordination-driver-vWb7kViA.js.map → coordination-driver-qdAwriPV.js.map} +1 -1
- package/dist/{delegate-Ch93bpH5.js → delegate-BWHG-zZW.js} +2 -2
- package/dist/{delegate-Ch93bpH5.js.map → delegate-BWHG-zZW.js.map} +1 -1
- package/dist/durable.d.ts +2 -2
- package/dist/durable.js +2 -2
- package/dist/{graph-C6pT08K-.js → graph-cCqKhLqz.js} +3 -3
- package/dist/{graph-C6pT08K-.js.map → graph-cCqKhLqz.js.map} +1 -1
- package/dist/{improvement-cycle-CxZKIbLD.js → improvement-cycle-yia1ST3c.js} +4 -4
- package/dist/{improvement-cycle-CxZKIbLD.js.map → improvement-cycle-yia1ST3c.js.map} +1 -1
- package/dist/{index-CMTUgh-T.d.ts → index-Br191WbE.d.ts} +190 -142
- package/dist/index.d.ts +4 -4
- package/dist/index.js +20 -46
- package/dist/index.js.map +1 -1
- package/dist/intelligence.d.ts +13 -104
- package/dist/intelligence.js +7 -406
- package/dist/intelligence.js.map +1 -1
- package/dist/kernel.d.ts +3 -3
- package/dist/kernel.js +8 -9
- package/dist/{loop-runner-bin-CAf1OQot.d.ts → loop-runner-bin-BNRdsDOn.d.ts} +3 -3
- package/dist/{loop-runner-bin-SNrh585k.js → loop-runner-bin-jQ8hXO9J.js} +4 -4
- package/dist/{loop-runner-bin-SNrh585k.js.map → loop-runner-bin-jQ8hXO9J.js.map} +1 -1
- package/dist/loop-runner-bin.d.ts +1 -1
- package/dist/loop-runner-bin.js +1 -1
- package/dist/mcp/bin.js +3 -3
- package/dist/mcp/index.d.ts +2 -2
- package/dist/mcp/index.js +4 -4
- package/dist/{provision-supervisor-D5YSCDVD.js → provision-supervisor-BA2-GPth.js} +3 -3
- package/dist/{provision-supervisor-D5YSCDVD.js.map → provision-supervisor-BA2-GPth.js.map} +1 -1
- package/dist/{redact-Dv8WcCKy.js → redact-h-oaw11Q.js} +13093 -11221
- package/dist/redact-h-oaw11Q.js.map +1 -0
- package/dist/{runtime-BN16GbTA.js → runtime-DeqdBVeC.js} +8 -9
- package/dist/{runtime-BN16GbTA.js.map → runtime-DeqdBVeC.js.map} +1 -1
- package/dist/{server-CbNb_N1x.js → server-U_k9NUbk.js} +3 -3
- package/dist/{server-CbNb_N1x.js.map → server-U_k9NUbk.js.map} +1 -1
- package/dist/{stream-agent-turn-CLOQr497.d.ts → stream-agent-turn-CJWthifS.d.ts} +42 -13
- package/dist/{structural-rollout-DANCelu6.js → structural-rollout-BcMmrpv4.js} +2 -3
- package/dist/{structural-rollout-DANCelu6.js.map → structural-rollout-BcMmrpv4.js.map} +1 -1
- package/dist/{supervise-BGNXB8No.js → supervise-DmFO50N6.js} +497 -90
- package/dist/supervise-DmFO50N6.js.map +1 -0
- package/dist/testing.d.ts +2 -2
- package/dist/testing.js +12 -12
- package/dist/tui/index.d.ts +1 -1
- package/dist/tui/index.js +1 -1
- package/package.json +3 -3
- package/dist/analyst-loop-BknOQUW5.js +0 -546
- package/dist/analyst-loop-BknOQUW5.js.map +0 -1
- package/dist/redact-Dv8WcCKy.js.map +0 -1
- package/dist/sandbox-events-DbC2WKKS.js +0 -929
- package/dist/sandbox-events-DbC2WKKS.js.map +0 -1
- package/dist/supervise-BGNXB8No.js.map +0 -1
package/dist/testing.d.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { An as RunGraphTestOptions,
|
|
2
|
-
import { Tn as ToolLoopChat, wn as ToolLoopCallContext } from "./stream-agent-turn-
|
|
1
|
+
import { An as RunGraphTestOptions, Bo as SuperviseTestOptions, Ho as superviseWithTestBrain, Pn as runGraphWithTestBrain, _s as supervisorAgentWithTestBrain, ar as DriverAgentOptions, cs as SupervisorAgentTestDeps, or as driverAgent } from "./index-Br191WbE.js";
|
|
2
|
+
import { Tn as ToolLoopChat, wn as ToolLoopCallContext } from "./stream-agent-turn-CJWthifS.js";
|
|
3
3
|
import { AgentImprovementProposal, AgentProfile, AgentProfileImprovementMeasuredComparison, SandboxSizePreset } from "@tangle-network/agent-interface";
|
|
4
4
|
//#region src/testing/index.d.ts
|
|
5
5
|
/** A proposal produced by Runtime's opaque profile-improvement path. */
|
package/dist/testing.js
CHANGED
|
@@ -1,14 +1,14 @@
|
|
|
1
|
-
import { f as verifyAgentImprovementProposal } from "./improvement-cycle-
|
|
1
|
+
import { f as verifyAgentImprovementProposal } from "./improvement-cycle-yia1ST3c.js";
|
|
2
2
|
import { o as canonicalCandidateDigest$1, u as immutableCandidateValue } from "./protected-redaction-wGo44k2K.js";
|
|
3
3
|
import { M as parseExactAgentProfile, w as applyExactAgentProfileDiff } from "./prepare-CAO1yXov.js";
|
|
4
|
-
import { p as supervisorAgentWithTestBrain, r as superviseWithTestBrain } from "./supervise-
|
|
5
|
-
import { r as driverAgent } from "./coordination-driver-
|
|
6
|
-
import { i as runGraphWithTestBrain } from "./graph-
|
|
4
|
+
import { p as supervisorAgentWithTestBrain, r as superviseWithTestBrain } from "./supervise-DmFO50N6.js";
|
|
5
|
+
import { r as driverAgent } from "./coordination-driver-qdAwriPV.js";
|
|
6
|
+
import { i as runGraphWithTestBrain } from "./graph-cCqKhLqz.js";
|
|
7
7
|
import { SANDBOX_SIZE_PRESET_NAMES } from "@tangle-network/agent-interface";
|
|
8
8
|
//#region src/testing/fixtures/agent-improvement-proposal.json
|
|
9
9
|
var agent_improvement_proposal_default = {
|
|
10
10
|
changedSurfaces: ["prompt"],
|
|
11
|
-
digest: "sha256:
|
|
11
|
+
digest: "sha256:240186a31dc1452a779e85c4b174090e3f8fb84494ac87c5f82e0280598c2bb1",
|
|
12
12
|
evaluation: {
|
|
13
13
|
"decision": {
|
|
14
14
|
"contributingChecks": [
|
|
@@ -4579,7 +4579,7 @@ var agent_improvement_proposal_default = {
|
|
|
4579
4579
|
],
|
|
4580
4580
|
"metadata": {
|
|
4581
4581
|
"fixture": "agent-improvement-proposal",
|
|
4582
|
-
"runtimeVersion": "0.
|
|
4582
|
+
"runtimeVersion": "0.200.0"
|
|
4583
4583
|
},
|
|
4584
4584
|
"objectives": [
|
|
4585
4585
|
{
|
|
@@ -4690,8 +4690,8 @@ var agent_improvement_proposal_default = {
|
|
|
4690
4690
|
"baselineContentHash": "sha256:5c21ee53e513fc604cb09754e21c392b24a424da0ef37dbf8f1ee4a8a0b08f09",
|
|
4691
4691
|
"candidateContentHash": "sha256:60fcbb1c728194bd51d7d19cb732d1c3f1881dce7e0a6266b41c8b98cfd65693",
|
|
4692
4692
|
"kind": "agent-eval-loop",
|
|
4693
|
-
"recordDigest": "sha256:
|
|
4694
|
-
"runId": "agent-runtime-0.
|
|
4693
|
+
"recordDigest": "sha256:797845421140b7afdfaeef76018260dd0775bdc2f0acb51e4748f6f22233d699",
|
|
4694
|
+
"runId": "agent-runtime-0.200.0-proposal-fixture",
|
|
4695
4695
|
"schema": "agent-candidate-experiment"
|
|
4696
4696
|
}
|
|
4697
4697
|
},
|
|
@@ -4714,13 +4714,13 @@ var agent_improvement_proposal_default = {
|
|
|
4714
4714
|
}],
|
|
4715
4715
|
kind: "agent-improvement-proposal",
|
|
4716
4716
|
proposedAt: "2026-07-10T01:00:00.000Z",
|
|
4717
|
-
runId: "agent-runtime-0.
|
|
4717
|
+
runId: "agent-runtime-0.200.0-proposal-fixture"
|
|
4718
4718
|
};
|
|
4719
4719
|
//#endregion
|
|
4720
4720
|
//#region src/testing/fixtures/agent-profile-improvement-proposal.json
|
|
4721
4721
|
var agent_profile_improvement_proposal_default = {
|
|
4722
4722
|
changedSurfaces: ["prompt", "skills"],
|
|
4723
|
-
digest: "sha256:
|
|
4723
|
+
digest: "sha256:6e4a8ef4eab876ad23c1eac8e76e20f9f58506c25f64591b51d8e041414b8a3e",
|
|
4724
4724
|
evaluation: {
|
|
4725
4725
|
"decision": {
|
|
4726
4726
|
"contributingChecks": [
|
|
@@ -6354,7 +6354,7 @@ var agent_profile_improvement_proposal_default = {
|
|
|
6354
6354
|
],
|
|
6355
6355
|
"metadata": {
|
|
6356
6356
|
"fixture": "agent-profile-improvement-proposal",
|
|
6357
|
-
"runtimeVersion": "0.
|
|
6357
|
+
"runtimeVersion": "0.200.0"
|
|
6358
6358
|
},
|
|
6359
6359
|
"objectives": [
|
|
6360
6360
|
{
|
|
@@ -6465,7 +6465,7 @@ var agent_profile_improvement_proposal_default = {
|
|
|
6465
6465
|
"baselineContentHash": "sha256:21c495a37c418c10bde64fbaa188beddeed31f1f051ea60a6a6582a9ee0db704",
|
|
6466
6466
|
"candidateContentHash": "sha256:103f77bc8481601eef1ad5fe6ba84a40dffabc3a44f421f8c8559121edab84e9",
|
|
6467
6467
|
"kind": "agent-eval-loop",
|
|
6468
|
-
"recordDigest": "sha256:
|
|
6468
|
+
"recordDigest": "sha256:5a209130c625a814b2f0498d629d9927d5d0658649f3852a551ada9a153e8df6",
|
|
6469
6469
|
"runId": "profile-improvement-1",
|
|
6470
6470
|
"schema": "agent-profile-improvement-experiment"
|
|
6471
6471
|
}
|
package/dist/tui/index.d.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { an as SupervisorCleanupReceipt, in as ProvisionedSupervisor, nn as ProvisionSupervisorConnection, on as provisionSupervisor, rn as ProvisionSupervisorRequest } from "../index-
|
|
1
|
+
import { an as SupervisorCleanupReceipt, in as ProvisionedSupervisor, nn as ProvisionSupervisorConnection, on as provisionSupervisor, rn as ProvisionSupervisorRequest } from "../index-Br191WbE.js";
|
|
2
2
|
//#region src/tui/top-app.d.ts
|
|
3
3
|
/**
|
|
4
4
|
* The interactive side of the supervisor-run TUI: a keypress/mouse loop over the frames
|
package/dist/tui/index.js
CHANGED
|
@@ -1,3 +1,3 @@
|
|
|
1
|
-
import { t as provisionSupervisor } from "../provision-supervisor-
|
|
1
|
+
import { t as provisionSupervisor } from "../provision-supervisor-BA2-GPth.js";
|
|
2
2
|
import { a as renderTopFrameWithLayout, i as renderTopFrame, n as runTopApp, r as loadTopSnapshot, t as renderTopOnce } from "../top-app-KgIFpD9H.js";
|
|
3
3
|
export { loadTopSnapshot, provisionSupervisor, renderTopFrame, renderTopFrameWithLayout, renderTopOnce, runTopApp };
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@tangle-network/agent-runtime",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.200.0",
|
|
4
4
|
"description": "Shared task-lifecycle skeleton for agents: a recursive loop kernel for chat turns, one-shot tasks, and multi-attempt loops, with trace capture and eval-gated self-improvement. Domain behavior lives in adapters; scoring and ship-gates in @tangle-network/agent-eval.",
|
|
5
5
|
"homepage": "https://github.com/tangle-network/agent-runtime#readme",
|
|
6
6
|
"repository": {
|
|
@@ -99,7 +99,7 @@
|
|
|
99
99
|
"@biomejs/biome": "^2.5.11",
|
|
100
100
|
"@modelcontextprotocol/sdk": "1.30.0",
|
|
101
101
|
"@tangle-network/agent-eval": ">=0.175.0 <0.176.0",
|
|
102
|
-
"@tangle-network/agent-interface": "^2.
|
|
102
|
+
"@tangle-network/agent-interface": "^2.4.0",
|
|
103
103
|
"@tangle-network/sandbox": ">=0.36.4 <0.38.0",
|
|
104
104
|
"@types/node": "26.4.0",
|
|
105
105
|
"@types/tar-stream": "3.1.4",
|
|
@@ -122,7 +122,7 @@
|
|
|
122
122
|
"license": "MIT",
|
|
123
123
|
"peerDependencies": {
|
|
124
124
|
"@tangle-network/agent-eval": ">=0.175.0 <0.176.0",
|
|
125
|
-
"@tangle-network/agent-interface": "^2.
|
|
125
|
+
"@tangle-network/agent-interface": "^2.4.0",
|
|
126
126
|
"@tangle-network/sandbox": ">=0.36.4 <0.38.0"
|
|
127
127
|
},
|
|
128
128
|
"dependencies": {
|
|
@@ -1,546 +0,0 @@
|
|
|
1
|
-
import { n as AnalystError } from "./errors-DodWX-cb.js";
|
|
2
|
-
import { a as extractLlmCallEvent } from "./sandbox-events-DbC2WKKS.js";
|
|
3
|
-
import { DEFAULT_TRACE_ANALYST_BUDGETS, TRACE_ANALYST_TRUNCATION_MARKER_PREFIX, diffFindings } from "@tangle-network/agent-eval";
|
|
4
|
-
//#region src/analyst-loop/iterations-to-trace-store.ts
|
|
5
|
-
/**
|
|
6
|
-
*
|
|
7
|
-
* The read seam that closes the autonomous loop: project a round's `Iteration[]`
|
|
8
|
-
* (each carrying its raw `SandboxEvent[]`) into an in-memory `TraceAnalysisStore`, the
|
|
9
|
-
* read interface the trace analysts query. `runAnalystLoop` has had zero consumers
|
|
10
|
-
* because nothing turned a loop's iterations into a store — this is that bridge.
|
|
11
|
-
*
|
|
12
|
-
* One iteration → one trace. The iteration is the root AGENT span; each `SandboxEvent`
|
|
13
|
-
* becomes a child span (LLM for llm_call events, TOOL for tool events, else SPAN), so an
|
|
14
|
-
* analyst can walk a shot's trace, cluster its errors, and emit findings the driver
|
|
15
|
-
* steers on. Projection is best-effort over the FLAT SandboxEvent shape (no per-event
|
|
16
|
-
* lineage yet — that's the richer-trace gap); it never fabricates — an errored iteration
|
|
17
|
-
* surfaces a real ERROR span carrying the real message.
|
|
18
|
-
*
|
|
19
|
-
* @experimental
|
|
20
|
-
*/
|
|
21
|
-
const bytesOf = (v) => Buffer.byteLength(JSON.stringify(v) ?? "", "utf8");
|
|
22
|
-
const iso = (ms) => new Date(ms).toISOString();
|
|
23
|
-
/** Normalize volatile tokens out of a status message so semantically identical failures
|
|
24
|
-
* collapse to one signature (digits, hex/uuids, paths, durations → placeholders). */
|
|
25
|
-
function normalizeSignature(message) {
|
|
26
|
-
return message.replace(/0x[0-9a-fA-F]+/g, "HEX").replace(/[0-9a-fA-F]{8}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{4}-[0-9a-fA-F]{12}/g, "UUID").replace(/(\/[\w.-]+){2,}/g, "PATH").replace(/\b\d+(\.\d+)?(ms|s|m|h)\b/g, "DUR").replace(/\b\d+\b/g, "#").replace(/\s+/g, " ").trim().slice(0, 200);
|
|
27
|
-
}
|
|
28
|
-
function spanKindFor(event, agentRunName) {
|
|
29
|
-
const llm = extractLlmCallEvent(event, agentRunName);
|
|
30
|
-
if (llm) return {
|
|
31
|
-
kind: "LLM",
|
|
32
|
-
model: llm.model ?? null,
|
|
33
|
-
tool: null
|
|
34
|
-
};
|
|
35
|
-
const type = String(event?.type ?? "");
|
|
36
|
-
if (/tool/i.test(type)) {
|
|
37
|
-
const d = event?.data;
|
|
38
|
-
return {
|
|
39
|
-
kind: "TOOL",
|
|
40
|
-
model: null,
|
|
41
|
-
tool: typeof d?.name === "string" ? d.name : typeof d?.tool === "string" ? d.tool : type
|
|
42
|
-
};
|
|
43
|
-
}
|
|
44
|
-
return {
|
|
45
|
-
kind: "SPAN",
|
|
46
|
-
model: null,
|
|
47
|
-
tool: null
|
|
48
|
-
};
|
|
49
|
-
}
|
|
50
|
-
function errorMessageOf(event) {
|
|
51
|
-
const type = String(event?.type ?? "");
|
|
52
|
-
const d = event?.data;
|
|
53
|
-
if (/error|fail/i.test(type) || d?.error) {
|
|
54
|
-
const m = d?.error ?? d?.message;
|
|
55
|
-
return typeof m === "string" ? m : `${type} error`;
|
|
56
|
-
}
|
|
57
|
-
}
|
|
58
|
-
/** Project one iteration → one trace: root AGENT span + a child span per event. */
|
|
59
|
-
function projectIteration(iter) {
|
|
60
|
-
const traceId = (iter.events.find((e) => (e?.data)?.sandboxId)?.data)?.sandboxId ?? `iter-${iter.index}`;
|
|
61
|
-
const start = iso(iter.startedAt);
|
|
62
|
-
const end = iso(iter.endedAt || iter.startedAt);
|
|
63
|
-
const durationMs = Math.max(0, (iter.endedAt || iter.startedAt) - iter.startedAt);
|
|
64
|
-
const rootId = `${traceId}:root`;
|
|
65
|
-
const iterErrored = Boolean(iter.error) || iter.verdict?.valid === false;
|
|
66
|
-
const spans = [{
|
|
67
|
-
trace_id: traceId,
|
|
68
|
-
span_id: rootId,
|
|
69
|
-
parent_span_id: null,
|
|
70
|
-
name: iter.agentRunName,
|
|
71
|
-
kind: "AGENT",
|
|
72
|
-
start_time: start,
|
|
73
|
-
end_time: end,
|
|
74
|
-
duration_ms: durationMs,
|
|
75
|
-
status: iter.error ? "ERROR" : "OK",
|
|
76
|
-
status_message: iter.error?.message,
|
|
77
|
-
service_name: "agent-runtime",
|
|
78
|
-
agent_name: iter.agentRunName,
|
|
79
|
-
model_name: null,
|
|
80
|
-
tool_name: null,
|
|
81
|
-
attributes: {
|
|
82
|
-
"iteration.index": iter.index,
|
|
83
|
-
"verdict.valid": iter.verdict?.valid,
|
|
84
|
-
"verdict.score": iter.verdict?.score,
|
|
85
|
-
"output.preview": iter.output === void 0 ? void 0 : String(iter.output).slice(0, 2e3)
|
|
86
|
-
}
|
|
87
|
-
}];
|
|
88
|
-
const models = /* @__PURE__ */ new Set();
|
|
89
|
-
const tools = /* @__PURE__ */ new Set();
|
|
90
|
-
iter.events.forEach((event, i) => {
|
|
91
|
-
const { kind, model, tool } = spanKindFor(event, iter.agentRunName);
|
|
92
|
-
if (model) models.add(model);
|
|
93
|
-
if (tool) tools.add(tool);
|
|
94
|
-
const errMsg = errorMessageOf(event);
|
|
95
|
-
spans.push({
|
|
96
|
-
trace_id: traceId,
|
|
97
|
-
span_id: `${traceId}:e${i}`,
|
|
98
|
-
parent_span_id: rootId,
|
|
99
|
-
name: String(event?.type ?? "event"),
|
|
100
|
-
kind,
|
|
101
|
-
start_time: start,
|
|
102
|
-
end_time: end,
|
|
103
|
-
duration_ms: 0,
|
|
104
|
-
status: errMsg ? "ERROR" : "OK",
|
|
105
|
-
status_message: errMsg,
|
|
106
|
-
service_name: "agent-runtime",
|
|
107
|
-
agent_name: iter.agentRunName,
|
|
108
|
-
model_name: model,
|
|
109
|
-
tool_name: tool,
|
|
110
|
-
attributes: event?.data ?? {}
|
|
111
|
-
});
|
|
112
|
-
});
|
|
113
|
-
const hasErrors = spans.some((s) => s.status === "ERROR") || iterErrored;
|
|
114
|
-
const summary = {
|
|
115
|
-
trace_id: traceId,
|
|
116
|
-
service_name: "agent-runtime",
|
|
117
|
-
agent_name: iter.agentRunName,
|
|
118
|
-
span_count: spans.length,
|
|
119
|
-
has_errors: hasErrors,
|
|
120
|
-
start_time: start,
|
|
121
|
-
end_time: end,
|
|
122
|
-
duration_ms: durationMs,
|
|
123
|
-
raw_jsonl_bytes: bytesOf(spans),
|
|
124
|
-
models: [...models],
|
|
125
|
-
tools: [...tools]
|
|
126
|
-
};
|
|
127
|
-
return {
|
|
128
|
-
summary,
|
|
129
|
-
spans,
|
|
130
|
-
rawBytes: summary.raw_jsonl_bytes
|
|
131
|
-
};
|
|
132
|
-
}
|
|
133
|
-
function matchesFilters(t, f) {
|
|
134
|
-
if (!f) return true;
|
|
135
|
-
if (f.has_errors !== void 0 && t.summary.has_errors !== f.has_errors) return false;
|
|
136
|
-
if (f.service_names?.length && !f.service_names.includes(t.summary.service_name ?? "")) return false;
|
|
137
|
-
if (f.agent_names?.length && !f.agent_names.includes(t.summary.agent_name ?? "")) return false;
|
|
138
|
-
if (f.model_names?.length && !f.model_names.some((m) => t.summary.models.includes(m))) return false;
|
|
139
|
-
if (f.tool_names?.length && !f.tool_names.some((tn) => t.summary.tools.includes(tn))) return false;
|
|
140
|
-
if (f.start_time_after && t.summary.start_time < f.start_time_after) return false;
|
|
141
|
-
if (f.start_time_before && t.summary.start_time > f.start_time_before) return false;
|
|
142
|
-
if (f.regex_pattern && !new RegExp(f.regex_pattern).test(JSON.stringify(t.spans))) return false;
|
|
143
|
-
return true;
|
|
144
|
-
}
|
|
145
|
-
function capAttributes(attributes, perAttrCap) {
|
|
146
|
-
let truncated = 0;
|
|
147
|
-
const capped = {};
|
|
148
|
-
for (const [k, v] of Object.entries(attributes)) {
|
|
149
|
-
const s = typeof v === "string" ? v : JSON.stringify(v);
|
|
150
|
-
if (typeof s === "string" && s.length > perAttrCap) {
|
|
151
|
-
truncated += 1;
|
|
152
|
-
capped[k] = `${TRACE_ANALYST_TRUNCATION_MARKER_PREFIX} ${s.length}b]${s.slice(0, perAttrCap)}`;
|
|
153
|
-
} else capped[k] = v;
|
|
154
|
-
}
|
|
155
|
-
return {
|
|
156
|
-
capped,
|
|
157
|
-
truncated
|
|
158
|
-
};
|
|
159
|
-
}
|
|
160
|
-
/**
|
|
161
|
-
* Build an in-memory `TraceAnalysisStore` over a loop round's iterations. Fail-loud on an
|
|
162
|
-
* empty round — there is nothing for an analyst to read, and a silent empty store would
|
|
163
|
-
* mask a broken capture path.
|
|
164
|
-
*/
|
|
165
|
-
function iterationsToTraceStore(iterations, budgets = DEFAULT_TRACE_ANALYST_BUDGETS) {
|
|
166
|
-
if (iterations.length === 0) throw new AnalystError("iterationsToTraceStore: no iterations to analyze (empty round)");
|
|
167
|
-
const traces = iterations.map((it) => projectIteration(it));
|
|
168
|
-
const byId = new Map(traces.map((t) => [t.summary.trace_id, t]));
|
|
169
|
-
const buildClusters = (set) => {
|
|
170
|
-
const map = /* @__PURE__ */ new Map();
|
|
171
|
-
for (const t of set) for (const s of t.spans) {
|
|
172
|
-
if (s.status !== "ERROR" || !s.status_message) continue;
|
|
173
|
-
const sig = normalizeSignature(s.status_message);
|
|
174
|
-
const c = map.get(sig) ?? {
|
|
175
|
-
signature: sig,
|
|
176
|
-
status_message_sample: s.status_message,
|
|
177
|
-
span_name: s.name,
|
|
178
|
-
tool_name: s.tool_name,
|
|
179
|
-
trace_count: 0,
|
|
180
|
-
span_count: 0,
|
|
181
|
-
prevalence: 0,
|
|
182
|
-
exemplar_trace_ids: [],
|
|
183
|
-
exemplar_span_ids: []
|
|
184
|
-
};
|
|
185
|
-
c.span_count += 1;
|
|
186
|
-
if (!c.exemplar_trace_ids.includes(t.summary.trace_id) && c.exemplar_trace_ids.length < 10) {
|
|
187
|
-
c.exemplar_trace_ids.push(t.summary.trace_id);
|
|
188
|
-
c.trace_count += 1;
|
|
189
|
-
}
|
|
190
|
-
if (c.exemplar_span_ids.length < 10) c.exemplar_span_ids.push(s.span_id);
|
|
191
|
-
map.set(sig, c);
|
|
192
|
-
}
|
|
193
|
-
const errorTraces = set.filter((t) => t.summary.has_errors).length || 1;
|
|
194
|
-
return [...map.values()].map((c) => ({
|
|
195
|
-
...c,
|
|
196
|
-
prevalence: c.trace_count / errorTraces
|
|
197
|
-
})).sort((a, b) => b.trace_count - a.trace_count);
|
|
198
|
-
};
|
|
199
|
-
return {
|
|
200
|
-
async hasTrace(trace_id) {
|
|
201
|
-
return byId.has(trace_id);
|
|
202
|
-
},
|
|
203
|
-
async hasSpans(opts) {
|
|
204
|
-
const t = byId.get(opts.trace_id);
|
|
205
|
-
if (!t) return [];
|
|
206
|
-
const present = new Set(t.spans.map((s) => s.span_id));
|
|
207
|
-
return opts.span_ids.filter((id) => present.has(id));
|
|
208
|
-
},
|
|
209
|
-
async getOverview(filters) {
|
|
210
|
-
const set = traces.filter((t) => matchesFilters(t, filters));
|
|
211
|
-
const services = /* @__PURE__ */ new Set();
|
|
212
|
-
const agents = /* @__PURE__ */ new Set();
|
|
213
|
-
const models = /* @__PURE__ */ new Set();
|
|
214
|
-
const tools = /* @__PURE__ */ new Set();
|
|
215
|
-
let errorSpans = 0;
|
|
216
|
-
for (const t of set) {
|
|
217
|
-
if (t.summary.service_name) services.add(t.summary.service_name);
|
|
218
|
-
if (t.summary.agent_name) agents.add(t.summary.agent_name);
|
|
219
|
-
for (const m of t.summary.models) models.add(m);
|
|
220
|
-
for (const tn of t.summary.tools) tools.add(tn);
|
|
221
|
-
errorSpans += t.spans.filter((s) => s.status === "ERROR").length;
|
|
222
|
-
}
|
|
223
|
-
const times = set.map((t) => t.summary.start_time).sort();
|
|
224
|
-
return {
|
|
225
|
-
total_traces: set.length,
|
|
226
|
-
raw_jsonl_bytes: set.reduce((n, t) => n + t.rawBytes, 0),
|
|
227
|
-
services: [...services],
|
|
228
|
-
agents: [...agents],
|
|
229
|
-
models: [...models],
|
|
230
|
-
tool_names: [...tools],
|
|
231
|
-
sample_trace_ids: set.slice(0, 20).map((t) => t.summary.trace_id),
|
|
232
|
-
errors: {
|
|
233
|
-
trace_count: set.filter((t) => t.summary.has_errors).length,
|
|
234
|
-
span_count: errorSpans
|
|
235
|
-
},
|
|
236
|
-
error_clusters: buildClusters(set),
|
|
237
|
-
time_range: times.length ? {
|
|
238
|
-
earliest: times[0],
|
|
239
|
-
latest: times[times.length - 1]
|
|
240
|
-
} : null
|
|
241
|
-
};
|
|
242
|
-
},
|
|
243
|
-
async queryTraces(opts) {
|
|
244
|
-
const set = traces.filter((t) => matchesFilters(t, opts.filters));
|
|
245
|
-
const offset = opts.offset ?? 0;
|
|
246
|
-
return {
|
|
247
|
-
traces: set.slice(offset, offset + opts.limit).map((t) => t.summary),
|
|
248
|
-
total: set.length,
|
|
249
|
-
has_more: offset + opts.limit < set.length
|
|
250
|
-
};
|
|
251
|
-
},
|
|
252
|
-
async countTraces(filters) {
|
|
253
|
-
return traces.filter((t) => matchesFilters(t, filters)).length;
|
|
254
|
-
},
|
|
255
|
-
async viewTrace(opts) {
|
|
256
|
-
const t = byId.get(opts.trace_id);
|
|
257
|
-
if (!t) return {
|
|
258
|
-
trace_id: opts.trace_id,
|
|
259
|
-
spans: []
|
|
260
|
-
};
|
|
261
|
-
const cap = opts.per_attribute_byte_cap ?? budgets.perAttributeViewBudget;
|
|
262
|
-
const projected = t.spans.map((s) => ({
|
|
263
|
-
...s,
|
|
264
|
-
attributes: capAttributes(s.attributes, cap).capped
|
|
265
|
-
}));
|
|
266
|
-
if (bytesOf(projected) > budgets.perCallByteCeiling) {
|
|
267
|
-
const names = /* @__PURE__ */ new Map();
|
|
268
|
-
for (const s of t.spans) names.set(s.name, (names.get(s.name) ?? 0) + 1);
|
|
269
|
-
return {
|
|
270
|
-
trace_id: opts.trace_id,
|
|
271
|
-
oversized: {
|
|
272
|
-
span_count: t.spans.length,
|
|
273
|
-
top_span_names: [...names.entries()].sort((a, b) => b[1] - a[1]).slice(0, 20),
|
|
274
|
-
span_response_bytes_max: Math.max(...t.spans.map((s) => bytesOf(s))),
|
|
275
|
-
error_span_count: t.spans.filter((s) => s.status === "ERROR").length
|
|
276
|
-
}
|
|
277
|
-
};
|
|
278
|
-
}
|
|
279
|
-
return {
|
|
280
|
-
trace_id: opts.trace_id,
|
|
281
|
-
spans: projected
|
|
282
|
-
};
|
|
283
|
-
},
|
|
284
|
-
async viewSpans(opts) {
|
|
285
|
-
const t = byId.get(opts.trace_id);
|
|
286
|
-
const cap = opts.per_attribute_byte_cap ?? budgets.perAttributeSpanBudget;
|
|
287
|
-
const want = new Set(opts.span_ids);
|
|
288
|
-
const found = (t?.spans ?? []).filter((s) => want.has(s.span_id));
|
|
289
|
-
let truncated = 0;
|
|
290
|
-
let bytes = 0;
|
|
291
|
-
const spans = [];
|
|
292
|
-
const omitted = [];
|
|
293
|
-
for (const s of found) {
|
|
294
|
-
const { capped, truncated: n } = capAttributes(s.attributes, cap);
|
|
295
|
-
const projected = {
|
|
296
|
-
...s,
|
|
297
|
-
attributes: capped
|
|
298
|
-
};
|
|
299
|
-
const size = bytesOf(projected);
|
|
300
|
-
if (spans.length > 0 && bytes + size > budgets.perCallByteCeiling) {
|
|
301
|
-
omitted.push(s.span_id);
|
|
302
|
-
continue;
|
|
303
|
-
}
|
|
304
|
-
truncated += n;
|
|
305
|
-
bytes += size;
|
|
306
|
-
spans.push(projected);
|
|
307
|
-
}
|
|
308
|
-
const foundIds = new Set(found.map((s) => s.span_id));
|
|
309
|
-
return {
|
|
310
|
-
trace_id: opts.trace_id,
|
|
311
|
-
spans,
|
|
312
|
-
missing_span_ids: opts.span_ids.filter((id) => !foundIds.has(id)),
|
|
313
|
-
omitted_span_ids: omitted,
|
|
314
|
-
has_more: omitted.length > 0,
|
|
315
|
-
truncated_attribute_count: truncated
|
|
316
|
-
};
|
|
317
|
-
},
|
|
318
|
-
async searchTrace(opts) {
|
|
319
|
-
const t = byId.get(opts.trace_id);
|
|
320
|
-
const max = opts.max_matches ?? 50;
|
|
321
|
-
const hits = [];
|
|
322
|
-
for (const s of t?.spans ?? []) for (const hit of searchSpanAttrs(s, opts.regex_pattern, budgets.perMatchTextBudget)) {
|
|
323
|
-
if (hits.length >= max) break;
|
|
324
|
-
hits.push(hit);
|
|
325
|
-
}
|
|
326
|
-
return {
|
|
327
|
-
trace_id: opts.trace_id,
|
|
328
|
-
hits,
|
|
329
|
-
has_more: hits.length >= max
|
|
330
|
-
};
|
|
331
|
-
},
|
|
332
|
-
async searchSpan(opts) {
|
|
333
|
-
const t = byId.get(opts.trace_id);
|
|
334
|
-
const max = opts.max_matches ?? 50;
|
|
335
|
-
const span = (t?.spans ?? []).find((s) => s.span_id === opts.span_id);
|
|
336
|
-
const all = span ? searchSpanAttrs(span, opts.regex_pattern, budgets.perMatchTextBudget) : [];
|
|
337
|
-
const hits = all.slice(0, max);
|
|
338
|
-
return {
|
|
339
|
-
trace_id: opts.trace_id,
|
|
340
|
-
span_id: opts.span_id,
|
|
341
|
-
hits,
|
|
342
|
-
has_more: all.length > hits.length
|
|
343
|
-
};
|
|
344
|
-
}
|
|
345
|
-
};
|
|
346
|
-
}
|
|
347
|
-
function searchSpanAttrs(span, pattern, textCap) {
|
|
348
|
-
const re = new RegExp(pattern, "g");
|
|
349
|
-
const hits = [];
|
|
350
|
-
for (const [k, v] of Object.entries(span.attributes)) {
|
|
351
|
-
const text = typeof v === "string" ? v : JSON.stringify(v);
|
|
352
|
-
if (typeof text !== "string") continue;
|
|
353
|
-
re.lastIndex = 0;
|
|
354
|
-
const m = re.exec(text);
|
|
355
|
-
if (!m) continue;
|
|
356
|
-
const at = m.index;
|
|
357
|
-
hits.push({
|
|
358
|
-
trace_id: span.trace_id,
|
|
359
|
-
span_id: span.span_id,
|
|
360
|
-
span_name: span.name,
|
|
361
|
-
span_kind: span.kind,
|
|
362
|
-
attribute_path: `attributes.${k}`,
|
|
363
|
-
matched_text: m[0].slice(0, textCap),
|
|
364
|
-
context_before: text.slice(Math.max(0, at - textCap / 2), at),
|
|
365
|
-
context_after: text.slice(at + m[0].length, at + m[0].length + textCap / 2),
|
|
366
|
-
match_offset: at
|
|
367
|
-
});
|
|
368
|
-
}
|
|
369
|
-
return hits;
|
|
370
|
-
}
|
|
371
|
-
//#endregion
|
|
372
|
-
//#region src/analyst-loop/run-analyst-loop.ts
|
|
373
|
-
/** Analyze a run and apply accepted knowledge and agent-surface proposals. */
|
|
374
|
-
async function runAnalystLoop(opts) {
|
|
375
|
-
const log = opts.log ?? defaultLog;
|
|
376
|
-
const strategy = opts.priorFindingsStrategy ?? "per-kind";
|
|
377
|
-
const emit = makeEmitter(opts.onEvent);
|
|
378
|
-
const startedAt = Date.now();
|
|
379
|
-
const baselineRunId = resolveBaselineRunId(opts);
|
|
380
|
-
const priorAll = baselineRunId ? opts.findingsStore?.loadRun(baselineRunId) ?? [] : [];
|
|
381
|
-
log("baseline resolved", {
|
|
382
|
-
baselineRunId,
|
|
383
|
-
prior_findings: priorAll.length
|
|
384
|
-
});
|
|
385
|
-
await emit({
|
|
386
|
-
type: "baseline-resolved",
|
|
387
|
-
runId: opts.runId,
|
|
388
|
-
baselineRunId,
|
|
389
|
-
priorFindingCount: priorAll.length
|
|
390
|
-
});
|
|
391
|
-
const analystResult = await runRegistry(opts, buildPriorFindingsInput(priorAll, strategy, opts.registry.list()), emit);
|
|
392
|
-
log("analyst run complete", {
|
|
393
|
-
findings: analystResult.findings.length,
|
|
394
|
-
cost_usd: analystResult.total_cost_usd,
|
|
395
|
-
per_analyst: analystResult.per_analyst.map((s) => ({
|
|
396
|
-
id: s.analyst_id,
|
|
397
|
-
status: s.status,
|
|
398
|
-
n: s.findings_count
|
|
399
|
-
}))
|
|
400
|
-
});
|
|
401
|
-
if (opts.findingsStore && analystResult.findings.length > 0) {
|
|
402
|
-
await opts.findingsStore.append(opts.runId, analystResult.findings);
|
|
403
|
-
await emit({
|
|
404
|
-
type: "findings-persisted",
|
|
405
|
-
runId: opts.runId,
|
|
406
|
-
count: analystResult.findings.length
|
|
407
|
-
});
|
|
408
|
-
}
|
|
409
|
-
let diff = null;
|
|
410
|
-
if (baselineRunId && analystResult.findings.length > 0) {
|
|
411
|
-
diff = diffFindings(priorAll.map((f) => ({ ...f })), analystResult.findings.map((f) => ({
|
|
412
|
-
...f,
|
|
413
|
-
run_id: opts.runId
|
|
414
|
-
})));
|
|
415
|
-
log("diff vs baseline", {
|
|
416
|
-
appeared: diff.appeared.length,
|
|
417
|
-
disappeared: diff.disappeared.length,
|
|
418
|
-
persisted: diff.persisted.length,
|
|
419
|
-
changed: diff.changed.length
|
|
420
|
-
});
|
|
421
|
-
await emit({
|
|
422
|
-
type: "diff-computed",
|
|
423
|
-
runId: opts.runId,
|
|
424
|
-
baselineRunId,
|
|
425
|
-
appeared: diff.appeared.length,
|
|
426
|
-
disappeared: diff.disappeared.length,
|
|
427
|
-
persisted: diff.persisted.length,
|
|
428
|
-
changed: diff.changed.length
|
|
429
|
-
});
|
|
430
|
-
}
|
|
431
|
-
let knowledge = null;
|
|
432
|
-
if (opts.knowledgeProposalSource) knowledge = await runKnowledgeProposalSource(opts, analystResult.findings, log, emit);
|
|
433
|
-
let improvement = null;
|
|
434
|
-
if (opts.improvementProposalSource) improvement = await runImprovementProposalSource(opts, analystResult.findings, log, emit);
|
|
435
|
-
const durationMs = Math.max(0, Date.now() - startedAt);
|
|
436
|
-
await emit({
|
|
437
|
-
type: "loop-completed",
|
|
438
|
-
runId: opts.runId,
|
|
439
|
-
durationMs
|
|
440
|
-
});
|
|
441
|
-
return {
|
|
442
|
-
runId: opts.runId,
|
|
443
|
-
baselineRunId,
|
|
444
|
-
durationMs,
|
|
445
|
-
analystResult,
|
|
446
|
-
diff,
|
|
447
|
-
knowledge,
|
|
448
|
-
improvement
|
|
449
|
-
};
|
|
450
|
-
}
|
|
451
|
-
function makeEmitter(onEvent) {
|
|
452
|
-
if (!onEvent) return async () => {};
|
|
453
|
-
return async (event) => {
|
|
454
|
-
await onEvent(event);
|
|
455
|
-
};
|
|
456
|
-
}
|
|
457
|
-
async function runRegistry(opts, priorFindings, emit) {
|
|
458
|
-
const reg = opts.registry;
|
|
459
|
-
const registryOptions = {
|
|
460
|
-
...priorFindings ? { priorFindings } : {},
|
|
461
|
-
...opts.chainFindings !== void 0 ? { chainFindings: opts.chainFindings } : {},
|
|
462
|
-
...opts.costLedger ? { costLedger: opts.costLedger } : {},
|
|
463
|
-
...opts.costPhase ? { costPhase: opts.costPhase } : {},
|
|
464
|
-
...opts.signal ? { signal: opts.signal } : {}
|
|
465
|
-
};
|
|
466
|
-
if (typeof reg.runStream === "function" && opts.onEvent) {
|
|
467
|
-
let final = null;
|
|
468
|
-
for await (const ev of reg.runStream(opts.runId, opts.inputs, registryOptions)) {
|
|
469
|
-
await emit({
|
|
470
|
-
type: "analyst",
|
|
471
|
-
runId: opts.runId,
|
|
472
|
-
event: ev
|
|
473
|
-
});
|
|
474
|
-
if (ev.type === "run-completed") final = ev.result;
|
|
475
|
-
}
|
|
476
|
-
if (!final) throw new Error("runAnalystLoop: registry.runStream ended without run-completed event");
|
|
477
|
-
return final;
|
|
478
|
-
}
|
|
479
|
-
return opts.registry.run(opts.runId, opts.inputs, registryOptions);
|
|
480
|
-
}
|
|
481
|
-
function resolveBaselineRunId(opts) {
|
|
482
|
-
if (opts.baselineRunId === null) return null;
|
|
483
|
-
if (typeof opts.baselineRunId === "string") return opts.baselineRunId;
|
|
484
|
-
if (!opts.findingsStore) return null;
|
|
485
|
-
const all = opts.findingsStore.loadAll();
|
|
486
|
-
let last = null;
|
|
487
|
-
for (const row of all) {
|
|
488
|
-
if (row.run_id === opts.runId) continue;
|
|
489
|
-
last = row.run_id;
|
|
490
|
-
}
|
|
491
|
-
return last;
|
|
492
|
-
}
|
|
493
|
-
function buildPriorFindingsInput(prior, strategy, registry) {
|
|
494
|
-
if (strategy === "none" || prior.length === 0) return void 0;
|
|
495
|
-
const stripped = prior.map(({ run_id: _run_id, ...rest }) => rest);
|
|
496
|
-
if (strategy === "wildcard") return { "*": stripped };
|
|
497
|
-
return stripped;
|
|
498
|
-
}
|
|
499
|
-
async function runKnowledgeProposalSource(opts, findings, log, emit) {
|
|
500
|
-
const batch = await opts.knowledgeProposalSource.proposeFromFindings(findings);
|
|
501
|
-
log("knowledge.proposeFromFindings", {
|
|
502
|
-
proposals: batch.proposals.length,
|
|
503
|
-
skipped: batch.skipped,
|
|
504
|
-
errors: batch.errors.length
|
|
505
|
-
});
|
|
506
|
-
await emit({
|
|
507
|
-
type: "knowledge-proposed",
|
|
508
|
-
runId: opts.runId,
|
|
509
|
-
proposalCount: batch.proposals.length,
|
|
510
|
-
skipped: batch.skipped,
|
|
511
|
-
errors: batch.errors.length
|
|
512
|
-
});
|
|
513
|
-
return {
|
|
514
|
-
proposals: batch.proposals,
|
|
515
|
-
skipped: batch.skipped,
|
|
516
|
-
errors: batch.errors
|
|
517
|
-
};
|
|
518
|
-
}
|
|
519
|
-
async function runImprovementProposalSource(opts, findings, log, emit) {
|
|
520
|
-
const batch = await opts.improvementProposalSource.proposeFromFindings(findings);
|
|
521
|
-
log("improvement.proposeFromFindings", {
|
|
522
|
-
edits: batch.edits.length,
|
|
523
|
-
skipped: batch.skipped,
|
|
524
|
-
errors: batch.errors.length
|
|
525
|
-
});
|
|
526
|
-
await emit({
|
|
527
|
-
type: "improvement-proposed",
|
|
528
|
-
runId: opts.runId,
|
|
529
|
-
editCount: batch.edits.length,
|
|
530
|
-
skipped: batch.skipped,
|
|
531
|
-
errors: batch.errors.length
|
|
532
|
-
});
|
|
533
|
-
return {
|
|
534
|
-
edits: batch.edits,
|
|
535
|
-
skipped: batch.skipped,
|
|
536
|
-
errors: batch.errors
|
|
537
|
-
};
|
|
538
|
-
}
|
|
539
|
-
function defaultLog(msg, fields) {
|
|
540
|
-
if (fields) console.log(`[analyst-loop] ${msg}`, fields);
|
|
541
|
-
else console.log(`[analyst-loop] ${msg}`);
|
|
542
|
-
}
|
|
543
|
-
//#endregion
|
|
544
|
-
export { iterationsToTraceStore as n, runAnalystLoop as t };
|
|
545
|
-
|
|
546
|
-
//# sourceMappingURL=analyst-loop-BknOQUW5.js.map
|