sentinelayer-cli 0.1.1 → 0.3.0
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/README.md +996 -996
- package/bin/create-sentinelayer.js +5 -5
- package/bin/sentinelayer-cli.js +4 -4
- package/bin/sl.js +5 -5
- package/package.json +62 -54
- package/src/agents/jules/config/definition.js +209 -209
- package/src/agents/jules/config/system-prompt.js +175 -175
- package/src/agents/jules/error-intake.js +51 -51
- package/src/agents/jules/fix-cycle.js +377 -377
- package/src/agents/jules/loop.js +367 -367
- package/src/agents/jules/pulse.js +319 -319
- package/src/agents/jules/stream.js +186 -186
- package/src/agents/jules/swarm/file-scanner.js +74 -74
- package/src/agents/jules/swarm/index.js +11 -11
- package/src/agents/jules/swarm/orchestrator.js +362 -362
- package/src/agents/jules/swarm/pattern-hunter.js +123 -123
- package/src/agents/jules/swarm/sub-agent.js +308 -308
- package/src/agents/jules/tools/auth-audit.js +226 -222
- package/src/agents/jules/tools/dispatch.js +327 -327
- package/src/agents/jules/tools/file-edit.js +180 -180
- package/src/agents/jules/tools/file-read.js +100 -100
- package/src/agents/jules/tools/frontend-analyze.js +570 -570
- package/src/agents/jules/tools/glob.js +168 -168
- package/src/agents/jules/tools/grep.js +228 -228
- package/src/agents/jules/tools/index.js +29 -29
- package/src/agents/jules/tools/path-guards.js +161 -161
- package/src/agents/jules/tools/runtime-audit.js +493 -493
- package/src/agents/jules/tools/shell.js +383 -383
- package/src/ai/aidenid.js +972 -945
- package/src/ai/client.js +508 -508
- package/src/ai/domain-target-store.js +268 -268
- package/src/ai/identity-store.js +270 -270
- package/src/ai/site-store.js +145 -145
- package/src/audit/agents/architecture.js +180 -180
- package/src/audit/agents/compliance.js +179 -179
- package/src/audit/agents/documentation.js +165 -165
- package/src/audit/agents/performance.js +145 -145
- package/src/audit/agents/security.js +215 -215
- package/src/audit/agents/testing.js +172 -172
- package/src/audit/orchestrator.js +557 -557
- package/src/audit/package.js +204 -204
- package/src/audit/registry.js +284 -284
- package/src/audit/replay.js +103 -103
- package/src/auth/http.js +113 -113
- package/src/auth/service.js +891 -848
- package/src/auth/session-store.js +359 -345
- package/src/cli.js +252 -252
- package/src/commands/ai/identity-lifecycle.js +1338 -1337
- package/src/commands/ai/provision-governance.js +1272 -1246
- package/src/commands/ai/shared.js +147 -147
- package/src/commands/ai.js +11 -11
- package/src/commands/apply.js +12 -12
- package/src/commands/audit.js +1166 -1147
- package/src/commands/auth.js +375 -366
- package/src/commands/chat.js +191 -191
- package/src/commands/config.js +184 -184
- package/src/commands/cost.js +311 -311
- package/src/commands/daemon/core.js +850 -850
- package/src/commands/daemon/extended.js +1048 -1048
- package/src/commands/daemon/shared.js +213 -213
- package/src/commands/daemon.js +11 -11
- package/src/commands/guide.js +174 -174
- package/src/commands/ingest.js +58 -58
- package/src/commands/init.js +55 -55
- package/src/commands/legacy-args.js +10 -10
- package/src/commands/mcp.js +461 -404
- package/src/commands/omargate.js +15 -15
- package/src/commands/persona.js +20 -20
- package/src/commands/plugin.js +260 -260
- package/src/commands/policy.js +132 -132
- package/src/commands/prompt.js +238 -238
- package/src/commands/review.js +704 -704
- package/src/commands/scan.js +866 -788
- package/src/commands/spec.js +716 -716
- package/src/commands/swarm.js +651 -651
- package/src/commands/telemetry.js +202 -202
- package/src/commands/watch.js +510 -510
- package/src/config/agent-dictionary.js +182 -182
- package/src/config/io.js +56 -56
- package/src/config/paths.js +18 -18
- package/src/config/schema.js +55 -55
- package/src/config/service.js +184 -184
- package/src/cost/budget.js +235 -235
- package/src/cost/history.js +188 -188
- package/src/cost/tracker.js +171 -171
- package/src/daemon/artifact-lineage.js +534 -534
- package/src/daemon/assignment-ledger.js +770 -770
- package/src/daemon/ast-parser-layer.js +258 -258
- package/src/daemon/budget-governor.js +633 -633
- package/src/daemon/callgraph-overlay.js +646 -646
- package/src/daemon/error-worker.js +626 -626
- package/src/daemon/hybrid-mapper.js +929 -929
- package/src/daemon/ingest-refresh.js +195 -0
- package/src/daemon/jira-lifecycle.js +632 -632
- package/src/daemon/operator-control.js +657 -657
- package/src/daemon/reliability-lane.js +471 -471
- package/src/daemon/watchdog.js +971 -971
- package/src/guide/generator.js +316 -316
- package/src/ingest/engine.js +918 -918
- package/src/interactive/action-menu.js +132 -0
- package/src/interactive/auto-ingest.js +111 -0
- package/src/interactive/index.js +95 -0
- package/src/interactive/workspace.js +92 -0
- package/src/legacy-cli.js +2548 -2435
- package/src/mcp/registry.js +695 -695
- package/src/memory/blackboard.js +301 -301
- package/src/memory/retrieval.js +581 -581
- package/src/plugin/manifest.js +553 -553
- package/src/policy/packs.js +144 -144
- package/src/prompt/generator.js +118 -106
- package/src/review/ai-review.js +669 -669
- package/src/review/local-review.js +1284 -1284
- package/src/review/replay.js +235 -235
- package/src/review/report.js +664 -664
- package/src/review/spec-binding.js +487 -487
- package/src/scaffold/generator.js +67 -0
- package/src/scaffold/templates.js +150 -0
- package/src/scan/generator.js +418 -351
- package/src/scan/gh-secrets.js +107 -0
- package/src/spec/generator.js +519 -519
- package/src/spec/regenerate.js +237 -237
- package/src/spec/templates.js +91 -91
- package/src/swarm/dashboard.js +247 -247
- package/src/swarm/factory.js +363 -363
- package/src/swarm/pentest.js +934 -934
- package/src/swarm/registry.js +419 -419
- package/src/swarm/report.js +158 -158
- package/src/swarm/runtime.js +576 -576
- package/src/swarm/scenario-dsl.js +272 -272
- package/src/telemetry/ledger.js +302 -302
- package/src/telemetry/session-tracker.js +118 -0
- package/src/telemetry/sync.js +190 -0
- package/src/ui/markdown.js +220 -220
package/src/swarm/runtime.js
CHANGED
|
@@ -1,576 +1,576 @@
|
|
|
1
|
-
import { randomUUID } from "node:crypto";
|
|
2
|
-
import fsp from "node:fs/promises";
|
|
3
|
-
import path from "node:path";
|
|
4
|
-
|
|
5
|
-
import { resolveOutputRoot } from "../config/service.js";
|
|
6
|
-
import { evaluateBudget } from "../cost/budget.js";
|
|
7
|
-
|
|
8
|
-
function normalizeString(value) {
|
|
9
|
-
return String(value || "").trim();
|
|
10
|
-
}
|
|
11
|
-
|
|
12
|
-
function formatTimestampToken() {
|
|
13
|
-
const now = new Date();
|
|
14
|
-
const pad = (value) => String(value).padStart(2, "0");
|
|
15
|
-
return `${now.getUTCFullYear()}${pad(now.getUTCMonth() + 1)}${pad(now.getUTCDate())}-${pad(
|
|
16
|
-
now.getUTCHours()
|
|
17
|
-
)}${pad(now.getUTCMinutes())}${pad(now.getUTCSeconds())}`;
|
|
18
|
-
}
|
|
19
|
-
|
|
20
|
-
function createRuntimeRunId() {
|
|
21
|
-
return `swarm-runtime-${formatTimestampToken()}-${randomUUID().slice(0, 8)}`;
|
|
22
|
-
}
|
|
23
|
-
|
|
24
|
-
function toPosixPath(value) {
|
|
25
|
-
return String(value || "").replace(/\\/g, "/");
|
|
26
|
-
}
|
|
27
|
-
|
|
28
|
-
function estimateTokens(text) {
|
|
29
|
-
const normalized = normalizeString(text);
|
|
30
|
-
if (!normalized) {
|
|
31
|
-
return 0;
|
|
32
|
-
}
|
|
33
|
-
return Math.max(1, Math.ceil(normalized.length / 4));
|
|
34
|
-
}
|
|
35
|
-
|
|
36
|
-
function normalizeEngine(value) {
|
|
37
|
-
const normalized = normalizeString(value).toLowerCase();
|
|
38
|
-
if (normalized === "playwright") {
|
|
39
|
-
return "playwright";
|
|
40
|
-
}
|
|
41
|
-
return "mock";
|
|
42
|
-
}
|
|
43
|
-
|
|
44
|
-
function normalizeMaxSteps(value) {
|
|
45
|
-
const normalized = Number(value || 0);
|
|
46
|
-
if (!Number.isFinite(normalized) || normalized < 1) {
|
|
47
|
-
throw new Error("maxSteps must be an integer >= 1.");
|
|
48
|
-
}
|
|
49
|
-
return Math.floor(normalized);
|
|
50
|
-
}
|
|
51
|
-
|
|
52
|
-
function normalizePlaybookAction(action = {}) {
|
|
53
|
-
const type = normalizeString(action.type).toLowerCase();
|
|
54
|
-
if (!type) {
|
|
55
|
-
return null;
|
|
56
|
-
}
|
|
57
|
-
return {
|
|
58
|
-
type,
|
|
59
|
-
url: normalizeString(action.url),
|
|
60
|
-
selector: normalizeString(action.selector),
|
|
61
|
-
text: normalizeString(action.text),
|
|
62
|
-
ms: Math.max(0, Math.floor(Number(action.ms || 0))),
|
|
63
|
-
path: normalizeString(action.path),
|
|
64
|
-
};
|
|
65
|
-
}
|
|
66
|
-
|
|
67
|
-
export async function loadSwarmPlaybook(playbookFile = "") {
|
|
68
|
-
const normalizedPath = normalizeString(playbookFile);
|
|
69
|
-
if (!normalizedPath) {
|
|
70
|
-
return [];
|
|
71
|
-
}
|
|
72
|
-
const resolved = path.resolve(process.cwd(), normalizedPath);
|
|
73
|
-
const raw = await fsp.readFile(resolved, "utf-8");
|
|
74
|
-
const parsed = JSON.parse(raw);
|
|
75
|
-
if (!parsed || typeof parsed !== "object" || !Array.isArray(parsed.actions)) {
|
|
76
|
-
throw new Error("Invalid playbook file: expected { actions: [...] }.");
|
|
77
|
-
}
|
|
78
|
-
return parsed.actions.map((action) => normalizePlaybookAction(action)).filter(Boolean);
|
|
79
|
-
}
|
|
80
|
-
|
|
81
|
-
export async function loadSwarmPlanFile(planFile = "") {
|
|
82
|
-
const normalizedPath = normalizeString(planFile);
|
|
83
|
-
if (!normalizedPath) {
|
|
84
|
-
throw new Error("planFile is required.");
|
|
85
|
-
}
|
|
86
|
-
const resolved = path.resolve(process.cwd(), normalizedPath);
|
|
87
|
-
const raw = await fsp.readFile(resolved, "utf-8");
|
|
88
|
-
const parsed = JSON.parse(raw);
|
|
89
|
-
if (!parsed || typeof parsed !== "object") {
|
|
90
|
-
throw new Error("Invalid plan file: expected object payload.");
|
|
91
|
-
}
|
|
92
|
-
if (!Array.isArray(parsed.assignments) || parsed.assignments.length === 0) {
|
|
93
|
-
throw new Error("Invalid plan file: assignments are required.");
|
|
94
|
-
}
|
|
95
|
-
return parsed;
|
|
96
|
-
}
|
|
97
|
-
|
|
98
|
-
function createEvent({
|
|
99
|
-
runId,
|
|
100
|
-
step,
|
|
101
|
-
eventType,
|
|
102
|
-
agentId = "",
|
|
103
|
-
message = "",
|
|
104
|
-
metadata = {},
|
|
105
|
-
usage = {},
|
|
106
|
-
} = {}) {
|
|
107
|
-
return {
|
|
108
|
-
timestamp: new Date().toISOString(),
|
|
109
|
-
runId,
|
|
110
|
-
step,
|
|
111
|
-
eventType,
|
|
112
|
-
agentId: normalizeString(agentId).toLowerCase(),
|
|
113
|
-
message: normalizeString(message),
|
|
114
|
-
usage: {
|
|
115
|
-
outputTokens: Number(usage.outputTokens || 0),
|
|
116
|
-
toolCalls: Number(usage.toolCalls || 0),
|
|
117
|
-
durationMs: Number(usage.durationMs || 0),
|
|
118
|
-
costUsd: Number(usage.costUsd || 0),
|
|
119
|
-
},
|
|
120
|
-
metadata,
|
|
121
|
-
};
|
|
122
|
-
}
|
|
123
|
-
|
|
124
|
-
async function executePlaywrightAction({ page, action, runDirectory, usage, runId, step }) {
|
|
125
|
-
const type = action.type;
|
|
126
|
-
const metadata = {
|
|
127
|
-
action: type,
|
|
128
|
-
};
|
|
129
|
-
if (type === "goto") {
|
|
130
|
-
const url = action.url || "about:blank";
|
|
131
|
-
await page.goto(url, { waitUntil: "domcontentloaded" });
|
|
132
|
-
metadata.url = url;
|
|
133
|
-
usage.toolCalls += 1;
|
|
134
|
-
usage.outputTokens += estimateTokens(`goto:${url}`);
|
|
135
|
-
return createEvent({
|
|
136
|
-
runId,
|
|
137
|
-
step,
|
|
138
|
-
eventType: "tool_call",
|
|
139
|
-
agentId: "omar",
|
|
140
|
-
message: `Playwright goto ${url}`,
|
|
141
|
-
metadata,
|
|
142
|
-
usage,
|
|
143
|
-
});
|
|
144
|
-
}
|
|
145
|
-
if (type === "click") {
|
|
146
|
-
if (!action.selector) {
|
|
147
|
-
throw new Error("Playbook click action requires selector.");
|
|
148
|
-
}
|
|
149
|
-
await page.click(action.selector);
|
|
150
|
-
metadata.selector = action.selector;
|
|
151
|
-
usage.toolCalls += 1;
|
|
152
|
-
usage.outputTokens += estimateTokens(`click:${action.selector}`);
|
|
153
|
-
return createEvent({
|
|
154
|
-
runId,
|
|
155
|
-
step,
|
|
156
|
-
eventType: "tool_call",
|
|
157
|
-
agentId: "omar",
|
|
158
|
-
message: `Playwright click ${action.selector}`,
|
|
159
|
-
metadata,
|
|
160
|
-
usage,
|
|
161
|
-
});
|
|
162
|
-
}
|
|
163
|
-
if (type === "fill") {
|
|
164
|
-
if (!action.selector) {
|
|
165
|
-
throw new Error("Playbook fill action requires selector.");
|
|
166
|
-
}
|
|
167
|
-
await page.fill(action.selector, action.text || "");
|
|
168
|
-
metadata.selector = action.selector;
|
|
169
|
-
usage.toolCalls += 1;
|
|
170
|
-
usage.outputTokens += estimateTokens(`fill:${action.selector}:${action.text || ""}`);
|
|
171
|
-
return createEvent({
|
|
172
|
-
runId,
|
|
173
|
-
step,
|
|
174
|
-
eventType: "tool_call",
|
|
175
|
-
agentId: "omar",
|
|
176
|
-
message: `Playwright fill ${action.selector}`,
|
|
177
|
-
metadata,
|
|
178
|
-
usage,
|
|
179
|
-
});
|
|
180
|
-
}
|
|
181
|
-
if (type === "wait") {
|
|
182
|
-
const waitMs = Math.max(0, Number(action.ms || 0));
|
|
183
|
-
await page.waitForTimeout(waitMs);
|
|
184
|
-
metadata.ms = waitMs;
|
|
185
|
-
usage.toolCalls += 1;
|
|
186
|
-
usage.outputTokens += estimateTokens(`wait:${waitMs}`);
|
|
187
|
-
return createEvent({
|
|
188
|
-
runId,
|
|
189
|
-
step,
|
|
190
|
-
eventType: "tool_call",
|
|
191
|
-
agentId: "omar",
|
|
192
|
-
message: `Playwright wait ${waitMs}ms`,
|
|
193
|
-
metadata,
|
|
194
|
-
usage,
|
|
195
|
-
});
|
|
196
|
-
}
|
|
197
|
-
if (type === "screenshot") {
|
|
198
|
-
const outputPath = action.path
|
|
199
|
-
? path.resolve(runDirectory, action.path)
|
|
200
|
-
: path.join(runDirectory, "runtime", `step-${String(step).padStart(3, "0")}.png`);
|
|
201
|
-
await fsp.mkdir(path.dirname(outputPath), { recursive: true });
|
|
202
|
-
await page.screenshot({
|
|
203
|
-
path: outputPath,
|
|
204
|
-
fullPage: true,
|
|
205
|
-
});
|
|
206
|
-
metadata.path = toPosixPath(path.relative(runDirectory, outputPath));
|
|
207
|
-
usage.toolCalls += 1;
|
|
208
|
-
usage.outputTokens += estimateTokens(`screenshot:${metadata.path}`);
|
|
209
|
-
return createEvent({
|
|
210
|
-
runId,
|
|
211
|
-
step,
|
|
212
|
-
eventType: "tool_call",
|
|
213
|
-
agentId: "omar",
|
|
214
|
-
message: `Playwright screenshot ${metadata.path}`,
|
|
215
|
-
metadata,
|
|
216
|
-
usage,
|
|
217
|
-
});
|
|
218
|
-
}
|
|
219
|
-
|
|
220
|
-
usage.toolCalls += 1;
|
|
221
|
-
usage.outputTokens += estimateTokens(`unsupported:${type}`);
|
|
222
|
-
return createEvent({
|
|
223
|
-
runId,
|
|
224
|
-
step,
|
|
225
|
-
eventType: "tool_call",
|
|
226
|
-
agentId: "omar",
|
|
227
|
-
message: `Unsupported Playwright action skipped: ${type}`,
|
|
228
|
-
metadata: {
|
|
229
|
-
action: type,
|
|
230
|
-
skipped: true,
|
|
231
|
-
},
|
|
232
|
-
usage,
|
|
233
|
-
});
|
|
234
|
-
}
|
|
235
|
-
|
|
236
|
-
function buildRuntimeMarkdown(summary = {}) {
|
|
237
|
-
return `# SWARM_RUNTIME
|
|
238
|
-
|
|
239
|
-
Generated: ${summary.generatedAt}
|
|
240
|
-
Run ID: ${summary.runId}
|
|
241
|
-
Plan run ID: ${summary.planRunId}
|
|
242
|
-
Target: ${summary.targetPath}
|
|
243
|
-
Scenario: ${summary.scenario}
|
|
244
|
-
Engine: ${summary.engine}
|
|
245
|
-
Execute: ${summary.execute ? "yes" : "no"}
|
|
246
|
-
|
|
247
|
-
Status:
|
|
248
|
-
- completed: ${summary.completed ? "yes" : "no"}
|
|
249
|
-
- stop_class: ${summary.stop?.stopClass || "NONE"}
|
|
250
|
-
- stop_reason: ${summary.stop?.reason || "none"}
|
|
251
|
-
|
|
252
|
-
Usage:
|
|
253
|
-
- output_tokens: ${summary.usage.outputTokens}
|
|
254
|
-
- tool_calls: ${summary.usage.toolCalls}
|
|
255
|
-
- duration_ms: ${summary.usage.durationMs}
|
|
256
|
-
- cost_usd: ${summary.usage.costUsd}
|
|
257
|
-
|
|
258
|
-
Artifacts:
|
|
259
|
-
- events: ${summary.runtimeEventsPath}
|
|
260
|
-
- summary_json: ${summary.runtimeJsonPath}
|
|
261
|
-
`;
|
|
262
|
-
}
|
|
263
|
-
|
|
264
|
-
async function writeRuntimeArtifacts({
|
|
265
|
-
summary,
|
|
266
|
-
events,
|
|
267
|
-
outputDir = "",
|
|
268
|
-
env,
|
|
269
|
-
homeDir,
|
|
270
|
-
} = {}) {
|
|
271
|
-
const outputRoot = await resolveOutputRoot({
|
|
272
|
-
cwd: path.resolve(String(summary.targetPath || ".")),
|
|
273
|
-
outputDirOverride: outputDir,
|
|
274
|
-
env,
|
|
275
|
-
homeDir,
|
|
276
|
-
});
|
|
277
|
-
const runDirectory = path.join(outputRoot, "swarms", summary.runId);
|
|
278
|
-
const runtimeDirectory = path.join(runDirectory, "runtime");
|
|
279
|
-
const runtimeJsonPath = path.join(runtimeDirectory, "SWARM_RUNTIME.json");
|
|
280
|
-
const runtimeMarkdownPath = path.join(runtimeDirectory, "SWARM_RUNTIME.md");
|
|
281
|
-
const runtimeEventsPath = path.join(runtimeDirectory, "events.ndjson");
|
|
282
|
-
await fsp.mkdir(runtimeDirectory, { recursive: true });
|
|
283
|
-
await fsp.writeFile(runtimeEventsPath, `${events.map((event) => JSON.stringify(event)).join("\n")}\n`, "utf-8");
|
|
284
|
-
|
|
285
|
-
const fullSummary = {
|
|
286
|
-
...summary,
|
|
287
|
-
outputRoot,
|
|
288
|
-
runDirectory,
|
|
289
|
-
runtimeDirectory,
|
|
290
|
-
runtimeJsonPath,
|
|
291
|
-
runtimeMarkdownPath,
|
|
292
|
-
runtimeEventsPath,
|
|
293
|
-
};
|
|
294
|
-
|
|
295
|
-
await fsp.writeFile(runtimeJsonPath, `${JSON.stringify(fullSummary, null, 2)}\n`, "utf-8");
|
|
296
|
-
await fsp.writeFile(runtimeMarkdownPath, `${buildRuntimeMarkdown(fullSummary).trim()}\n`, "utf-8");
|
|
297
|
-
|
|
298
|
-
return fullSummary;
|
|
299
|
-
}
|
|
300
|
-
|
|
301
|
-
export async function runSwarmRuntime({
|
|
302
|
-
plan,
|
|
303
|
-
targetPath,
|
|
304
|
-
engine = "mock",
|
|
305
|
-
execute = false,
|
|
306
|
-
maxSteps = 20,
|
|
307
|
-
startUrl = "about:blank",
|
|
308
|
-
playbookActions = [],
|
|
309
|
-
outputDir = "",
|
|
310
|
-
env,
|
|
311
|
-
} = {}) {
|
|
312
|
-
if (!plan || typeof plan !== "object") {
|
|
313
|
-
throw new Error("plan is required.");
|
|
314
|
-
}
|
|
315
|
-
if (!Array.isArray(plan.assignments) || plan.assignments.length === 0) {
|
|
316
|
-
throw new Error("plan.assignments must include at least one assignment.");
|
|
317
|
-
}
|
|
318
|
-
|
|
319
|
-
const normalizedEngine = normalizeEngine(engine);
|
|
320
|
-
const normalizedTargetPath = path.resolve(String(targetPath || plan.targetPath || "."));
|
|
321
|
-
const normalizedMaxSteps = normalizeMaxSteps(maxSteps);
|
|
322
|
-
const runId = createRuntimeRunId();
|
|
323
|
-
const resolvedOutputRoot = await resolveOutputRoot({
|
|
324
|
-
cwd: normalizedTargetPath,
|
|
325
|
-
outputDirOverride: outputDir,
|
|
326
|
-
env,
|
|
327
|
-
});
|
|
328
|
-
const runtimeRunDirectory = path.join(resolvedOutputRoot, "swarms", runId);
|
|
329
|
-
const runStartedAt = Date.now();
|
|
330
|
-
const events = [];
|
|
331
|
-
let step = 0;
|
|
332
|
-
|
|
333
|
-
const usage = {
|
|
334
|
-
outputTokens: 0,
|
|
335
|
-
toolCalls: 0,
|
|
336
|
-
durationMs: 0,
|
|
337
|
-
costUsd: 0,
|
|
338
|
-
};
|
|
339
|
-
let stop = {
|
|
340
|
-
stopClass: "NONE",
|
|
341
|
-
reason: "",
|
|
342
|
-
blocking: false,
|
|
343
|
-
};
|
|
344
|
-
|
|
345
|
-
events.push(
|
|
346
|
-
createEvent({
|
|
347
|
-
runId,
|
|
348
|
-
step,
|
|
349
|
-
eventType: "run_start",
|
|
350
|
-
agentId: "omar",
|
|
351
|
-
message: `Swarm runtime started with engine=${normalizedEngine}, execute=${Boolean(execute)}`,
|
|
352
|
-
metadata: {
|
|
353
|
-
planRunId: normalizeString(plan.runId),
|
|
354
|
-
scenario: normalizeString(plan.scenario),
|
|
355
|
-
},
|
|
356
|
-
usage,
|
|
357
|
-
})
|
|
358
|
-
);
|
|
359
|
-
|
|
360
|
-
let browser = null;
|
|
361
|
-
let page = null;
|
|
362
|
-
let playwrightActions = playbookActions;
|
|
363
|
-
|
|
364
|
-
try {
|
|
365
|
-
if (normalizedEngine === "playwright" && execute) {
|
|
366
|
-
const playwright = await import("playwright");
|
|
367
|
-
browser = await playwright.chromium.launch({
|
|
368
|
-
headless: true,
|
|
369
|
-
});
|
|
370
|
-
page = await browser.newPage();
|
|
371
|
-
await page.goto(startUrl || "about:blank", { waitUntil: "domcontentloaded" });
|
|
372
|
-
usage.toolCalls += 1;
|
|
373
|
-
usage.outputTokens += estimateTokens(`goto:${startUrl || "about:blank"}`);
|
|
374
|
-
step += 1;
|
|
375
|
-
events.push(
|
|
376
|
-
createEvent({
|
|
377
|
-
runId,
|
|
378
|
-
step,
|
|
379
|
-
eventType: "tool_call",
|
|
380
|
-
agentId: "omar",
|
|
381
|
-
message: `Playwright runtime initialized at ${startUrl || "about:blank"}`,
|
|
382
|
-
metadata: {
|
|
383
|
-
action: "goto",
|
|
384
|
-
url: startUrl || "about:blank",
|
|
385
|
-
},
|
|
386
|
-
usage,
|
|
387
|
-
})
|
|
388
|
-
);
|
|
389
|
-
}
|
|
390
|
-
|
|
391
|
-
for (const assignment of plan.assignments) {
|
|
392
|
-
if (step >= normalizedMaxSteps) {
|
|
393
|
-
stop = {
|
|
394
|
-
stopClass: "MAX_STEPS_EXCEEDED",
|
|
395
|
-
reason: `max-steps reached (${normalizedMaxSteps})`,
|
|
396
|
-
blocking: true,
|
|
397
|
-
};
|
|
398
|
-
break;
|
|
399
|
-
}
|
|
400
|
-
|
|
401
|
-
step += 1;
|
|
402
|
-
usage.outputTokens += estimateTokens(assignment.objective);
|
|
403
|
-
events.push(
|
|
404
|
-
createEvent({
|
|
405
|
-
runId,
|
|
406
|
-
step,
|
|
407
|
-
eventType: "run_step",
|
|
408
|
-
agentId: assignment.agentId,
|
|
409
|
-
message: `Assignment started: ${assignment.objective}`,
|
|
410
|
-
metadata: {
|
|
411
|
-
assignmentId: assignment.assignmentId,
|
|
412
|
-
role: assignment.role,
|
|
413
|
-
domain: assignment.domain,
|
|
414
|
-
},
|
|
415
|
-
usage,
|
|
416
|
-
})
|
|
417
|
-
);
|
|
418
|
-
|
|
419
|
-
if (normalizedEngine === "mock" || !execute) {
|
|
420
|
-
usage.toolCalls += 1;
|
|
421
|
-
usage.outputTokens += estimateTokens(`mock:${assignment.agentId}`);
|
|
422
|
-
step += 1;
|
|
423
|
-
events.push(
|
|
424
|
-
createEvent({
|
|
425
|
-
runId,
|
|
426
|
-
step,
|
|
427
|
-
eventType: "tool_call",
|
|
428
|
-
agentId: assignment.agentId,
|
|
429
|
-
message: `Mock runtime action completed for ${assignment.agentId}`,
|
|
430
|
-
metadata: {
|
|
431
|
-
engine: normalizedEngine,
|
|
432
|
-
execute: Boolean(execute),
|
|
433
|
-
},
|
|
434
|
-
usage,
|
|
435
|
-
})
|
|
436
|
-
);
|
|
437
|
-
} else if (normalizedEngine === "playwright" && execute && page) {
|
|
438
|
-
let resolvedActions;
|
|
439
|
-
if (Array.isArray(playwrightActions) && playwrightActions.length > 0) {
|
|
440
|
-
resolvedActions = playwrightActions;
|
|
441
|
-
} else {
|
|
442
|
-
resolvedActions = [{ type: "wait", ms: 250 }, { type: "screenshot" }];
|
|
443
|
-
}
|
|
444
|
-
|
|
445
|
-
for (const action of resolvedActions) {
|
|
446
|
-
if (step >= normalizedMaxSteps) {
|
|
447
|
-
stop = {
|
|
448
|
-
stopClass: "MAX_STEPS_EXCEEDED",
|
|
449
|
-
reason: `max-steps reached (${normalizedMaxSteps})`,
|
|
450
|
-
blocking: true,
|
|
451
|
-
};
|
|
452
|
-
break;
|
|
453
|
-
}
|
|
454
|
-
step += 1;
|
|
455
|
-
const event = await executePlaywrightAction({
|
|
456
|
-
page,
|
|
457
|
-
action,
|
|
458
|
-
runDirectory: runtimeRunDirectory,
|
|
459
|
-
usage,
|
|
460
|
-
runId,
|
|
461
|
-
step,
|
|
462
|
-
});
|
|
463
|
-
events.push(event);
|
|
464
|
-
}
|
|
465
|
-
if (stop.blocking) {
|
|
466
|
-
break;
|
|
467
|
-
}
|
|
468
|
-
}
|
|
469
|
-
|
|
470
|
-
usage.durationMs = Date.now() - runStartedAt;
|
|
471
|
-
usage.costUsd = Number((usage.outputTokens * 0.000003).toFixed(6));
|
|
472
|
-
const budgetStatus = evaluateBudget({
|
|
473
|
-
sessionSummary: {
|
|
474
|
-
costUsd: usage.costUsd,
|
|
475
|
-
outputTokens: usage.outputTokens,
|
|
476
|
-
noProgressStreak: 0,
|
|
477
|
-
durationMs: usage.durationMs,
|
|
478
|
-
toolCalls: usage.toolCalls,
|
|
479
|
-
},
|
|
480
|
-
maxCostUsd: Number(plan.globalBudget?.maxCostUsd || 5),
|
|
481
|
-
maxOutputTokens: Number(plan.globalBudget?.maxOutputTokens || 20000),
|
|
482
|
-
maxNoProgress: Number.MAX_SAFE_INTEGER,
|
|
483
|
-
maxRuntimeMs: Number(plan.globalBudget?.maxRuntimeMs || 3600000),
|
|
484
|
-
maxToolCalls: Number(plan.globalBudget?.maxToolCalls || 500),
|
|
485
|
-
warningThresholdPercent: Number(plan.globalBudget?.warningThresholdPercent || 80),
|
|
486
|
-
});
|
|
487
|
-
if (budgetStatus.blocking) {
|
|
488
|
-
stop = {
|
|
489
|
-
stopClass: String(budgetStatus.reasons[0]?.code || "BUDGET_EXCEEDED"),
|
|
490
|
-
reason: String(budgetStatus.reasons[0]?.message || "Budget exceeded."),
|
|
491
|
-
blocking: true,
|
|
492
|
-
};
|
|
493
|
-
step += 1;
|
|
494
|
-
events.push(
|
|
495
|
-
createEvent({
|
|
496
|
-
runId,
|
|
497
|
-
step,
|
|
498
|
-
eventType: "budget_stop",
|
|
499
|
-
agentId: "omar",
|
|
500
|
-
message: stop.reason,
|
|
501
|
-
metadata: {
|
|
502
|
-
reasonCodes: budgetStatus.reasons.map((reason) => reason.code),
|
|
503
|
-
},
|
|
504
|
-
usage,
|
|
505
|
-
})
|
|
506
|
-
);
|
|
507
|
-
break;
|
|
508
|
-
}
|
|
509
|
-
}
|
|
510
|
-
} finally {
|
|
511
|
-
if (browser) {
|
|
512
|
-
await browser.close();
|
|
513
|
-
}
|
|
514
|
-
}
|
|
515
|
-
|
|
516
|
-
usage.durationMs = Date.now() - runStartedAt;
|
|
517
|
-
usage.costUsd = Number((usage.outputTokens * 0.000003).toFixed(6));
|
|
518
|
-
const completed = !stop.blocking;
|
|
519
|
-
if (completed) {
|
|
520
|
-
step += 1;
|
|
521
|
-
events.push(
|
|
522
|
-
createEvent({
|
|
523
|
-
runId,
|
|
524
|
-
step,
|
|
525
|
-
eventType: "run_stop",
|
|
526
|
-
agentId: "omar",
|
|
527
|
-
message: "Swarm runtime completed successfully.",
|
|
528
|
-
metadata: {
|
|
529
|
-
completed: true,
|
|
530
|
-
},
|
|
531
|
-
usage,
|
|
532
|
-
})
|
|
533
|
-
);
|
|
534
|
-
} else {
|
|
535
|
-
step += 1;
|
|
536
|
-
events.push(
|
|
537
|
-
createEvent({
|
|
538
|
-
runId,
|
|
539
|
-
step,
|
|
540
|
-
eventType: "run_stop",
|
|
541
|
-
agentId: "omar",
|
|
542
|
-
message: stop.reason || "Swarm runtime stopped.",
|
|
543
|
-
metadata: {
|
|
544
|
-
completed: false,
|
|
545
|
-
stopClass: stop.stopClass,
|
|
546
|
-
},
|
|
547
|
-
usage,
|
|
548
|
-
})
|
|
549
|
-
);
|
|
550
|
-
}
|
|
551
|
-
|
|
552
|
-
const summary = {
|
|
553
|
-
schemaVersion: 1,
|
|
554
|
-
generatedAt: new Date().toISOString(),
|
|
555
|
-
runId,
|
|
556
|
-
planRunId: normalizeString(plan.runId),
|
|
557
|
-
targetPath: normalizedTargetPath,
|
|
558
|
-
scenario: normalizeString(plan.scenario || "qa_audit"),
|
|
559
|
-
engine: normalizedEngine,
|
|
560
|
-
execute: Boolean(execute),
|
|
561
|
-
startUrl: normalizeString(startUrl || "about:blank"),
|
|
562
|
-
maxSteps: normalizedMaxSteps,
|
|
563
|
-
completed,
|
|
564
|
-
stop,
|
|
565
|
-
usage,
|
|
566
|
-
eventCount: events.length,
|
|
567
|
-
selectedAgents: Array.isArray(plan.selectedAgents) ? [...plan.selectedAgents] : [],
|
|
568
|
-
};
|
|
569
|
-
|
|
570
|
-
return writeRuntimeArtifacts({
|
|
571
|
-
summary,
|
|
572
|
-
events,
|
|
573
|
-
outputDir,
|
|
574
|
-
env,
|
|
575
|
-
});
|
|
576
|
-
}
|
|
1
|
+
import { randomUUID } from "node:crypto";
|
|
2
|
+
import fsp from "node:fs/promises";
|
|
3
|
+
import path from "node:path";
|
|
4
|
+
|
|
5
|
+
import { resolveOutputRoot } from "../config/service.js";
|
|
6
|
+
import { evaluateBudget } from "../cost/budget.js";
|
|
7
|
+
|
|
8
|
+
function normalizeString(value) {
|
|
9
|
+
return String(value || "").trim();
|
|
10
|
+
}
|
|
11
|
+
|
|
12
|
+
function formatTimestampToken() {
|
|
13
|
+
const now = new Date();
|
|
14
|
+
const pad = (value) => String(value).padStart(2, "0");
|
|
15
|
+
return `${now.getUTCFullYear()}${pad(now.getUTCMonth() + 1)}${pad(now.getUTCDate())}-${pad(
|
|
16
|
+
now.getUTCHours()
|
|
17
|
+
)}${pad(now.getUTCMinutes())}${pad(now.getUTCSeconds())}`;
|
|
18
|
+
}
|
|
19
|
+
|
|
20
|
+
function createRuntimeRunId() {
|
|
21
|
+
return `swarm-runtime-${formatTimestampToken()}-${randomUUID().slice(0, 8)}`;
|
|
22
|
+
}
|
|
23
|
+
|
|
24
|
+
function toPosixPath(value) {
|
|
25
|
+
return String(value || "").replace(/\\/g, "/");
|
|
26
|
+
}
|
|
27
|
+
|
|
28
|
+
function estimateTokens(text) {
|
|
29
|
+
const normalized = normalizeString(text);
|
|
30
|
+
if (!normalized) {
|
|
31
|
+
return 0;
|
|
32
|
+
}
|
|
33
|
+
return Math.max(1, Math.ceil(normalized.length / 4));
|
|
34
|
+
}
|
|
35
|
+
|
|
36
|
+
function normalizeEngine(value) {
|
|
37
|
+
const normalized = normalizeString(value).toLowerCase();
|
|
38
|
+
if (normalized === "playwright") {
|
|
39
|
+
return "playwright";
|
|
40
|
+
}
|
|
41
|
+
return "mock";
|
|
42
|
+
}
|
|
43
|
+
|
|
44
|
+
function normalizeMaxSteps(value) {
|
|
45
|
+
const normalized = Number(value || 0);
|
|
46
|
+
if (!Number.isFinite(normalized) || normalized < 1) {
|
|
47
|
+
throw new Error("maxSteps must be an integer >= 1.");
|
|
48
|
+
}
|
|
49
|
+
return Math.floor(normalized);
|
|
50
|
+
}
|
|
51
|
+
|
|
52
|
+
function normalizePlaybookAction(action = {}) {
|
|
53
|
+
const type = normalizeString(action.type).toLowerCase();
|
|
54
|
+
if (!type) {
|
|
55
|
+
return null;
|
|
56
|
+
}
|
|
57
|
+
return {
|
|
58
|
+
type,
|
|
59
|
+
url: normalizeString(action.url),
|
|
60
|
+
selector: normalizeString(action.selector),
|
|
61
|
+
text: normalizeString(action.text),
|
|
62
|
+
ms: Math.max(0, Math.floor(Number(action.ms || 0))),
|
|
63
|
+
path: normalizeString(action.path),
|
|
64
|
+
};
|
|
65
|
+
}
|
|
66
|
+
|
|
67
|
+
export async function loadSwarmPlaybook(playbookFile = "") {
|
|
68
|
+
const normalizedPath = normalizeString(playbookFile);
|
|
69
|
+
if (!normalizedPath) {
|
|
70
|
+
return [];
|
|
71
|
+
}
|
|
72
|
+
const resolved = path.resolve(process.cwd(), normalizedPath);
|
|
73
|
+
const raw = await fsp.readFile(resolved, "utf-8");
|
|
74
|
+
const parsed = JSON.parse(raw);
|
|
75
|
+
if (!parsed || typeof parsed !== "object" || !Array.isArray(parsed.actions)) {
|
|
76
|
+
throw new Error("Invalid playbook file: expected { actions: [...] }.");
|
|
77
|
+
}
|
|
78
|
+
return parsed.actions.map((action) => normalizePlaybookAction(action)).filter(Boolean);
|
|
79
|
+
}
|
|
80
|
+
|
|
81
|
+
export async function loadSwarmPlanFile(planFile = "") {
|
|
82
|
+
const normalizedPath = normalizeString(planFile);
|
|
83
|
+
if (!normalizedPath) {
|
|
84
|
+
throw new Error("planFile is required.");
|
|
85
|
+
}
|
|
86
|
+
const resolved = path.resolve(process.cwd(), normalizedPath);
|
|
87
|
+
const raw = await fsp.readFile(resolved, "utf-8");
|
|
88
|
+
const parsed = JSON.parse(raw);
|
|
89
|
+
if (!parsed || typeof parsed !== "object") {
|
|
90
|
+
throw new Error("Invalid plan file: expected object payload.");
|
|
91
|
+
}
|
|
92
|
+
if (!Array.isArray(parsed.assignments) || parsed.assignments.length === 0) {
|
|
93
|
+
throw new Error("Invalid plan file: assignments are required.");
|
|
94
|
+
}
|
|
95
|
+
return parsed;
|
|
96
|
+
}
|
|
97
|
+
|
|
98
|
+
function createEvent({
|
|
99
|
+
runId,
|
|
100
|
+
step,
|
|
101
|
+
eventType,
|
|
102
|
+
agentId = "",
|
|
103
|
+
message = "",
|
|
104
|
+
metadata = {},
|
|
105
|
+
usage = {},
|
|
106
|
+
} = {}) {
|
|
107
|
+
return {
|
|
108
|
+
timestamp: new Date().toISOString(),
|
|
109
|
+
runId,
|
|
110
|
+
step,
|
|
111
|
+
eventType,
|
|
112
|
+
agentId: normalizeString(agentId).toLowerCase(),
|
|
113
|
+
message: normalizeString(message),
|
|
114
|
+
usage: {
|
|
115
|
+
outputTokens: Number(usage.outputTokens || 0),
|
|
116
|
+
toolCalls: Number(usage.toolCalls || 0),
|
|
117
|
+
durationMs: Number(usage.durationMs || 0),
|
|
118
|
+
costUsd: Number(usage.costUsd || 0),
|
|
119
|
+
},
|
|
120
|
+
metadata,
|
|
121
|
+
};
|
|
122
|
+
}
|
|
123
|
+
|
|
124
|
+
async function executePlaywrightAction({ page, action, runDirectory, usage, runId, step }) {
|
|
125
|
+
const type = action.type;
|
|
126
|
+
const metadata = {
|
|
127
|
+
action: type,
|
|
128
|
+
};
|
|
129
|
+
if (type === "goto") {
|
|
130
|
+
const url = action.url || "about:blank";
|
|
131
|
+
await page.goto(url, { waitUntil: "domcontentloaded" });
|
|
132
|
+
metadata.url = url;
|
|
133
|
+
usage.toolCalls += 1;
|
|
134
|
+
usage.outputTokens += estimateTokens(`goto:${url}`);
|
|
135
|
+
return createEvent({
|
|
136
|
+
runId,
|
|
137
|
+
step,
|
|
138
|
+
eventType: "tool_call",
|
|
139
|
+
agentId: "omar",
|
|
140
|
+
message: `Playwright goto ${url}`,
|
|
141
|
+
metadata,
|
|
142
|
+
usage,
|
|
143
|
+
});
|
|
144
|
+
}
|
|
145
|
+
if (type === "click") {
|
|
146
|
+
if (!action.selector) {
|
|
147
|
+
throw new Error("Playbook click action requires selector.");
|
|
148
|
+
}
|
|
149
|
+
await page.click(action.selector);
|
|
150
|
+
metadata.selector = action.selector;
|
|
151
|
+
usage.toolCalls += 1;
|
|
152
|
+
usage.outputTokens += estimateTokens(`click:${action.selector}`);
|
|
153
|
+
return createEvent({
|
|
154
|
+
runId,
|
|
155
|
+
step,
|
|
156
|
+
eventType: "tool_call",
|
|
157
|
+
agentId: "omar",
|
|
158
|
+
message: `Playwright click ${action.selector}`,
|
|
159
|
+
metadata,
|
|
160
|
+
usage,
|
|
161
|
+
});
|
|
162
|
+
}
|
|
163
|
+
if (type === "fill") {
|
|
164
|
+
if (!action.selector) {
|
|
165
|
+
throw new Error("Playbook fill action requires selector.");
|
|
166
|
+
}
|
|
167
|
+
await page.fill(action.selector, action.text || "");
|
|
168
|
+
metadata.selector = action.selector;
|
|
169
|
+
usage.toolCalls += 1;
|
|
170
|
+
usage.outputTokens += estimateTokens(`fill:${action.selector}:${action.text || ""}`);
|
|
171
|
+
return createEvent({
|
|
172
|
+
runId,
|
|
173
|
+
step,
|
|
174
|
+
eventType: "tool_call",
|
|
175
|
+
agentId: "omar",
|
|
176
|
+
message: `Playwright fill ${action.selector}`,
|
|
177
|
+
metadata,
|
|
178
|
+
usage,
|
|
179
|
+
});
|
|
180
|
+
}
|
|
181
|
+
if (type === "wait") {
|
|
182
|
+
const waitMs = Math.max(0, Number(action.ms || 0));
|
|
183
|
+
await page.waitForTimeout(waitMs);
|
|
184
|
+
metadata.ms = waitMs;
|
|
185
|
+
usage.toolCalls += 1;
|
|
186
|
+
usage.outputTokens += estimateTokens(`wait:${waitMs}`);
|
|
187
|
+
return createEvent({
|
|
188
|
+
runId,
|
|
189
|
+
step,
|
|
190
|
+
eventType: "tool_call",
|
|
191
|
+
agentId: "omar",
|
|
192
|
+
message: `Playwright wait ${waitMs}ms`,
|
|
193
|
+
metadata,
|
|
194
|
+
usage,
|
|
195
|
+
});
|
|
196
|
+
}
|
|
197
|
+
if (type === "screenshot") {
|
|
198
|
+
const outputPath = action.path
|
|
199
|
+
? path.resolve(runDirectory, action.path)
|
|
200
|
+
: path.join(runDirectory, "runtime", `step-${String(step).padStart(3, "0")}.png`);
|
|
201
|
+
await fsp.mkdir(path.dirname(outputPath), { recursive: true });
|
|
202
|
+
await page.screenshot({
|
|
203
|
+
path: outputPath,
|
|
204
|
+
fullPage: true,
|
|
205
|
+
});
|
|
206
|
+
metadata.path = toPosixPath(path.relative(runDirectory, outputPath));
|
|
207
|
+
usage.toolCalls += 1;
|
|
208
|
+
usage.outputTokens += estimateTokens(`screenshot:${metadata.path}`);
|
|
209
|
+
return createEvent({
|
|
210
|
+
runId,
|
|
211
|
+
step,
|
|
212
|
+
eventType: "tool_call",
|
|
213
|
+
agentId: "omar",
|
|
214
|
+
message: `Playwright screenshot ${metadata.path}`,
|
|
215
|
+
metadata,
|
|
216
|
+
usage,
|
|
217
|
+
});
|
|
218
|
+
}
|
|
219
|
+
|
|
220
|
+
usage.toolCalls += 1;
|
|
221
|
+
usage.outputTokens += estimateTokens(`unsupported:${type}`);
|
|
222
|
+
return createEvent({
|
|
223
|
+
runId,
|
|
224
|
+
step,
|
|
225
|
+
eventType: "tool_call",
|
|
226
|
+
agentId: "omar",
|
|
227
|
+
message: `Unsupported Playwright action skipped: ${type}`,
|
|
228
|
+
metadata: {
|
|
229
|
+
action: type,
|
|
230
|
+
skipped: true,
|
|
231
|
+
},
|
|
232
|
+
usage,
|
|
233
|
+
});
|
|
234
|
+
}
|
|
235
|
+
|
|
236
|
+
function buildRuntimeMarkdown(summary = {}) {
|
|
237
|
+
return `# SWARM_RUNTIME
|
|
238
|
+
|
|
239
|
+
Generated: ${summary.generatedAt}
|
|
240
|
+
Run ID: ${summary.runId}
|
|
241
|
+
Plan run ID: ${summary.planRunId}
|
|
242
|
+
Target: ${summary.targetPath}
|
|
243
|
+
Scenario: ${summary.scenario}
|
|
244
|
+
Engine: ${summary.engine}
|
|
245
|
+
Execute: ${summary.execute ? "yes" : "no"}
|
|
246
|
+
|
|
247
|
+
Status:
|
|
248
|
+
- completed: ${summary.completed ? "yes" : "no"}
|
|
249
|
+
- stop_class: ${summary.stop?.stopClass || "NONE"}
|
|
250
|
+
- stop_reason: ${summary.stop?.reason || "none"}
|
|
251
|
+
|
|
252
|
+
Usage:
|
|
253
|
+
- output_tokens: ${summary.usage.outputTokens}
|
|
254
|
+
- tool_calls: ${summary.usage.toolCalls}
|
|
255
|
+
- duration_ms: ${summary.usage.durationMs}
|
|
256
|
+
- cost_usd: ${summary.usage.costUsd}
|
|
257
|
+
|
|
258
|
+
Artifacts:
|
|
259
|
+
- events: ${summary.runtimeEventsPath}
|
|
260
|
+
- summary_json: ${summary.runtimeJsonPath}
|
|
261
|
+
`;
|
|
262
|
+
}
|
|
263
|
+
|
|
264
|
+
async function writeRuntimeArtifacts({
|
|
265
|
+
summary,
|
|
266
|
+
events,
|
|
267
|
+
outputDir = "",
|
|
268
|
+
env,
|
|
269
|
+
homeDir,
|
|
270
|
+
} = {}) {
|
|
271
|
+
const outputRoot = await resolveOutputRoot({
|
|
272
|
+
cwd: path.resolve(String(summary.targetPath || ".")),
|
|
273
|
+
outputDirOverride: outputDir,
|
|
274
|
+
env,
|
|
275
|
+
homeDir,
|
|
276
|
+
});
|
|
277
|
+
const runDirectory = path.join(outputRoot, "swarms", summary.runId);
|
|
278
|
+
const runtimeDirectory = path.join(runDirectory, "runtime");
|
|
279
|
+
const runtimeJsonPath = path.join(runtimeDirectory, "SWARM_RUNTIME.json");
|
|
280
|
+
const runtimeMarkdownPath = path.join(runtimeDirectory, "SWARM_RUNTIME.md");
|
|
281
|
+
const runtimeEventsPath = path.join(runtimeDirectory, "events.ndjson");
|
|
282
|
+
await fsp.mkdir(runtimeDirectory, { recursive: true });
|
|
283
|
+
await fsp.writeFile(runtimeEventsPath, `${events.map((event) => JSON.stringify(event)).join("\n")}\n`, "utf-8");
|
|
284
|
+
|
|
285
|
+
const fullSummary = {
|
|
286
|
+
...summary,
|
|
287
|
+
outputRoot,
|
|
288
|
+
runDirectory,
|
|
289
|
+
runtimeDirectory,
|
|
290
|
+
runtimeJsonPath,
|
|
291
|
+
runtimeMarkdownPath,
|
|
292
|
+
runtimeEventsPath,
|
|
293
|
+
};
|
|
294
|
+
|
|
295
|
+
await fsp.writeFile(runtimeJsonPath, `${JSON.stringify(fullSummary, null, 2)}\n`, "utf-8");
|
|
296
|
+
await fsp.writeFile(runtimeMarkdownPath, `${buildRuntimeMarkdown(fullSummary).trim()}\n`, "utf-8");
|
|
297
|
+
|
|
298
|
+
return fullSummary;
|
|
299
|
+
}
|
|
300
|
+
|
|
301
|
+
export async function runSwarmRuntime({
|
|
302
|
+
plan,
|
|
303
|
+
targetPath,
|
|
304
|
+
engine = "mock",
|
|
305
|
+
execute = false,
|
|
306
|
+
maxSteps = 20,
|
|
307
|
+
startUrl = "about:blank",
|
|
308
|
+
playbookActions = [],
|
|
309
|
+
outputDir = "",
|
|
310
|
+
env,
|
|
311
|
+
} = {}) {
|
|
312
|
+
if (!plan || typeof plan !== "object") {
|
|
313
|
+
throw new Error("plan is required.");
|
|
314
|
+
}
|
|
315
|
+
if (!Array.isArray(plan.assignments) || plan.assignments.length === 0) {
|
|
316
|
+
throw new Error("plan.assignments must include at least one assignment.");
|
|
317
|
+
}
|
|
318
|
+
|
|
319
|
+
const normalizedEngine = normalizeEngine(engine);
|
|
320
|
+
const normalizedTargetPath = path.resolve(String(targetPath || plan.targetPath || "."));
|
|
321
|
+
const normalizedMaxSteps = normalizeMaxSteps(maxSteps);
|
|
322
|
+
const runId = createRuntimeRunId();
|
|
323
|
+
const resolvedOutputRoot = await resolveOutputRoot({
|
|
324
|
+
cwd: normalizedTargetPath,
|
|
325
|
+
outputDirOverride: outputDir,
|
|
326
|
+
env,
|
|
327
|
+
});
|
|
328
|
+
const runtimeRunDirectory = path.join(resolvedOutputRoot, "swarms", runId);
|
|
329
|
+
const runStartedAt = Date.now();
|
|
330
|
+
const events = [];
|
|
331
|
+
let step = 0;
|
|
332
|
+
|
|
333
|
+
const usage = {
|
|
334
|
+
outputTokens: 0,
|
|
335
|
+
toolCalls: 0,
|
|
336
|
+
durationMs: 0,
|
|
337
|
+
costUsd: 0,
|
|
338
|
+
};
|
|
339
|
+
let stop = {
|
|
340
|
+
stopClass: "NONE",
|
|
341
|
+
reason: "",
|
|
342
|
+
blocking: false,
|
|
343
|
+
};
|
|
344
|
+
|
|
345
|
+
events.push(
|
|
346
|
+
createEvent({
|
|
347
|
+
runId,
|
|
348
|
+
step,
|
|
349
|
+
eventType: "run_start",
|
|
350
|
+
agentId: "omar",
|
|
351
|
+
message: `Swarm runtime started with engine=${normalizedEngine}, execute=${Boolean(execute)}`,
|
|
352
|
+
metadata: {
|
|
353
|
+
planRunId: normalizeString(plan.runId),
|
|
354
|
+
scenario: normalizeString(plan.scenario),
|
|
355
|
+
},
|
|
356
|
+
usage,
|
|
357
|
+
})
|
|
358
|
+
);
|
|
359
|
+
|
|
360
|
+
let browser = null;
|
|
361
|
+
let page = null;
|
|
362
|
+
let playwrightActions = playbookActions;
|
|
363
|
+
|
|
364
|
+
try {
|
|
365
|
+
if (normalizedEngine === "playwright" && execute) {
|
|
366
|
+
const playwright = await import("playwright");
|
|
367
|
+
browser = await playwright.chromium.launch({
|
|
368
|
+
headless: true,
|
|
369
|
+
});
|
|
370
|
+
page = await browser.newPage();
|
|
371
|
+
await page.goto(startUrl || "about:blank", { waitUntil: "domcontentloaded" });
|
|
372
|
+
usage.toolCalls += 1;
|
|
373
|
+
usage.outputTokens += estimateTokens(`goto:${startUrl || "about:blank"}`);
|
|
374
|
+
step += 1;
|
|
375
|
+
events.push(
|
|
376
|
+
createEvent({
|
|
377
|
+
runId,
|
|
378
|
+
step,
|
|
379
|
+
eventType: "tool_call",
|
|
380
|
+
agentId: "omar",
|
|
381
|
+
message: `Playwright runtime initialized at ${startUrl || "about:blank"}`,
|
|
382
|
+
metadata: {
|
|
383
|
+
action: "goto",
|
|
384
|
+
url: startUrl || "about:blank",
|
|
385
|
+
},
|
|
386
|
+
usage,
|
|
387
|
+
})
|
|
388
|
+
);
|
|
389
|
+
}
|
|
390
|
+
|
|
391
|
+
for (const assignment of plan.assignments) {
|
|
392
|
+
if (step >= normalizedMaxSteps) {
|
|
393
|
+
stop = {
|
|
394
|
+
stopClass: "MAX_STEPS_EXCEEDED",
|
|
395
|
+
reason: `max-steps reached (${normalizedMaxSteps})`,
|
|
396
|
+
blocking: true,
|
|
397
|
+
};
|
|
398
|
+
break;
|
|
399
|
+
}
|
|
400
|
+
|
|
401
|
+
step += 1;
|
|
402
|
+
usage.outputTokens += estimateTokens(assignment.objective);
|
|
403
|
+
events.push(
|
|
404
|
+
createEvent({
|
|
405
|
+
runId,
|
|
406
|
+
step,
|
|
407
|
+
eventType: "run_step",
|
|
408
|
+
agentId: assignment.agentId,
|
|
409
|
+
message: `Assignment started: ${assignment.objective}`,
|
|
410
|
+
metadata: {
|
|
411
|
+
assignmentId: assignment.assignmentId,
|
|
412
|
+
role: assignment.role,
|
|
413
|
+
domain: assignment.domain,
|
|
414
|
+
},
|
|
415
|
+
usage,
|
|
416
|
+
})
|
|
417
|
+
);
|
|
418
|
+
|
|
419
|
+
if (normalizedEngine === "mock" || !execute) {
|
|
420
|
+
usage.toolCalls += 1;
|
|
421
|
+
usage.outputTokens += estimateTokens(`mock:${assignment.agentId}`);
|
|
422
|
+
step += 1;
|
|
423
|
+
events.push(
|
|
424
|
+
createEvent({
|
|
425
|
+
runId,
|
|
426
|
+
step,
|
|
427
|
+
eventType: "tool_call",
|
|
428
|
+
agentId: assignment.agentId,
|
|
429
|
+
message: `Mock runtime action completed for ${assignment.agentId}`,
|
|
430
|
+
metadata: {
|
|
431
|
+
engine: normalizedEngine,
|
|
432
|
+
execute: Boolean(execute),
|
|
433
|
+
},
|
|
434
|
+
usage,
|
|
435
|
+
})
|
|
436
|
+
);
|
|
437
|
+
} else if (normalizedEngine === "playwright" && execute && page) {
|
|
438
|
+
let resolvedActions;
|
|
439
|
+
if (Array.isArray(playwrightActions) && playwrightActions.length > 0) {
|
|
440
|
+
resolvedActions = playwrightActions;
|
|
441
|
+
} else {
|
|
442
|
+
resolvedActions = [{ type: "wait", ms: 250 }, { type: "screenshot" }];
|
|
443
|
+
}
|
|
444
|
+
|
|
445
|
+
for (const action of resolvedActions) {
|
|
446
|
+
if (step >= normalizedMaxSteps) {
|
|
447
|
+
stop = {
|
|
448
|
+
stopClass: "MAX_STEPS_EXCEEDED",
|
|
449
|
+
reason: `max-steps reached (${normalizedMaxSteps})`,
|
|
450
|
+
blocking: true,
|
|
451
|
+
};
|
|
452
|
+
break;
|
|
453
|
+
}
|
|
454
|
+
step += 1;
|
|
455
|
+
const event = await executePlaywrightAction({
|
|
456
|
+
page,
|
|
457
|
+
action,
|
|
458
|
+
runDirectory: runtimeRunDirectory,
|
|
459
|
+
usage,
|
|
460
|
+
runId,
|
|
461
|
+
step,
|
|
462
|
+
});
|
|
463
|
+
events.push(event);
|
|
464
|
+
}
|
|
465
|
+
if (stop.blocking) {
|
|
466
|
+
break;
|
|
467
|
+
}
|
|
468
|
+
}
|
|
469
|
+
|
|
470
|
+
usage.durationMs = Date.now() - runStartedAt;
|
|
471
|
+
usage.costUsd = Number((usage.outputTokens * 0.000003).toFixed(6));
|
|
472
|
+
const budgetStatus = evaluateBudget({
|
|
473
|
+
sessionSummary: {
|
|
474
|
+
costUsd: usage.costUsd,
|
|
475
|
+
outputTokens: usage.outputTokens,
|
|
476
|
+
noProgressStreak: 0,
|
|
477
|
+
durationMs: usage.durationMs,
|
|
478
|
+
toolCalls: usage.toolCalls,
|
|
479
|
+
},
|
|
480
|
+
maxCostUsd: Number(plan.globalBudget?.maxCostUsd || 5),
|
|
481
|
+
maxOutputTokens: Number(plan.globalBudget?.maxOutputTokens || 20000),
|
|
482
|
+
maxNoProgress: Number.MAX_SAFE_INTEGER,
|
|
483
|
+
maxRuntimeMs: Number(plan.globalBudget?.maxRuntimeMs || 3600000),
|
|
484
|
+
maxToolCalls: Number(plan.globalBudget?.maxToolCalls || 500),
|
|
485
|
+
warningThresholdPercent: Number(plan.globalBudget?.warningThresholdPercent || 80),
|
|
486
|
+
});
|
|
487
|
+
if (budgetStatus.blocking) {
|
|
488
|
+
stop = {
|
|
489
|
+
stopClass: String(budgetStatus.reasons[0]?.code || "BUDGET_EXCEEDED"),
|
|
490
|
+
reason: String(budgetStatus.reasons[0]?.message || "Budget exceeded."),
|
|
491
|
+
blocking: true,
|
|
492
|
+
};
|
|
493
|
+
step += 1;
|
|
494
|
+
events.push(
|
|
495
|
+
createEvent({
|
|
496
|
+
runId,
|
|
497
|
+
step,
|
|
498
|
+
eventType: "budget_stop",
|
|
499
|
+
agentId: "omar",
|
|
500
|
+
message: stop.reason,
|
|
501
|
+
metadata: {
|
|
502
|
+
reasonCodes: budgetStatus.reasons.map((reason) => reason.code),
|
|
503
|
+
},
|
|
504
|
+
usage,
|
|
505
|
+
})
|
|
506
|
+
);
|
|
507
|
+
break;
|
|
508
|
+
}
|
|
509
|
+
}
|
|
510
|
+
} finally {
|
|
511
|
+
if (browser) {
|
|
512
|
+
await browser.close();
|
|
513
|
+
}
|
|
514
|
+
}
|
|
515
|
+
|
|
516
|
+
usage.durationMs = Date.now() - runStartedAt;
|
|
517
|
+
usage.costUsd = Number((usage.outputTokens * 0.000003).toFixed(6));
|
|
518
|
+
const completed = !stop.blocking;
|
|
519
|
+
if (completed) {
|
|
520
|
+
step += 1;
|
|
521
|
+
events.push(
|
|
522
|
+
createEvent({
|
|
523
|
+
runId,
|
|
524
|
+
step,
|
|
525
|
+
eventType: "run_stop",
|
|
526
|
+
agentId: "omar",
|
|
527
|
+
message: "Swarm runtime completed successfully.",
|
|
528
|
+
metadata: {
|
|
529
|
+
completed: true,
|
|
530
|
+
},
|
|
531
|
+
usage,
|
|
532
|
+
})
|
|
533
|
+
);
|
|
534
|
+
} else {
|
|
535
|
+
step += 1;
|
|
536
|
+
events.push(
|
|
537
|
+
createEvent({
|
|
538
|
+
runId,
|
|
539
|
+
step,
|
|
540
|
+
eventType: "run_stop",
|
|
541
|
+
agentId: "omar",
|
|
542
|
+
message: stop.reason || "Swarm runtime stopped.",
|
|
543
|
+
metadata: {
|
|
544
|
+
completed: false,
|
|
545
|
+
stopClass: stop.stopClass,
|
|
546
|
+
},
|
|
547
|
+
usage,
|
|
548
|
+
})
|
|
549
|
+
);
|
|
550
|
+
}
|
|
551
|
+
|
|
552
|
+
const summary = {
|
|
553
|
+
schemaVersion: 1,
|
|
554
|
+
generatedAt: new Date().toISOString(),
|
|
555
|
+
runId,
|
|
556
|
+
planRunId: normalizeString(plan.runId),
|
|
557
|
+
targetPath: normalizedTargetPath,
|
|
558
|
+
scenario: normalizeString(plan.scenario || "qa_audit"),
|
|
559
|
+
engine: normalizedEngine,
|
|
560
|
+
execute: Boolean(execute),
|
|
561
|
+
startUrl: normalizeString(startUrl || "about:blank"),
|
|
562
|
+
maxSteps: normalizedMaxSteps,
|
|
563
|
+
completed,
|
|
564
|
+
stop,
|
|
565
|
+
usage,
|
|
566
|
+
eventCount: events.length,
|
|
567
|
+
selectedAgents: Array.isArray(plan.selectedAgents) ? [...plan.selectedAgents] : [],
|
|
568
|
+
};
|
|
569
|
+
|
|
570
|
+
return writeRuntimeArtifacts({
|
|
571
|
+
summary,
|
|
572
|
+
events,
|
|
573
|
+
outputDir,
|
|
574
|
+
env,
|
|
575
|
+
});
|
|
576
|
+
}
|