@agentmeshhq/agent 0.1.12 → 0.1.14
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/__tests__/loader.test.js +44 -1
- package/dist/__tests__/loader.test.js.map +1 -1
- package/dist/__tests__/runner.test.js.map +1 -1
- package/dist/__tests__/sandbox.test.d.ts +1 -0
- package/dist/__tests__/sandbox.test.js +362 -0
- package/dist/__tests__/sandbox.test.js.map +1 -0
- package/dist/__tests__/watchdog.test.d.ts +1 -0
- package/dist/__tests__/watchdog.test.js +290 -0
- package/dist/__tests__/watchdog.test.js.map +1 -0
- package/dist/cli/attach.js +20 -1
- package/dist/cli/attach.js.map +1 -1
- package/dist/cli/build.js +8 -2
- package/dist/cli/build.js.map +1 -1
- package/dist/cli/context.js.map +1 -1
- package/dist/cli/deploy.js +1 -1
- package/dist/cli/deploy.js.map +1 -1
- package/dist/cli/inbox.d.ts +5 -0
- package/dist/cli/inbox.js +123 -0
- package/dist/cli/inbox.js.map +1 -0
- package/dist/cli/index.js +5 -1
- package/dist/cli/index.js.map +1 -1
- package/dist/cli/init.js +1 -1
- package/dist/cli/init.js.map +1 -1
- package/dist/cli/issue.d.ts +42 -0
- package/dist/cli/issue.js +297 -0
- package/dist/cli/issue.js.map +1 -0
- package/dist/cli/list.js +3 -3
- package/dist/cli/list.js.map +1 -1
- package/dist/cli/local.js +5 -3
- package/dist/cli/local.js.map +1 -1
- package/dist/cli/migrate.js +1 -1
- package/dist/cli/migrate.js.map +1 -1
- package/dist/cli/nudge.js +16 -3
- package/dist/cli/nudge.js.map +1 -1
- package/dist/cli/ready.d.ts +5 -0
- package/dist/cli/ready.js +131 -0
- package/dist/cli/ready.js.map +1 -0
- package/dist/cli/restart.js.map +1 -1
- package/dist/cli/slack.js +1 -1
- package/dist/cli/slack.js.map +1 -1
- package/dist/cli/start.d.ts +8 -0
- package/dist/cli/start.js +9 -0
- package/dist/cli/start.js.map +1 -1
- package/dist/cli/stop.js +13 -5
- package/dist/cli/stop.js.map +1 -1
- package/dist/cli/sync.d.ts +8 -0
- package/dist/cli/sync.js +154 -0
- package/dist/cli/sync.js.map +1 -0
- package/dist/cli/test.js +1 -1
- package/dist/cli/test.js.map +1 -1
- package/dist/cli/token.js +2 -2
- package/dist/cli/token.js.map +1 -1
- package/dist/config/loader.d.ts +5 -1
- package/dist/config/loader.js +27 -2
- package/dist/config/loader.js.map +1 -1
- package/dist/config/schema.d.ts +13 -0
- package/dist/core/daemon.d.ts +50 -0
- package/dist/core/daemon.js +445 -11
- package/dist/core/daemon.js.map +1 -1
- package/dist/core/injector.d.ts +2 -2
- package/dist/core/injector.js +23 -4
- package/dist/core/injector.js.map +1 -1
- package/dist/core/issue-cache.d.ts +44 -0
- package/dist/core/issue-cache.js +75 -0
- package/dist/core/issue-cache.js.map +1 -0
- package/dist/core/registry.d.ts +5 -0
- package/dist/core/registry.js +8 -1
- package/dist/core/registry.js.map +1 -1
- package/dist/core/runner.d.ts +1 -1
- package/dist/core/runner.js +23 -1
- package/dist/core/runner.js.map +1 -1
- package/dist/core/sandbox.d.ts +138 -0
- package/dist/core/sandbox.js +409 -0
- package/dist/core/sandbox.js.map +1 -0
- package/dist/core/tmux.d.ts +8 -0
- package/dist/core/tmux.js +28 -1
- package/dist/core/tmux.js.map +1 -1
- package/dist/core/watchdog.d.ts +41 -0
- package/dist/core/watchdog.js +198 -0
- package/dist/core/watchdog.js.map +1 -0
- package/dist/core/websocket.js +1 -1
- package/dist/core/websocket.js.map +1 -1
- package/dist/index.d.ts +5 -5
- package/dist/index.js +5 -5
- package/dist/index.js.map +1 -1
- package/package.json +1 -1
- package/src/__tests__/loader.test.ts +52 -4
- package/src/__tests__/runner.test.ts +1 -2
- package/src/__tests__/sandbox.test.ts +435 -0
- package/src/__tests__/watchdog.test.ts +368 -0
- package/src/cli/attach.ts +22 -1
- package/src/cli/build.ts +12 -4
- package/src/cli/context.ts +0 -1
- package/src/cli/deploy.ts +7 -5
- package/src/cli/index.ts +8 -1
- package/src/cli/init.ts +7 -19
- package/src/cli/list.ts +6 -10
- package/src/cli/local.ts +21 -12
- package/src/cli/migrate.ts +6 -4
- package/src/cli/nudge.ts +29 -14
- package/src/cli/restart.ts +1 -1
- package/src/cli/slack.ts +16 -15
- package/src/cli/start.ts +14 -0
- package/src/cli/stop.ts +14 -5
- package/src/cli/test.ts +5 -3
- package/src/cli/token.ts +4 -4
- package/src/config/loader.ts +29 -2
- package/src/config/schema.ts +14 -0
- package/src/core/daemon.ts +540 -17
- package/src/core/injector.ts +27 -4
- package/src/core/registry.ts +14 -1
- package/src/core/runner.ts +26 -1
- package/src/core/sandbox.ts +550 -0
- package/src/core/tmux.ts +35 -2
- package/src/core/watchdog.ts +238 -0
- package/src/core/websocket.ts +2 -2
- package/src/index.ts +6 -5
package/dist/core/daemon.d.ts
CHANGED
|
@@ -12,6 +12,14 @@ export interface DaemonOptions {
|
|
|
12
12
|
serve?: boolean;
|
|
13
13
|
/** Port for opencode serve (default: 3001) */
|
|
14
14
|
servePort?: number;
|
|
15
|
+
/** Run agent in Docker sandbox container */
|
|
16
|
+
sandbox?: boolean;
|
|
17
|
+
/** Docker image for sandbox (default: agentmesh/agent-sandbox:latest) */
|
|
18
|
+
sandboxImage?: string;
|
|
19
|
+
/** CPU limit for sandbox (default: 1) */
|
|
20
|
+
sandboxCpu?: string;
|
|
21
|
+
/** Memory limit for sandbox (default: 2g) */
|
|
22
|
+
sandboxMemory?: string;
|
|
15
23
|
}
|
|
16
24
|
export declare class AgentDaemon {
|
|
17
25
|
private agentName;
|
|
@@ -29,14 +37,51 @@ export declare class AgentDaemon {
|
|
|
29
37
|
private serveMode;
|
|
30
38
|
private servePort;
|
|
31
39
|
private serveProcess;
|
|
40
|
+
private sandboxMode;
|
|
41
|
+
private sandboxImage;
|
|
42
|
+
private sandboxCpu;
|
|
43
|
+
private sandboxMemory;
|
|
44
|
+
private sandbox;
|
|
45
|
+
private healthCheckInterval;
|
|
46
|
+
private serverContext;
|
|
47
|
+
private restartCount;
|
|
48
|
+
private lastStableTime;
|
|
49
|
+
private stuckSince;
|
|
50
|
+
private nudgeSentAt;
|
|
32
51
|
constructor(options: DaemonOptions);
|
|
33
52
|
start(): Promise<void>;
|
|
53
|
+
/**
|
|
54
|
+
* Starts periodic health monitoring for the tmux session
|
|
55
|
+
* Includes auto-restart logic and progress watchdog
|
|
56
|
+
*/
|
|
57
|
+
private startHealthMonitor;
|
|
58
|
+
/**
|
|
59
|
+
* Handles session death - logs crash and attempts auto-restart
|
|
60
|
+
*/
|
|
61
|
+
private handleSessionDeath;
|
|
62
|
+
/**
|
|
63
|
+
* Handles stuck agent - sends nudge first, then restarts if still stuck
|
|
64
|
+
*/
|
|
65
|
+
private handleStuckAgent;
|
|
66
|
+
/**
|
|
67
|
+
* Restarts the agent session (sandbox or non-sandbox)
|
|
68
|
+
*/
|
|
69
|
+
private restartSession;
|
|
34
70
|
stop(): Promise<void>;
|
|
35
71
|
/**
|
|
36
72
|
* Starts opencode serve mode (for Integration Service)
|
|
37
73
|
* Replaces tmux with a direct HTTP server
|
|
38
74
|
*/
|
|
39
75
|
private startServeMode;
|
|
76
|
+
/**
|
|
77
|
+
* Starts agent in Docker sandbox mode
|
|
78
|
+
* Provides filesystem isolation with only workspace mounted
|
|
79
|
+
*
|
|
80
|
+
* Strategy: Start Docker container with tail -f /dev/null, then create
|
|
81
|
+
* a tmux session on the HOST that runs `docker exec -it <container> opencode`.
|
|
82
|
+
* This way tmux provides the TTY that docker exec needs.
|
|
83
|
+
*/
|
|
84
|
+
private startSandboxMode;
|
|
40
85
|
/**
|
|
41
86
|
* Saves the current agent context to disk
|
|
42
87
|
*/
|
|
@@ -51,4 +96,9 @@ export declare class AgentDaemon {
|
|
|
51
96
|
* Returns the absolute path to the workspace
|
|
52
97
|
*/
|
|
53
98
|
private setupWorkspace;
|
|
99
|
+
/**
|
|
100
|
+
* Ensures the sandbox OpenCode config exists
|
|
101
|
+
* Creates ~/.agentmesh/opencode-sandbox.json with permissive permissions and model
|
|
102
|
+
*/
|
|
103
|
+
private ensureSandboxOpencodeConfig;
|
|
54
104
|
}
|
package/dist/core/daemon.js
CHANGED
|
@@ -2,14 +2,29 @@ import { execSync, spawn } from "node:child_process";
|
|
|
2
2
|
import fs from "node:fs";
|
|
3
3
|
import os from "node:os";
|
|
4
4
|
import path from "node:path";
|
|
5
|
-
import { addAgentToState, getAgentState, loadConfig, updateAgentInState, } from "../config/loader.js";
|
|
5
|
+
import { addAgentToState, getAgentState, loadConfig, resetAgentRestartCount, updateAgentInState, } from "../config/loader.js";
|
|
6
6
|
import { loadContext, loadOrCreateContext, saveContext } from "../context/index.js";
|
|
7
7
|
import { Heartbeat } from "./heartbeat.js";
|
|
8
8
|
import { handleWebSocketEvent, injectRestoredContext, injectStartupMessage } from "./injector.js";
|
|
9
9
|
import { checkInbox, fetchAssignments, registerAgent } from "./registry.js";
|
|
10
|
-
import { buildRunnerConfig, getRunnerDisplayName
|
|
11
|
-
import {
|
|
10
|
+
import { buildRunnerConfig, getRunnerDisplayName } from "./runner.js";
|
|
11
|
+
import { DockerSandbox } from "./sandbox.js";
|
|
12
|
+
import { captureSessionContext, captureSessionOutput, createSession, destroySession, getSessionName, isSessionHealthy, sessionExists, updateSessionEnvironment, } from "./tmux.js";
|
|
13
|
+
import { checkAgentProgress, cleanupOrphanContainers, isProcessRunning, sendNudge, } from "./watchdog.js";
|
|
12
14
|
import { AgentWebSocket } from "./websocket.js";
|
|
15
|
+
// Maximum number of auto-restart attempts
|
|
16
|
+
const MAX_RESTART_ATTEMPTS = 3;
|
|
17
|
+
// Time after which restart count resets (30 minutes of stable operation)
|
|
18
|
+
const RESTART_COUNT_RESET_MS = 30 * 60 * 1000;
|
|
19
|
+
// Time to wait after nudging before restarting (2 minutes)
|
|
20
|
+
const NUDGE_WAIT_MS = 2 * 60 * 1000;
|
|
21
|
+
// Path to the sandbox OpenCode config (permissive permissions)
|
|
22
|
+
const SANDBOX_OPENCODE_CONFIG_PATH = path.join(os.homedir(), ".agentmesh", "opencode-sandbox.json");
|
|
23
|
+
// Sandbox OpenCode config content - allow everything since container is sandboxed
|
|
24
|
+
const SANDBOX_OPENCODE_CONFIG = {
|
|
25
|
+
$schema: "https://opencode.ai/config.json",
|
|
26
|
+
permission: "allow",
|
|
27
|
+
};
|
|
13
28
|
export class AgentDaemon {
|
|
14
29
|
agentName;
|
|
15
30
|
config;
|
|
@@ -26,11 +41,28 @@ export class AgentDaemon {
|
|
|
26
41
|
serveMode;
|
|
27
42
|
servePort;
|
|
28
43
|
serveProcess = null;
|
|
44
|
+
sandboxMode;
|
|
45
|
+
sandboxImage;
|
|
46
|
+
sandboxCpu;
|
|
47
|
+
sandboxMemory;
|
|
48
|
+
sandbox = null;
|
|
49
|
+
healthCheckInterval = null;
|
|
50
|
+
serverContext;
|
|
51
|
+
// Auto-restart tracking
|
|
52
|
+
restartCount = 0;
|
|
53
|
+
lastStableTime = null;
|
|
54
|
+
stuckSince = null;
|
|
55
|
+
nudgeSentAt = null;
|
|
29
56
|
constructor(options) {
|
|
30
57
|
const config = loadConfig();
|
|
31
58
|
if (!config) {
|
|
32
59
|
throw new Error("No config found. Run 'agentmesh init' first.");
|
|
33
60
|
}
|
|
61
|
+
// Ensure config has required fields with defaults
|
|
62
|
+
if (!config.agents)
|
|
63
|
+
config.agents = [];
|
|
64
|
+
if (!config.defaults)
|
|
65
|
+
config.defaults = { command: "opencode", model: "claude-sonnet-4" };
|
|
34
66
|
this.config = config;
|
|
35
67
|
this.agentName = options.name;
|
|
36
68
|
this.shouldRestoreContext = options.restoreContext !== false;
|
|
@@ -55,6 +87,10 @@ export class AgentDaemon {
|
|
|
55
87
|
this.agentConfig = agentConfig;
|
|
56
88
|
this.serveMode = options.serve === true;
|
|
57
89
|
this.servePort = options.servePort || 3001;
|
|
90
|
+
this.sandboxMode = options.sandbox === true;
|
|
91
|
+
this.sandboxImage = options.sandboxImage || "agentmesh/agent-sandbox:latest";
|
|
92
|
+
this.sandboxCpu = options.sandboxCpu || "1";
|
|
93
|
+
this.sandboxMemory = options.sandboxMemory || "2g";
|
|
58
94
|
// Build runner configuration with model resolution
|
|
59
95
|
this.runnerConfig = buildRunnerConfig({
|
|
60
96
|
cliModel: options.model,
|
|
@@ -72,9 +108,28 @@ export class AgentDaemon {
|
|
|
72
108
|
return;
|
|
73
109
|
}
|
|
74
110
|
console.log(`Starting agent: ${this.agentName}`);
|
|
111
|
+
// Check for duplicate process
|
|
112
|
+
const existingState = getAgentState(this.agentName);
|
|
113
|
+
if (existingState && existingState.pid > 0) {
|
|
114
|
+
if (isProcessRunning(existingState.pid)) {
|
|
115
|
+
throw new Error(`Agent "${this.agentName}" is already running (PID: ${existingState.pid}). ` +
|
|
116
|
+
`Use 'agentmesh stop ${this.agentName}' first.`);
|
|
117
|
+
}
|
|
118
|
+
// Process not running, clean up stale state
|
|
119
|
+
console.log(`Cleaning up stale state for PID ${existingState.pid}`);
|
|
120
|
+
}
|
|
121
|
+
// Clean up orphan containers in sandbox mode
|
|
122
|
+
if (this.sandboxMode) {
|
|
123
|
+
const cleaned = cleanupOrphanContainers(this.agentName);
|
|
124
|
+
if (cleaned > 0) {
|
|
125
|
+
console.log(`Cleaned up ${cleaned} orphan container(s)`);
|
|
126
|
+
}
|
|
127
|
+
}
|
|
128
|
+
// Reset restart count on manual start
|
|
129
|
+
this.restartCount = 0;
|
|
130
|
+
this.lastStableTime = new Date();
|
|
75
131
|
// Register with hub first (needed for assignment check)
|
|
76
132
|
console.log("Registering with AgentMesh hub...");
|
|
77
|
-
const existingState = getAgentState(this.agentName);
|
|
78
133
|
console.log(`Existing state: ${existingState ? `agentId=${existingState.agentId}` : "none"}`);
|
|
79
134
|
const registration = await registerAgent({
|
|
80
135
|
url: this.config.hubUrl,
|
|
@@ -83,14 +138,27 @@ export class AgentDaemon {
|
|
|
83
138
|
agentId: existingState?.agentId || this.agentConfig.agentId,
|
|
84
139
|
agentName: this.agentName,
|
|
85
140
|
model: this.agentConfig.model || this.config.defaults.model,
|
|
141
|
+
restoreContext: this.shouldRestoreContext,
|
|
86
142
|
});
|
|
87
143
|
this.agentId = registration.agentId;
|
|
88
144
|
this.token = registration.token;
|
|
89
|
-
|
|
145
|
+
if (registration.status === "re-registered") {
|
|
146
|
+
console.log(`Re-registered as: ${this.agentId}`);
|
|
147
|
+
if (registration.context && Object.keys(registration.context).length > 0) {
|
|
148
|
+
this.serverContext = registration.context;
|
|
149
|
+
console.log(`Server context restored: ${Object.keys(registration.context).join(", ")}`);
|
|
150
|
+
}
|
|
151
|
+
}
|
|
152
|
+
else {
|
|
153
|
+
console.log(`Registered as: ${this.agentId}`);
|
|
154
|
+
}
|
|
90
155
|
// Check assignments and auto-setup workdir if needed (before creating tmux session)
|
|
91
156
|
await this.checkAssignments();
|
|
92
|
-
//
|
|
93
|
-
if (this.
|
|
157
|
+
// Choose runtime mode: sandbox > serve > tmux
|
|
158
|
+
if (this.sandboxMode) {
|
|
159
|
+
await this.startSandboxMode();
|
|
160
|
+
}
|
|
161
|
+
else if (this.serveMode) {
|
|
94
162
|
await this.startServeMode();
|
|
95
163
|
}
|
|
96
164
|
else {
|
|
@@ -131,6 +199,7 @@ export class AgentDaemon {
|
|
|
131
199
|
assignedProject: this.assignedProject,
|
|
132
200
|
runtimeModel: this.runnerConfig.model,
|
|
133
201
|
runnerType: this.runnerConfig.type,
|
|
202
|
+
sandboxContainer: this.sandbox?.getContainerName(),
|
|
134
203
|
});
|
|
135
204
|
// Start heartbeat with auto-refresh
|
|
136
205
|
console.log("Starting heartbeat...");
|
|
@@ -211,11 +280,11 @@ export class AgentDaemon {
|
|
|
211
280
|
},
|
|
212
281
|
});
|
|
213
282
|
this.ws.connect();
|
|
214
|
-
// Check inbox and auto-nudge
|
|
283
|
+
// Check inbox and auto-nudge with full handoff details
|
|
215
284
|
console.log("Checking inbox...");
|
|
216
285
|
try {
|
|
217
286
|
const inboxItems = await checkInbox(this.config.hubUrl, this.config.workspace, this.token);
|
|
218
|
-
injectStartupMessage(this.agentName, inboxItems.length);
|
|
287
|
+
injectStartupMessage(this.agentName, inboxItems.length, inboxItems);
|
|
219
288
|
}
|
|
220
289
|
catch (error) {
|
|
221
290
|
console.error("Failed to check inbox:", error);
|
|
@@ -233,6 +302,8 @@ export class AgentDaemon {
|
|
|
233
302
|
}
|
|
234
303
|
}
|
|
235
304
|
this.isRunning = true;
|
|
305
|
+
// Start session health monitoring (every 60 seconds)
|
|
306
|
+
this.startHealthMonitor();
|
|
236
307
|
console.log(`
|
|
237
308
|
Agent "${this.agentName}" is running.
|
|
238
309
|
|
|
@@ -249,9 +320,227 @@ Nudge agent:
|
|
|
249
320
|
process.on("SIGINT", () => this.stop());
|
|
250
321
|
process.on("SIGTERM", () => this.stop());
|
|
251
322
|
}
|
|
323
|
+
/**
|
|
324
|
+
* Starts periodic health monitoring for the tmux session
|
|
325
|
+
* Includes auto-restart logic and progress watchdog
|
|
326
|
+
*/
|
|
327
|
+
startHealthMonitor() {
|
|
328
|
+
// Skip health monitoring for serve mode (no tmux session)
|
|
329
|
+
if (this.serveMode)
|
|
330
|
+
return;
|
|
331
|
+
const logDir = path.join(os.homedir(), ".agentmesh", "logs");
|
|
332
|
+
if (!fs.existsSync(logDir)) {
|
|
333
|
+
fs.mkdirSync(logDir, { recursive: true });
|
|
334
|
+
}
|
|
335
|
+
this.healthCheckInterval = setInterval(async () => {
|
|
336
|
+
if (!this.isRunning)
|
|
337
|
+
return;
|
|
338
|
+
// Reset restart count after stable operation
|
|
339
|
+
if (this.lastStableTime && this.restartCount > 0) {
|
|
340
|
+
const stableTime = Date.now() - this.lastStableTime.getTime();
|
|
341
|
+
if (stableTime > RESTART_COUNT_RESET_MS) {
|
|
342
|
+
console.log(`[HEALTH] Agent stable for 30+ minutes, resetting restart count`);
|
|
343
|
+
this.restartCount = 0;
|
|
344
|
+
resetAgentRestartCount(this.agentName);
|
|
345
|
+
}
|
|
346
|
+
}
|
|
347
|
+
const health = isSessionHealthy(this.agentName);
|
|
348
|
+
if (!health.healthy) {
|
|
349
|
+
// Session died - attempt restart
|
|
350
|
+
await this.handleSessionDeath(health.reason || "unknown", logDir);
|
|
351
|
+
return;
|
|
352
|
+
}
|
|
353
|
+
// Session is alive - check progress watchdog
|
|
354
|
+
const containerName = this.sandboxMode ? this.sandbox?.getContainerName() : undefined;
|
|
355
|
+
const progress = checkAgentProgress(this.agentName, containerName);
|
|
356
|
+
if (progress.status === "permission_blocked" || progress.status === "stuck") {
|
|
357
|
+
await this.handleStuckAgent(progress);
|
|
358
|
+
}
|
|
359
|
+
else if (progress.status === "active") {
|
|
360
|
+
// Agent is working - reset stuck tracking
|
|
361
|
+
if (this.stuckSince) {
|
|
362
|
+
console.log(`[HEALTH] Agent resumed activity`);
|
|
363
|
+
this.stuckSince = null;
|
|
364
|
+
this.nudgeSentAt = null;
|
|
365
|
+
updateAgentInState(this.agentName, { stuckSince: undefined, status: "running" });
|
|
366
|
+
}
|
|
367
|
+
this.lastStableTime = new Date();
|
|
368
|
+
}
|
|
369
|
+
}, 60000); // Check every 60 seconds
|
|
370
|
+
}
|
|
371
|
+
/**
|
|
372
|
+
* Handles session death - logs crash and attempts auto-restart
|
|
373
|
+
*/
|
|
374
|
+
async handleSessionDeath(reason, logDir) {
|
|
375
|
+
const timestamp = new Date().toISOString();
|
|
376
|
+
const logFile = path.join(logDir, `crash-${this.agentName}.log`);
|
|
377
|
+
// Capture last session output before it's gone
|
|
378
|
+
let lastOutput = "";
|
|
379
|
+
try {
|
|
380
|
+
lastOutput = captureSessionOutput(this.agentName, 200) || "Unable to capture output";
|
|
381
|
+
}
|
|
382
|
+
catch {
|
|
383
|
+
lastOutput = "Failed to capture session output";
|
|
384
|
+
}
|
|
385
|
+
const crashLog = `
|
|
386
|
+
================================================================================
|
|
387
|
+
AGENT CRASH DETECTED
|
|
388
|
+
================================================================================
|
|
389
|
+
Timestamp: ${timestamp}
|
|
390
|
+
Agent: ${this.agentName}
|
|
391
|
+
Agent ID: ${this.agentId}
|
|
392
|
+
Reason: ${reason}
|
|
393
|
+
Restart Count: ${this.restartCount}/${MAX_RESTART_ATTEMPTS}
|
|
394
|
+
Sandbox: ${this.sandboxMode ? this.sandbox?.getContainerName() : "none"}
|
|
395
|
+
Workdir: ${this.agentConfig.workdir}
|
|
396
|
+
Model: ${this.runnerConfig.model}
|
|
397
|
+
|
|
398
|
+
--- Last Session Output ---
|
|
399
|
+
${lastOutput}
|
|
400
|
+
================================================================================
|
|
401
|
+
|
|
402
|
+
`;
|
|
403
|
+
fs.appendFileSync(logFile, crashLog);
|
|
404
|
+
// Check if we can restart
|
|
405
|
+
if (this.restartCount < MAX_RESTART_ATTEMPTS) {
|
|
406
|
+
this.restartCount++;
|
|
407
|
+
console.error(`[CRASH] Session died: ${reason}. Attempting restart (${this.restartCount}/${MAX_RESTART_ATTEMPTS})...`);
|
|
408
|
+
updateAgentInState(this.agentName, {
|
|
409
|
+
restartCount: this.restartCount,
|
|
410
|
+
lastRestartAt: timestamp,
|
|
411
|
+
status: "running",
|
|
412
|
+
});
|
|
413
|
+
try {
|
|
414
|
+
await this.restartSession();
|
|
415
|
+
console.log(`[RESTART] Agent restarted successfully`);
|
|
416
|
+
this.lastStableTime = new Date();
|
|
417
|
+
}
|
|
418
|
+
catch (error) {
|
|
419
|
+
console.error(`[RESTART] Failed to restart: ${error.message}`);
|
|
420
|
+
}
|
|
421
|
+
}
|
|
422
|
+
else {
|
|
423
|
+
// Exceeded restart limit - mark as failed
|
|
424
|
+
console.error(`[FAILED] Agent exceeded restart limit (${MAX_RESTART_ATTEMPTS}). Manual intervention required.`);
|
|
425
|
+
// Terminal bell to alert user
|
|
426
|
+
process.stdout.write("\x07");
|
|
427
|
+
updateAgentInState(this.agentName, {
|
|
428
|
+
status: "failed",
|
|
429
|
+
restartCount: this.restartCount,
|
|
430
|
+
});
|
|
431
|
+
// Stop monitoring
|
|
432
|
+
this.isRunning = false;
|
|
433
|
+
if (this.healthCheckInterval) {
|
|
434
|
+
clearInterval(this.healthCheckInterval);
|
|
435
|
+
this.healthCheckInterval = null;
|
|
436
|
+
}
|
|
437
|
+
}
|
|
438
|
+
}
|
|
439
|
+
/**
|
|
440
|
+
* Handles stuck agent - sends nudge first, then restarts if still stuck
|
|
441
|
+
*/
|
|
442
|
+
async handleStuckAgent(progress) {
|
|
443
|
+
const now = new Date();
|
|
444
|
+
if (!this.stuckSince) {
|
|
445
|
+
// First detection of stuck state
|
|
446
|
+
this.stuckSince = now;
|
|
447
|
+
console.log(`[HEALTH] Agent appears stuck: ${progress.details || progress.blockedOn || "no activity"}`);
|
|
448
|
+
updateAgentInState(this.agentName, {
|
|
449
|
+
stuckSince: now.toISOString(),
|
|
450
|
+
status: "stuck",
|
|
451
|
+
});
|
|
452
|
+
}
|
|
453
|
+
// If we haven't sent a nudge yet, send one
|
|
454
|
+
if (!this.nudgeSentAt) {
|
|
455
|
+
console.log(`[HEALTH] Sending nudge to unstick agent...`);
|
|
456
|
+
const nudgeMessage = progress.status === "permission_blocked"
|
|
457
|
+
? "Please continue with your task. If you see a permission prompt, try an alternative approach that doesn't require that permission."
|
|
458
|
+
: "Please continue with your current task.";
|
|
459
|
+
const sent = sendNudge(this.agentName, nudgeMessage);
|
|
460
|
+
if (sent) {
|
|
461
|
+
this.nudgeSentAt = now;
|
|
462
|
+
console.log(`[HEALTH] Nudge sent successfully`);
|
|
463
|
+
}
|
|
464
|
+
else {
|
|
465
|
+
console.log(`[HEALTH] Failed to send nudge`);
|
|
466
|
+
}
|
|
467
|
+
return;
|
|
468
|
+
}
|
|
469
|
+
// Check if enough time has passed since nudge
|
|
470
|
+
const timeSinceNudge = now.getTime() - this.nudgeSentAt.getTime();
|
|
471
|
+
if (timeSinceNudge < NUDGE_WAIT_MS) {
|
|
472
|
+
// Still waiting for agent to respond to nudge
|
|
473
|
+
return;
|
|
474
|
+
}
|
|
475
|
+
// Agent still stuck after nudge - trigger restart
|
|
476
|
+
console.log(`[HEALTH] Agent still stuck after nudge, triggering restart...`);
|
|
477
|
+
this.stuckSince = null;
|
|
478
|
+
this.nudgeSentAt = null;
|
|
479
|
+
await this.handleSessionDeath("stuck_after_nudge", path.join(os.homedir(), ".agentmesh", "logs"));
|
|
480
|
+
}
|
|
481
|
+
/**
|
|
482
|
+
* Restarts the agent session (sandbox or non-sandbox)
|
|
483
|
+
*/
|
|
484
|
+
async restartSession() {
|
|
485
|
+
// Destroy existing session
|
|
486
|
+
destroySession(this.agentName);
|
|
487
|
+
if (this.sandboxMode && this.sandbox) {
|
|
488
|
+
// Restart sandbox container
|
|
489
|
+
const newContainerId = await this.sandbox.restart();
|
|
490
|
+
console.log(`[RESTART] New container: ${newContainerId.substring(0, 12)}`);
|
|
491
|
+
// Recreate tmux session for sandbox
|
|
492
|
+
const containerName = this.sandbox.getContainerName();
|
|
493
|
+
const sessionName = getSessionName(this.agentName);
|
|
494
|
+
// Build environment args for docker exec
|
|
495
|
+
const envArgs = [];
|
|
496
|
+
const allEnv = {
|
|
497
|
+
...this.runnerConfig.env,
|
|
498
|
+
AGENT_TOKEN: this.token,
|
|
499
|
+
AGENTMESH_AGENT_ID: this.agentId,
|
|
500
|
+
};
|
|
501
|
+
for (const [key, value] of Object.entries(allEnv)) {
|
|
502
|
+
if (value !== undefined && value !== "") {
|
|
503
|
+
envArgs.push(`-e "${key}=${value}"`);
|
|
504
|
+
}
|
|
505
|
+
}
|
|
506
|
+
const envString = envArgs.join(" ");
|
|
507
|
+
const modelArg = this.runnerConfig.env?.OPENCODE_MODEL
|
|
508
|
+
? ` --model ${this.runnerConfig.env.OPENCODE_MODEL}`
|
|
509
|
+
: "";
|
|
510
|
+
const dockerExecCommand = `docker exec -it ${envString} ${containerName} opencode${modelArg}`;
|
|
511
|
+
const created = createSession(this.agentName, dockerExecCommand, undefined, undefined);
|
|
512
|
+
if (!created) {
|
|
513
|
+
throw new Error("Failed to create tmux session for restarted sandbox");
|
|
514
|
+
}
|
|
515
|
+
// Update state with new container name
|
|
516
|
+
updateAgentInState(this.agentName, {
|
|
517
|
+
sandboxContainer: containerName,
|
|
518
|
+
});
|
|
519
|
+
}
|
|
520
|
+
else {
|
|
521
|
+
// Non-sandbox restart - just recreate tmux session
|
|
522
|
+
const created = createSession(this.agentName, this.agentConfig.command, this.agentConfig.workdir, this.runnerConfig.env);
|
|
523
|
+
if (!created) {
|
|
524
|
+
throw new Error("Failed to create tmux session");
|
|
525
|
+
}
|
|
526
|
+
// Re-inject environment
|
|
527
|
+
updateSessionEnvironment(this.agentName, {
|
|
528
|
+
AGENT_TOKEN: this.token,
|
|
529
|
+
AGENTMESH_AGENT_ID: this.agentId,
|
|
530
|
+
...this.runnerConfig.env,
|
|
531
|
+
});
|
|
532
|
+
}
|
|
533
|
+
// Wait for session to be ready
|
|
534
|
+
await new Promise((resolve) => setTimeout(resolve, 2000));
|
|
535
|
+
}
|
|
252
536
|
async stop() {
|
|
253
537
|
console.log(`\nStopping agent: ${this.agentName}`);
|
|
254
538
|
this.isRunning = false;
|
|
539
|
+
// Stop health monitor
|
|
540
|
+
if (this.healthCheckInterval) {
|
|
541
|
+
clearInterval(this.healthCheckInterval);
|
|
542
|
+
this.healthCheckInterval = null;
|
|
543
|
+
}
|
|
255
544
|
// Save context before stopping
|
|
256
545
|
if (this.agentId) {
|
|
257
546
|
console.log("Saving agent context...");
|
|
@@ -267,8 +556,17 @@ Nudge agent:
|
|
|
267
556
|
this.ws.disconnect();
|
|
268
557
|
this.ws = null;
|
|
269
558
|
}
|
|
270
|
-
// Stop serve process or destroy tmux session
|
|
271
|
-
if (this.
|
|
559
|
+
// Stop sandbox, serve process, or destroy tmux session
|
|
560
|
+
if (this.sandboxMode && this.sandbox) {
|
|
561
|
+
console.log("Stopping sandbox...");
|
|
562
|
+
// In sandbox mode, we have both a tmux session (on host) and a Docker container
|
|
563
|
+
// Destroy tmux session first (this stops docker exec)
|
|
564
|
+
destroySession(this.agentName);
|
|
565
|
+
// Then destroy the container
|
|
566
|
+
await this.sandbox.destroy();
|
|
567
|
+
this.sandbox = null;
|
|
568
|
+
}
|
|
569
|
+
else if (this.serveMode && this.serveProcess) {
|
|
272
570
|
console.log("Stopping opencode serve...");
|
|
273
571
|
this.serveProcess.kill("SIGTERM");
|
|
274
572
|
this.serveProcess = null;
|
|
@@ -325,6 +623,120 @@ Nudge agent:
|
|
|
325
623
|
await new Promise((resolve) => setTimeout(resolve, 2000));
|
|
326
624
|
console.log(`opencode serve started on http://0.0.0.0:${this.servePort}`);
|
|
327
625
|
}
|
|
626
|
+
/**
|
|
627
|
+
* Starts agent in Docker sandbox mode
|
|
628
|
+
* Provides filesystem isolation with only workspace mounted
|
|
629
|
+
*
|
|
630
|
+
* Strategy: Start Docker container with tail -f /dev/null, then create
|
|
631
|
+
* a tmux session on the HOST that runs `docker exec -it <container> opencode`.
|
|
632
|
+
* This way tmux provides the TTY that docker exec needs.
|
|
633
|
+
*/
|
|
634
|
+
async startSandboxMode() {
|
|
635
|
+
console.log("Starting in Docker sandbox mode...");
|
|
636
|
+
// Check Docker availability
|
|
637
|
+
if (!DockerSandbox.checkDockerAvailable()) {
|
|
638
|
+
throw new Error("Docker is not available. Install Docker or use --sandbox host to run on host.");
|
|
639
|
+
}
|
|
640
|
+
const workdir = this.agentConfig.workdir || process.cwd();
|
|
641
|
+
// Check for existing sandbox container
|
|
642
|
+
const existingContainer = DockerSandbox.findExisting(this.agentName);
|
|
643
|
+
if (existingContainer) {
|
|
644
|
+
console.log(`Found existing sandbox container: ${existingContainer}`);
|
|
645
|
+
console.log(`Stop it with: agentmesh stop ${this.agentName}`);
|
|
646
|
+
throw new Error("Sandbox container already exists");
|
|
647
|
+
}
|
|
648
|
+
// Build additional mounts for credentials and config
|
|
649
|
+
// The entrypoint script copies these from /tmp/ to the correct locations
|
|
650
|
+
const additionalMounts = [];
|
|
651
|
+
// Mount git credentials
|
|
652
|
+
const gitCredentialsPath = path.join(os.homedir(), ".git-credentials");
|
|
653
|
+
if (fs.existsSync(gitCredentialsPath)) {
|
|
654
|
+
additionalMounts.push(`${gitCredentialsPath}:/tmp/.git-credentials-host:ro`);
|
|
655
|
+
}
|
|
656
|
+
// Mount OpenCode auth.json for API provider tokens (Anthropic, OpenAI, etc.)
|
|
657
|
+
const opencodeAuthPath = path.join(os.homedir(), ".local", "share", "opencode", "auth.json");
|
|
658
|
+
if (fs.existsSync(opencodeAuthPath)) {
|
|
659
|
+
additionalMounts.push(`${opencodeAuthPath}:/tmp/.opencode-auth-host:ro`);
|
|
660
|
+
}
|
|
661
|
+
// Mount AgentMesh config for hub URL, API key, workspace
|
|
662
|
+
const agentmeshConfigPath = path.join(os.homedir(), ".agentmesh", "config.json");
|
|
663
|
+
if (fs.existsSync(agentmeshConfigPath)) {
|
|
664
|
+
additionalMounts.push(`${agentmeshConfigPath}:/tmp/.agentmesh-config-host:ro`);
|
|
665
|
+
}
|
|
666
|
+
// Create and mount permissive OpenCode config for sandbox
|
|
667
|
+
// This allows all permissions since the container is already sandboxed
|
|
668
|
+
this.ensureSandboxOpencodeConfig();
|
|
669
|
+
additionalMounts.push(`${SANDBOX_OPENCODE_CONFIG_PATH}:/workspace/opencode.json:ro`);
|
|
670
|
+
// Pass GitHub token as environment variable for git operations
|
|
671
|
+
const gitCredentials = fs.existsSync(gitCredentialsPath)
|
|
672
|
+
? fs.readFileSync(gitCredentialsPath, "utf-8").trim()
|
|
673
|
+
: "";
|
|
674
|
+
const gitHubToken = gitCredentials.match(/github_pat_[^\s@]+/)?.[0] || "";
|
|
675
|
+
// Build the command to run inside the container
|
|
676
|
+
// The agentmesh CLI inside the container will create tmux + opencode
|
|
677
|
+
const model = this.runnerConfig.env?.OPENCODE_MODEL || this.runnerConfig.model || "claude-sonnet-4";
|
|
678
|
+
const containerCommand = [
|
|
679
|
+
"agentmesh",
|
|
680
|
+
"start",
|
|
681
|
+
"--name",
|
|
682
|
+
this.agentName,
|
|
683
|
+
"--model",
|
|
684
|
+
model,
|
|
685
|
+
"--foreground",
|
|
686
|
+
];
|
|
687
|
+
// Create sandbox configuration
|
|
688
|
+
this.sandbox = new DockerSandbox({
|
|
689
|
+
agentName: this.agentName,
|
|
690
|
+
image: this.sandboxImage,
|
|
691
|
+
workspacePath: workdir,
|
|
692
|
+
cpuLimit: this.sandboxCpu,
|
|
693
|
+
memoryLimit: this.sandboxMemory,
|
|
694
|
+
env: {
|
|
695
|
+
...this.runnerConfig.env,
|
|
696
|
+
AGENT_TOKEN: this.token,
|
|
697
|
+
AGENTMESH_AGENT_ID: this.agentId,
|
|
698
|
+
// Git credentials for pushing to GitHub
|
|
699
|
+
...(gitHubToken && { GH_TOKEN: gitHubToken, GITHUB_TOKEN: gitHubToken }),
|
|
700
|
+
},
|
|
701
|
+
serveMode: this.serveMode,
|
|
702
|
+
servePort: this.servePort,
|
|
703
|
+
networkMode: "bridge",
|
|
704
|
+
additionalMounts: additionalMounts.length > 0 ? additionalMounts : undefined,
|
|
705
|
+
command: this.serveMode ? undefined : containerCommand,
|
|
706
|
+
});
|
|
707
|
+
// Validate mount policy (will throw if denied)
|
|
708
|
+
this.sandbox.validateMountPolicy();
|
|
709
|
+
// Pull image if needed
|
|
710
|
+
await this.sandbox.pullImage();
|
|
711
|
+
// Start container with agentmesh running inside
|
|
712
|
+
// The entrypoint script sets up credentials before agentmesh starts
|
|
713
|
+
await this.sandbox.start();
|
|
714
|
+
const containerName = this.sandbox.getContainerName();
|
|
715
|
+
console.log(`
|
|
716
|
+
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
717
|
+
SANDBOX MODE ACTIVE
|
|
718
|
+
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
719
|
+
|
|
720
|
+
Container: ${containerName}
|
|
721
|
+
Image: ${this.sandboxImage}
|
|
722
|
+
Workspace: ${workdir} -> /workspace
|
|
723
|
+
CPU: ${this.sandboxCpu} core(s)
|
|
724
|
+
Memory: ${this.sandboxMemory}
|
|
725
|
+
Model: ${model}
|
|
726
|
+
|
|
727
|
+
The agent daemon is running INSIDE the Docker container.
|
|
728
|
+
tmux session and OpenCode are managed inside the container.
|
|
729
|
+
|
|
730
|
+
Attach: agentmesh attach ${this.agentName}
|
|
731
|
+
Nudge: agentmesh nudge ${this.agentName} "message"
|
|
732
|
+
Stop: agentmesh stop ${this.agentName}
|
|
733
|
+
Logs: docker logs ${containerName}
|
|
734
|
+
|
|
735
|
+
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
|
|
736
|
+
`);
|
|
737
|
+
// No host tmux session needed - the container runs agentmesh which creates its own tmux
|
|
738
|
+
// Heartbeats are sent by the daemon running inside the container
|
|
739
|
+
}
|
|
328
740
|
/**
|
|
329
741
|
* Saves the current agent context to disk
|
|
330
742
|
*/
|
|
@@ -484,5 +896,27 @@ Option 3: Use --auto-setup to automatically clone the repository:
|
|
|
484
896
|
console.log(`✓ Workspace ready: ${workspacePath}\n`);
|
|
485
897
|
return workspacePath;
|
|
486
898
|
}
|
|
899
|
+
/**
|
|
900
|
+
* Ensures the sandbox OpenCode config exists
|
|
901
|
+
* Creates ~/.agentmesh/opencode-sandbox.json with permissive permissions and model
|
|
902
|
+
*/
|
|
903
|
+
ensureSandboxOpencodeConfig() {
|
|
904
|
+
const configDir = path.dirname(SANDBOX_OPENCODE_CONFIG_PATH);
|
|
905
|
+
if (!fs.existsSync(configDir)) {
|
|
906
|
+
fs.mkdirSync(configDir, { recursive: true });
|
|
907
|
+
}
|
|
908
|
+
// Build config with model if available
|
|
909
|
+
const config = {
|
|
910
|
+
...SANDBOX_OPENCODE_CONFIG,
|
|
911
|
+
};
|
|
912
|
+
// Include model from runner config
|
|
913
|
+
const model = this.runnerConfig.env?.OPENCODE_MODEL;
|
|
914
|
+
if (model) {
|
|
915
|
+
config.model = model;
|
|
916
|
+
}
|
|
917
|
+
// Always write to ensure model is up to date
|
|
918
|
+
fs.writeFileSync(SANDBOX_OPENCODE_CONFIG_PATH, JSON.stringify(config, null, 2));
|
|
919
|
+
console.log(`Updated sandbox OpenCode config: ${SANDBOX_OPENCODE_CONFIG_PATH}`);
|
|
920
|
+
}
|
|
487
921
|
}
|
|
488
922
|
//# sourceMappingURL=daemon.js.map
|