@sagentlab/navarch-runtime 0.1.92 → 0.1.95

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 CHANGED
@@ -9,6 +9,16 @@ environment, outside the coding agent's OS sandbox, with a fresh browser
9
9
  context for desktop and touch-enabled mobile captures. It neither uses the
10
10
  operator's browser profile nor executes agent-provided scripts.
11
11
 
12
+ The same server offers interactive `local_browser_open`, `local_browser_click`,
13
+ `local_browser_fill`, `local_browser_press`, `local_browser_snapshot`, and
14
+ `local_browser_screenshot` tools for walking through a service the agent starts
15
+ on this machine. That browser is limited to loopback (`localhost`,
16
+ `*.localhost`, `127.0.0.0/8`, `::1`). Every request, WebSocket, or navigation
17
+ to any other host is aborted and listed in the tool result. The agent's own
18
+ sandbox can already reach these services, so the tools add a browser without
19
+ adding network reach. Each lease gets one tab at a time, a fresh context per
20
+ viewport, and at most 300 actions.
21
+
12
22
  Install the runtime's pinned browser once on the agent host (and after a
13
23
  Playwright version update):
14
24
 
@@ -36,9 +46,13 @@ lifetime, so restart the agent after fixing the browser to re-probe.
36
46
  `navarch-runtime doctor` runs the same check.
37
47
 
38
48
  Codex's macOS seatbelt denies Chromium's Mach port registration
39
- (`bootstrap_check_in ... MachPortRendezvousServer: Permission denied`), so a
40
- sandboxed Codex machine on macOS does not take browser tasks. Run those with
41
- another agent such as Claude Code, or use Docker mode. Docker mode is not
49
+ (`bootstrap_check_in ... MachPortRendezvousServer: Permission denied`), and
50
+ `--single-process` crashes once a second page opens. When Node runs in the
51
+ sandbox and only Chromium fails, the probe still passes, with that reason. Codex
52
+ sessions then browse local services through the `local_browser_*` tools above,
53
+ and the session prompt tells them to. With an operator `NAVARCH_MCP_CONFIG_PATH`
54
+ those tools are absent, so the machine does not take browser tasks; run those
55
+ with another agent such as Claude Code, or use Docker mode. Docker mode is not
42
56
  probed because its image supplies the browser. Set `NAVARCH_BROWSER_PROBE=off`
43
57
  to trust the configured capability list without a check.
44
58
 
@@ -316,13 +330,14 @@ own budgets. An absolute limit covers all guidance/remediation turns within the
316
330
  claim, rather than resetting with each turn. These are runtime limits, not
317
331
  provider spending caps.
318
332
 
319
- After a watchdog stop or lease loss during a turn, before worktree cleanup, the runtime attempts a private
333
+ After a watchdog stop, lease loss during a turn, or exhausted delivery correction
334
+ turns, before worktree cleanup, the runtime attempts a private
320
335
  local recovery checkpoint at
321
336
  `<workspaceRoot>/checkpoints/<project>/<task>/<session>/`. It contains a Git
322
337
  bundle of committed history, a binary patch for dirty tracked files, untracked
323
338
  non-ignored files, and recovery instructions. The bundle excludes repository
324
339
  configuration and broker credentials. Checkpoints are not pushed or uploaded;
325
- timeout reports and runtime logs name their location. They survive session cleanup and need
340
+ failure reports and runtime logs name their location. They survive session cleanup and need
326
341
  manual recovery and removal on that machine. Ignored files, agent conversation
327
342
  state, and checkpoints across machine/disk loss are not provided.
328
343
 
@@ -490,7 +505,9 @@ as an explicit read-only mount.
490
505
  - **Codex:** the runtime passes a one-off native permission profile with
491
506
  `approval_policy="on-request"` and `approvals_reviewer="auto_review"`.
492
507
  Codex's OS sandbox grants read/write access only to the allowed roots and
493
- denies the surrounding multi-session workspace. Its network policy allows
508
+ denies the surrounding multi-session workspace. After preparing the worktree,
509
+ the runtime explicitly grants its resolved linked Git metadata directory write
510
+ access so staging and committing work inside the sandbox. Its network policy allows
494
511
  GitHub and GitHub-hosted content so authenticated `gh` and git operations
495
512
  required by the task stay inside the sandbox. Other destinations remain
496
513
  blocked and eligible escalations are decided by the automatic reviewer
@@ -13,6 +13,7 @@ const mcp_js_1 = require("@modelcontextprotocol/sdk/server/mcp.js");
13
13
  const streamableHttp_js_1 = require("@modelcontextprotocol/sdk/server/streamableHttp.js");
14
14
  const playwright_1 = require("playwright");
15
15
  const zod_1 = require("zod");
16
+ const local_browser_cjs_1 = require("./local-browser.cjs");
16
17
  const node_child_process_1 = require("node:child_process");
17
18
  const node_util_1 = require("node:util");
18
19
  const node_path_1 = __importDefault(require("node:path"));
@@ -46,7 +47,8 @@ const captureShape = {
46
47
  exports.captureSchema = zod_1.z.object(captureShape);
47
48
  /** Runs in the supervisor, outside the coding agent's OS sandbox. Fresh
48
49
  * contexts have no user cookies, arbitrary script execution, or file access.
49
- * Only the narrow capture operation is exposed to the lease. */
50
+ * The lease gets the narrow capture operation plus the loopback-only
51
+ * local_browser_* tools (local-browser.cts). */
50
52
  async function startBrowserEvidence(options) {
51
53
  const token = (0, node_crypto_1.randomBytes)(32).toString("hex");
52
54
  const fetchImpl = options.fetchImpl ?? fetch;
@@ -57,6 +59,10 @@ async function startBrowserEvidence(options) {
57
59
  let browser;
58
60
  let activeCapture;
59
61
  let attempts = 0;
62
+ const localBrowser = (0, local_browser_cjs_1.createLocalBrowser)({
63
+ launch: options.launch ?? launchCaptureBrowser,
64
+ isActive: () => !closed && options.isActive(),
65
+ });
60
66
  const assertActive = () => {
61
67
  if (closed || !options.isActive())
62
68
  throw new Error("Browser evidence lease is no longer active.");
@@ -204,6 +210,7 @@ async function startBrowserEvidence(options) {
204
210
  return { isError: true, content: [{ type: "text", text: String(error) }] };
205
211
  }
206
212
  });
213
+ (0, local_browser_cjs_1.registerLocalBrowserTools)(server, localBrowser);
207
214
  res.on("close", () => { void transport.close(); void server.close(); });
208
215
  try {
209
216
  await server.connect(transport);
@@ -253,6 +260,7 @@ async function startBrowserEvidence(options) {
253
260
  const stopped = new Promise((resolve) => http.close(() => resolve()));
254
261
  http.closeAllConnections();
255
262
  await browser?.close().catch(() => undefined);
263
+ await localBrowser.close();
256
264
  await activeCapture?.catch(() => undefined);
257
265
  await stopped;
258
266
  },
@@ -74,8 +74,19 @@ async function probeBrowserCapability(config, deps = {}) {
74
74
  const platform = deps.platform ?? process.platform;
75
75
  if (codexSandboxNeedsProbe(config, platform)) {
76
76
  const sandboxFailure = await probeCodexSandbox(config, deps);
77
+ if (sandboxFailure?.browserOnly && !config.mcpConfigPath) {
78
+ // Node runs in the sandbox, so the agent can start its services there;
79
+ // only Chromium cannot. Sessions get the supervisor's local_browser_*
80
+ // tools (local-browser.cts) instead, which reach the same loopback.
81
+ return {
82
+ status: "passed",
83
+ reason: truncate(`Headless Chromium cannot start inside the Codex sandbox (${sandboxFailure.detail}); ` +
84
+ "Codex sessions browse local services through Navarch's local_browser tools, which run outside it."),
85
+ checked_at: now().toISOString(),
86
+ };
87
+ }
77
88
  if (sandboxFailure) {
78
- return { status: "failed", reason: truncate(sandboxFailure), checked_at: now().toISOString() };
89
+ return { status: "failed", reason: truncate(sandboxFailure.reason), checked_at: now().toISOString() };
79
90
  }
80
91
  }
81
92
  return { status: "passed", checked_at: now().toISOString() };
@@ -215,8 +226,9 @@ async function probeCodexSandbox(config, deps) {
215
226
  // sandbox view: a path outside the granted read roots, not an old CLI.
216
227
  const hidden = /bwrap: execvp .*No such file or directory/.test(output);
217
228
  const opensslConfig = /OpenSSL configuration error/.test(output);
218
- return ("Could not verify a browser inside the Codex sandbox: `codex sandbox` did not run the probe " +
219
- `(${result.timedOut ? "timed out" : output.trim() ? summarizeLaunchFailure(output) : `exit ${result.code}`}). ` +
229
+ const detail = result.timedOut ? "timed out" : output.trim() ? summarizeLaunchFailure(output) : `exit ${result.code}`;
230
+ const reason = ("Could not verify a browser inside the Codex sandbox: `codex sandbox` did not run the probe " +
231
+ `(${detail}). ` +
220
232
  (hidden
221
233
  ? "The sandbox cannot see that executable; add its install directory to NAVARCH_GUARD_EXTRA_ROOTS, " +
222
234
  "or run browser tasks with another agent."
@@ -224,19 +236,22 @@ async function probeCodexSandbox(config, deps) {
224
236
  ? "Node cannot read its OpenSSL config inside the sandbox, so Codex sessions cannot run node either; " +
225
237
  "update the Navarch runtime, or add that file's directory to NAVARCH_GUARD_EXTRA_ROOTS."
226
238
  : "Update the Codex CLI, or run browser tasks with another agent."));
239
+ return { reason, detail, browserOnly: false };
227
240
  }
228
241
  const detail = result.timedOut ? "the launch timed out." : summarizeLaunchFailure(output);
229
242
  const machPort = /bootstrap_check_in|MachPortRendezvous|mach_port_rendezvous/i.test(output);
230
- return (`Headless Chromium cannot start inside the Codex sandbox: ${detail} ` +
243
+ const reason = (`Headless Chromium cannot start inside the Codex sandbox: ${detail} ` +
231
244
  (machPort
232
245
  ? "Codex's macOS sandbox denies Chromium's Mach port registration. "
233
246
  : "") +
234
247
  "Run browser tasks with another agent (for example Claude Code), use NAVARCH_SANDBOX_MODE=docker, " +
235
248
  "or run Codex without its sandbox.");
249
+ return { reason, detail, browserOnly: true };
236
250
  }
237
251
  catch (error) {
238
252
  log.warn(`codex sandbox browser probe could not run: ${String(error)}`);
239
- return `Could not verify a browser inside the Codex sandbox: ${summarizeLaunchFailure(String(error))}`;
253
+ const detail = summarizeLaunchFailure(String(error));
254
+ return { reason: `Could not verify a browser inside the Codex sandbox: ${detail}`, detail, browserOnly: false };
240
255
  }
241
256
  finally {
242
257
  await node_fs_1.promises.rm(probeRoot, { recursive: true, force: true }).catch(() => undefined);
package/dist/cli.cjs CHANGED
@@ -248,7 +248,7 @@ async function startCommand(flags) {
248
248
  log.warn(`not advertising browser-use: ${browserProbe.reason}`);
249
249
  }
250
250
  else if (browserProbe.status === "passed") {
251
- log.info("browser probe passed; advertising browser-use");
251
+ log.info(`browser probe passed; advertising browser-use${browserProbe.reason ? ` (${browserProbe.reason})` : ""}`);
252
252
  }
253
253
  const config = {
254
254
  ...configuredConfig,
@@ -359,7 +359,10 @@ function mapExitCondition(result) {
359
359
  // including when that response says the task could not start. Treat an
360
360
  // explicit leading blocked verdict as a failed lease so the dispatcher can
361
361
  // retry/escalate it instead of falsely moving unfinished work to Done.
362
- if (/^blocked(?:\s|:|$)/i.test(reportText.trim())) {
362
+ // Build agents also use an explicit outcome marker. A zero process exit
363
+ // means the conversation finished, not that the requested build succeeded.
364
+ if (/^blocked(?:\s|:|$)/i.test(reportText.trim()) ||
365
+ /^Build outcome:\s*(?:FAILED|BLOCKED)\b/i.test(reportText.trim())) {
363
366
  return {
364
367
  leaseOutcome: "failed",
365
368
  exitStatus: "failed",
@@ -0,0 +1,279 @@
1
+ "use strict";
2
+ Object.defineProperty(exports, "__esModule", { value: true });
3
+ exports.localBrowserShapes = void 0;
4
+ exports.isLoopbackUrl = isLoopbackUrl;
5
+ exports.createLocalBrowser = createLocalBrowser;
6
+ exports.registerLocalBrowserTools = registerLocalBrowserTools;
7
+ const node_net_1 = require("node:net");
8
+ const playwright_1 = require("playwright");
9
+ const zod_1 = require("zod");
10
+ /**
11
+ * Interactive browser for services the agent starts on this machine.
12
+ *
13
+ * Codex's macOS seatbelt denies Chromium's Mach port registration
14
+ * (`bootstrap_check_in ... MachPortRendezvousServer: Permission denied`), so
15
+ * no browser can start inside a Codex session there, and `--single-process`
16
+ * crashes as soon as a second page opens. This Chromium runs in the
17
+ * supervisor instead, next to capture_visual_evidence, and is limited to
18
+ * loopback: the agent's own sandbox can already reach localhost, so these
19
+ * tools add a browser without adding network reach. Every request, WebSocket,
20
+ * and navigation to any other host is aborted and reported back.
21
+ */
22
+ const MAX_ACTIONS = 300;
23
+ const ACTION_TIMEOUT_MS = 10_000;
24
+ const NAVIGATION_TIMEOUT_MS = 30_000;
25
+ const MAX_SNAPSHOT_CHARS = 12_000;
26
+ const MAX_LOG_LINES = 15;
27
+ /** http(s)/ws(s) on localhost, *.localhost, 127.0.0.0/8, or ::1, without embedded credentials. */
28
+ function isLoopbackUrl(value) {
29
+ let url;
30
+ try {
31
+ url = new URL(value);
32
+ }
33
+ catch {
34
+ return false;
35
+ }
36
+ if (!["http:", "https:", "ws:", "wss:"].includes(url.protocol) || url.username || url.password)
37
+ return false;
38
+ const host = url.hostname.toLowerCase().replace(/^\[|\]$/g, "");
39
+ if (host === "localhost" || host.endsWith(".localhost"))
40
+ return true;
41
+ if ((0, node_net_1.isIP)(host) === 4)
42
+ return host.split(".")[0] === "127";
43
+ return host === "::1";
44
+ }
45
+ const loopbackUrl = zod_1.z.string().refine(isLoopbackUrl, "Only services on this machine are reachable: use http://localhost:<port> or http://127.0.0.1:<port>.");
46
+ const selector = zod_1.z.string().min(1).max(500).describe("Playwright selector, e.g. `text=Sign in`, `role=button[name=\"Save\"]`, or CSS.");
47
+ exports.localBrowserShapes = {
48
+ open: {
49
+ url: loopbackUrl,
50
+ viewport: zod_1.z.enum(["desktop", "mobile"]).default("desktop"),
51
+ },
52
+ click: { selector },
53
+ fill: { selector, value: zod_1.z.string().max(10_000) },
54
+ press: {
55
+ key: zod_1.z.string().min(1).max(50).describe("Key or chord, e.g. `Enter`, `Tab`, `Control+A`."),
56
+ selector: selector.optional(),
57
+ },
58
+ snapshot: {},
59
+ screenshot: { full_page: zod_1.z.boolean().default(false) },
60
+ };
61
+ function createLocalBrowser(options) {
62
+ let browser;
63
+ let context;
64
+ let page;
65
+ let viewport;
66
+ let actions = 0;
67
+ let closed = false;
68
+ let creatingPage = false;
69
+ let queue = Promise.resolve();
70
+ let notes = [];
71
+ const blocked = new Set();
72
+ const logs = [];
73
+ const log = (line) => {
74
+ logs.push(line.slice(0, 500));
75
+ if (logs.length > MAX_LOG_LINES)
76
+ logs.shift();
77
+ };
78
+ const block = (url) => {
79
+ try {
80
+ blocked.add(new URL(url).origin);
81
+ }
82
+ catch {
83
+ blocked.add(url.slice(0, 100));
84
+ }
85
+ };
86
+ function track(next) {
87
+ page = next;
88
+ next.on("console", (message) => {
89
+ if (["error", "warning"].includes(message.type()))
90
+ log(`console.${message.type()}: ${message.text()}`);
91
+ });
92
+ next.on("pageerror", (error) => log(`pageerror: ${error.message}`));
93
+ }
94
+ async function ensureContext(wanted) {
95
+ if (context && viewport === wanted)
96
+ return context;
97
+ await context?.close().catch(() => undefined);
98
+ context = undefined;
99
+ page = undefined;
100
+ try {
101
+ browser ??= await options.launch();
102
+ }
103
+ catch {
104
+ throw new Error("The local browser could not start. On the agent host run `navarch-runtime browser-install`.");
105
+ }
106
+ const created = await browser.newContext({
107
+ ...(wanted === "mobile" ? playwright_1.devices["iPhone 13"] : { viewport: { width: 1440, height: 900 } }),
108
+ acceptDownloads: false,
109
+ // Service workers would answer fetches before the route below sees them.
110
+ serviceWorkers: "block",
111
+ });
112
+ await created.route("**/*", (route) => {
113
+ const url = route.request().url();
114
+ if (isLoopbackUrl(url))
115
+ return route.continue();
116
+ block(url);
117
+ return route.abort("blockedbyclient");
118
+ });
119
+ await created.routeWebSocket(() => true, (socket) => {
120
+ if (isLoopbackUrl(socket.url()))
121
+ return socket.connectToServer();
122
+ block(socket.url());
123
+ return socket.close({ code: 1008, reason: "Only loopback services are reachable." });
124
+ });
125
+ created.on("page", (opened) => {
126
+ if (creatingPage || opened === page)
127
+ return;
128
+ track(opened);
129
+ notes.push("A new tab opened; it is now the active page.");
130
+ });
131
+ context = created;
132
+ viewport = wanted;
133
+ return created;
134
+ }
135
+ function currentPage() {
136
+ if (!page || page.isClosed())
137
+ throw new Error("No page is open. Call local_browser_open first.");
138
+ return page;
139
+ }
140
+ async function snapshot(heading) {
141
+ const active = currentPage();
142
+ let tree;
143
+ try {
144
+ tree = await active.locator("body").ariaSnapshot({ timeout: 5_000 });
145
+ }
146
+ catch (error) {
147
+ tree = `(accessibility snapshot unavailable: ${String(error).split("\n")[0]})`;
148
+ }
149
+ if (tree.length > MAX_SNAPSHOT_CHARS)
150
+ tree = `${tree.slice(0, MAX_SNAPSHOT_CHARS)}\n… (truncated)`;
151
+ const lines = [
152
+ ...(heading ? [heading] : []),
153
+ ...notes,
154
+ ...(active.url().startsWith("chrome-error://")
155
+ ? ["The last navigation failed or tried to leave this machine and was blocked. Use local_browser_open to continue."]
156
+ : []),
157
+ `URL: ${active.url()}`,
158
+ `Title: ${await active.title().catch(() => "")}`,
159
+ `Viewport: ${viewport}`,
160
+ ...(logs.length ? ["Recent console/page errors:", ...logs.map((line) => ` ${line}`)] : []),
161
+ ...(blocked.size
162
+ ? [`Blocked non-loopback requests (this browser reaches only this machine): ${[...blocked].slice(-10).join(", ")}`]
163
+ : []),
164
+ "",
165
+ tree,
166
+ ];
167
+ notes = [];
168
+ return [{ type: "text", text: lines.join("\n") }];
169
+ }
170
+ /** Runs tool calls one at a time; a lease ending stops further use. */
171
+ function run(action) {
172
+ const next = queue.then(async () => {
173
+ if (closed || !options.isActive())
174
+ throw new Error("The local browser lease is no longer active.");
175
+ if (++actions > MAX_ACTIONS)
176
+ throw new Error(`This lease reached its ${MAX_ACTIONS} local browser actions.`);
177
+ return action();
178
+ });
179
+ queue = next.catch(() => undefined);
180
+ return next;
181
+ }
182
+ async function settle(active) {
183
+ await active.waitForLoadState("load", { timeout: 5_000 }).catch(() => undefined);
184
+ }
185
+ return {
186
+ open(args) {
187
+ return run(async () => {
188
+ if (!isLoopbackUrl(args.url))
189
+ throw new Error("Only loopback URLs are reachable.");
190
+ const created = await ensureContext(args.viewport ?? "desktop");
191
+ if (!page || page.isClosed()) {
192
+ creatingPage = true;
193
+ try {
194
+ track(await created.newPage());
195
+ }
196
+ finally {
197
+ creatingPage = false;
198
+ }
199
+ }
200
+ const response = await currentPage().goto(args.url, { waitUntil: "load", timeout: NAVIGATION_TIMEOUT_MS });
201
+ return snapshot(`HTTP ${response?.status() ?? "(no response)"}`);
202
+ });
203
+ },
204
+ click(args) {
205
+ return run(async () => {
206
+ const active = currentPage();
207
+ await active.locator(args.selector).first().click({ timeout: ACTION_TIMEOUT_MS });
208
+ await settle(active);
209
+ return snapshot();
210
+ });
211
+ },
212
+ fill(args) {
213
+ return run(async () => {
214
+ await currentPage().locator(args.selector).first().fill(args.value, { timeout: ACTION_TIMEOUT_MS });
215
+ return snapshot();
216
+ });
217
+ },
218
+ press(args) {
219
+ return run(async () => {
220
+ const active = currentPage();
221
+ if (args.selector)
222
+ await active.locator(args.selector).first().press(args.key, { timeout: ACTION_TIMEOUT_MS });
223
+ else
224
+ await active.keyboard.press(args.key);
225
+ await settle(active);
226
+ return snapshot();
227
+ });
228
+ },
229
+ snapshot() {
230
+ return run(() => snapshot());
231
+ },
232
+ screenshot(args) {
233
+ return run(async () => {
234
+ const bytes = await currentPage().screenshot({ type: "png", fullPage: args.full_page ?? false, timeout: ACTION_TIMEOUT_MS });
235
+ return [
236
+ { type: "text", text: `Screenshot of ${currentPage().url()} (${viewport}). Not uploaded; use capture_visual_evidence for PR evidence.` },
237
+ { type: "image", data: bytes.toString("base64"), mimeType: "image/png" },
238
+ ];
239
+ });
240
+ },
241
+ async close() {
242
+ closed = true;
243
+ await queue.catch(() => undefined);
244
+ await context?.close().catch(() => undefined);
245
+ await browser?.close().catch(() => undefined);
246
+ browser = undefined;
247
+ context = undefined;
248
+ page = undefined;
249
+ },
250
+ };
251
+ }
252
+ const toolResult = (call) => call().then((content) => ({ content }), (error) => ({ isError: true, content: [{ type: "text", text: String(error) }] }));
253
+ /** Registers the local_browser_* tools on a lease's navarch-browser MCP server. */
254
+ function registerLocalBrowserTools(server, browser) {
255
+ server.registerTool("local_browser_open", {
256
+ description: "Open a page served on this machine (http://localhost:<port> or 127.0.0.1) in Navarch's browser, which runs outside the coding-agent sandbox. Start your dev server first. Reuses one tab; switching viewport starts a fresh session. Returns the page's accessibility snapshot, console errors, and any blocked non-loopback requests.",
257
+ inputSchema: exports.localBrowserShapes.open,
258
+ }, (args) => toolResult(() => browser.open(args)));
259
+ server.registerTool("local_browser_click", {
260
+ description: "Click the first element matching a Playwright selector on the open local page, then return the updated snapshot.",
261
+ inputSchema: exports.localBrowserShapes.click,
262
+ }, (args) => toolResult(() => browser.click(args)));
263
+ server.registerTool("local_browser_fill", {
264
+ description: "Replace the value of an input, textarea, or contenteditable on the open local page, then return the snapshot.",
265
+ inputSchema: exports.localBrowserShapes.fill,
266
+ }, (args) => toolResult(() => browser.fill(args)));
267
+ server.registerTool("local_browser_press", {
268
+ description: "Press a key (optionally focused on a selector) on the open local page, then return the snapshot.",
269
+ inputSchema: exports.localBrowserShapes.press,
270
+ }, (args) => toolResult(() => browser.press(args)));
271
+ server.registerTool("local_browser_snapshot", {
272
+ description: "Return the open local page's URL, title, accessibility snapshot, console errors, and blocked requests.",
273
+ inputSchema: exports.localBrowserShapes.snapshot,
274
+ }, () => toolResult(() => browser.snapshot()));
275
+ server.registerTool("local_browser_screenshot", {
276
+ description: "Return a PNG of the open local page for inspection. Not uploaded as evidence.",
277
+ inputSchema: exports.localBrowserShapes.screenshot,
278
+ }, (args) => toolResult(() => browser.screenshot(args)));
279
+ }
package/dist/session.cjs CHANGED
@@ -232,13 +232,14 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
232
232
  let retainWorktree = false;
233
233
  let heartbeatInFlight = null;
234
234
  let heartbeatWorkspaceUpdate = null;
235
+ let completingLease = false;
235
236
  const sessionOutcome = {};
236
- const checkpointForRecovery = async (result) => {
237
+ const checkpointForRecovery = async (result, reason) => {
237
238
  if (result.checkpointPath || result.retainedWorktreePath)
238
239
  return;
239
240
  try {
240
241
  result.checkpointPath = await gitWorktree.saveCheckpoint();
241
- log.warn(`session ${leaseId} stopped for ${leaseLost ? "lease loss" : result.capacityError ? "provider capacity" : result.timeoutReason ?? "duration"}; recovery checkpoint: ${result.checkpointPath}`);
242
+ log.warn(`session ${leaseId} stopped for ${reason ?? (leaseLost ? "lease loss" : result.capacityError ? "provider capacity" : result.timeoutReason ?? "duration")}; recovery checkpoint: ${result.checkpointPath}`);
242
243
  }
243
244
  catch (error) {
244
245
  retainWorktree = true;
@@ -285,7 +286,15 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
285
286
  log.warn(`lease heartbeat failed for ${leaseId}: ${String(err)} — retrying on the next heartbeat while the confirmed lease remains valid.`);
286
287
  return;
287
288
  }
288
- log.warn(`lease heartbeat failed for ${leaseId}: ${String(err)} — killing session.`);
289
+ if (completingLease && isTerminalLeaseHeartbeatError(err)) {
290
+ // Completion can release the lease before its HTTP response arrives.
291
+ // Keep awaiting/re-sending that report to resolve the ambiguous write;
292
+ // there is no agent turn to kill during this phase.
293
+ log.info(`lease ${leaseId} is no longer active while completion is pending; awaiting completion acknowledgement.`);
294
+ }
295
+ else {
296
+ log.warn(`lease heartbeat failed for ${leaseId}: ${String(err)} — killing session.`);
297
+ }
289
298
  leaseLost = true;
290
299
  leaseGone = isTerminalLeaseHeartbeatError(err);
291
300
  heartbeatGuard.stop();
@@ -385,14 +394,6 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
385
394
  await node_fs_1.promises.mkdir(tmpDir, { recursive: true, mode: 0o700 });
386
395
  Object.assign(sessionEnv, { TMPDIR: tmpDir, TMP: tmpDir, TEMP: tmpDir });
387
396
  }
388
- codexGuardArgs = (0, worktree_guard_cjs_1.codexWorktreeGuardArgs)({
389
- workDir,
390
- worktreePath: gitWorktree.worktreePath,
391
- repositoryPath: gitWorktree.repositoryPath,
392
- workspaceRoot: config.workspaceRoot,
393
- extraRoots: config.guardExtraRoots,
394
- codexBin: config.codexBin,
395
- });
396
397
  }
397
398
  else if (runtime === "gemini") {
398
399
  geminiGuardMounts = (0, worktree_guard_cjs_1.geminiSandboxMounts)({
@@ -459,12 +460,22 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
459
460
  const mcp = JSON.parse(await node_fs_1.promises.readFile(mcpConfigPath, "utf8"));
460
461
  mcp.mcpServers.navarch_browser = browserEvidence.config;
461
462
  await node_fs_1.promises.writeFile(mcpConfigPath, JSON.stringify(mcp), { mode: 0o600 });
462
- promptText += "\n\nNavarch supplies capture_visual_evidence through navarch_browser. Use it first for desktop/mobile screenshots of the verified preview URL and source commit. It runs outside the coding-agent sandbox, uploads images as task artifacts, and the runtime attaches them to the delivery PR before completion. Inspect the returned images before accepting the UI. Do not use GitHub's upload UI or add capture tools to the target repository. If the base commit has no runnable UI, report before evidence as not applicable with the base SHA and reason; still capture the current head. Authentication or interactive walkthroughs may require the existing browser tools.\n";
463
+ promptText += "\n\nNavarch supplies capture_visual_evidence through navarch_browser. Use it first for desktop/mobile screenshots of the verified preview URL and source commit. It runs outside the coding-agent sandbox, uploads images as task artifacts, and the runtime attaches them to the delivery PR before completion. Inspect the returned images before accepting the UI. Do not use GitHub's upload UI or add capture tools to the target repository. If the base commit has no runnable UI, report before evidence as not applicable with the base SHA and reason; still capture the current head. For interactive walkthroughs of a service you start on this machine, use navarch_browser's local_browser_open, local_browser_click, local_browser_fill, local_browser_press, local_browser_snapshot, and local_browser_screenshot with http://localhost:<port> or http://127.0.0.1:<port>. That browser also runs outside the sandbox and reaches only this machine's loopback services; prefer it over launching Chromium yourself, which Codex's macOS sandbox kills at startup.\n";
463
464
  }
464
465
  if (gitWorktree.hostedCheckpointRestoreFailure) {
465
466
  log.warn(`hosted checkpoint for ${leaseId} could not be restored: ` +
466
467
  `${(0, redact_cjs_1.redactText)(gitWorktree.hostedCheckpointRestoreFailure, registry.list())}; starting without it.`);
467
468
  }
469
+ if (config.sandboxMode === "host" && config.worktreeGuard && runtime === "codex") {
470
+ codexGuardArgs = (0, worktree_guard_cjs_1.codexWorktreeGuardArgs)({
471
+ workDir,
472
+ worktreePath: gitWorktree.worktreePath,
473
+ repositoryPath: gitWorktree.repositoryPath,
474
+ workspaceRoot: config.workspaceRoot,
475
+ extraRoots: config.guardExtraRoots,
476
+ codexBin: config.codexBin,
477
+ });
478
+ }
468
479
  const restoredCheckpoint = gitWorktree.restoredHostedCheckpoint;
469
480
  if (restoredCheckpoint) {
470
481
  log.info(`resumed hosted checkpoint ${restoredCheckpoint.sha.slice(0, 12)} for ${leaseId} ` +
@@ -701,11 +712,14 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
701
712
  });
702
713
  continue;
703
714
  }
715
+ await checkpointForRecovery(result, "incomplete delivery");
704
716
  deliveryFailureSummary =
705
717
  `Session ended without delivering the task: ${delivery.reason}. ` +
706
718
  `${deliveryCorrectionRetries} correction turn(s) did not recover it. ` +
707
- "The session worktree is deleted at teardown. Pushed commits remain on origin; " +
708
- "uncommitted or unpushed work may be lost. The task needs another attempt.";
719
+ (result.checkpointPath
720
+ ? `Local recovery checkpoint: ${result.checkpointPath}. `
721
+ : `Recovery worktree retained: ${result.retainedWorktreePath}. `) +
722
+ "The task needs another attempt.";
709
723
  log.warn(`failing ${leaseId}: ${deliveryFailureSummary}`);
710
724
  mapping.leaseOutcome = "failed";
711
725
  mapping.exitStatus = "failed";
@@ -757,6 +771,7 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
757
771
  transcript_url: completion.transcript_url,
758
772
  };
759
773
  try {
774
+ completingLease = true;
760
775
  await (0, completion_retry_cjs_1.completeWithTransientRetry)(() => api.completeLease(leaseId, completion), {
761
776
  onRetry: (attempt, delayMs, err) => log.warn(`completion for ${leaseId} did not reach the control plane (${(0, redact_cjs_1.redactText)(String(err), knownSecrets)}); re-sending in ${delayMs / 1000}s (${attempt}).`),
762
777
  });
@@ -778,10 +793,11 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
778
793
  if (!rejection)
779
794
  throw err;
780
795
  const redactedRejection = (0, redact_cjs_1.redactText)(rejection.message, knownSecrets);
796
+ const rejectionDetail = describeResponseBody(`${(0, redact_cjs_1.redactText)(rejection.code, knownSecrets)}: ${redactedRejection}`);
781
797
  const remediable = mapping.leaseOutcome === "completed" && rejection.remediable;
782
798
  if (remediable && completionRemediationRetries < COMPLETION_REMEDIATION_RETRIES) {
783
799
  completionRemediationRetries += 1;
784
- log.warn(`completion for ${leaseId} needs more work; restarting agent turn ${completionRemediationRetries}/${COMPLETION_REMEDIATION_RETRIES} in the same worktree.`);
800
+ log.warn(`completion for ${leaseId} needs more work (${rejectionDetail}); restarting agent turn ${completionRemediationRetries}/${COMPLETION_REMEDIATION_RETRIES} in the same worktree.`);
785
801
  // A new turn is a fresh agent invocation: the rejection alone would
786
802
  // reach an agent that has never seen the task.
787
803
  nextPrompt = (0, prompt_cjs_1.renderCompletionRemediationPrompt)({
@@ -794,22 +810,24 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
794
810
  continue;
795
811
  }
796
812
  if (remediable) {
797
- log.warn(`completion for ${leaseId} is still not ready after ${COMPLETION_REMEDIATION_RETRIES} retries; failing with the control-plane rejection.`);
813
+ log.warn(`completion for ${leaseId} is still not ready after ${COMPLETION_REMEDIATION_RETRIES} retries (${rejectionDetail}); failing with the control-plane rejection.`);
798
814
  }
799
815
  else {
800
816
  // Not something another turn can fix (e.g. review_evidence_tampered).
801
817
  // Fail the lease once, keeping this attempt's real report, cost and
802
818
  // evidence — this is a rejected attempt, not a crashed session.
803
- log.error(`completion for ${leaseId} was rejected as ${rejection.code}, which is not remediable; failing the lease.`);
819
+ log.error(`completion for ${leaseId} was rejected (${rejectionDetail}), which is not remediable; failing the lease.`);
804
820
  }
805
821
  try {
806
822
  await (0, completion_retry_cjs_1.completeWithTransientRetry)(() => api.completeLease(leaseId, {
807
823
  ...completion,
808
824
  status: "failed",
809
- report: remediable
810
- ? redactedRejection
811
- : `${redactedRejection}\n\n---\n\n${completion.report}`,
812
- failure_summary: redactedRejection,
825
+ report: `${redactedRejection}\n\n---\n\n${completion.report}`,
826
+ // Keep the agent's blocker first: the API bounds failure_summary,
827
+ // and retry attribution must not lose it behind a delivery error.
828
+ failure_summary: remediable
829
+ ? `${completion.report.slice(0, 1000)}\n\nControl-plane rejection:\n${redactedRejection.slice(0, 800)}`
830
+ : redactedRejection,
813
831
  exit_status: "failed",
814
832
  }));
815
833
  }
@@ -824,6 +842,9 @@ async function runClaimedSession(deps, claimed, sessionId, lifecycle) {
824
842
  }
825
843
  break;
826
844
  }
845
+ finally {
846
+ completingLease = false;
847
+ }
827
848
  }
828
849
  }
829
850
  catch (err) {
@@ -291,6 +291,13 @@ function codexWorktreeGuardArgs(options, platform = process.platform) {
291
291
  [node_path_1.default.resolve(options.repositoryPath)]: "write",
292
292
  [sessionGhConfigDir(options.workDir)]: "read",
293
293
  };
294
+ // Codex protects a linked worktree's gitdir with a more specific read-only
295
+ // rule. A write grant on the bare repository alone does not override it:
296
+ // fetch/checkout/commit fail on FETCH_HEAD, HEAD.lock, or index.lock. Grant
297
+ // the runtime-created gitdir explicitly, without reopening session metadata.
298
+ const gitdir = linkedWorktreeGitDir(options.worktreePath, options.repositoryPath);
299
+ if (gitdir)
300
+ filesystem[gitdir] = "write";
294
301
  const overlapsWorkspace = (root) => isPathInside(root, workspaceRoot) || (linux && isPathInside(workspaceRoot, root));
295
302
  const readRoots = [
296
303
  ...codexToolReadRoots(process.env, platform),
@@ -333,6 +340,36 @@ function codexWorktreeGuardArgs(options, platform = process.platform) {
333
340
  `permissions.${CODEX_GUARD_PROFILE}.network.domains=${tomlInlineTable(CODEX_GITHUB_NETWORK_DOMAINS)}`,
334
341
  ];
335
342
  }
343
+ function linkedWorktreeGitDir(worktreePath, repositoryPath) {
344
+ let dotGit;
345
+ try {
346
+ dotGit = (0, node_fs_1.readFileSync)(node_path_1.default.join(worktreePath, ".git"), "utf8");
347
+ }
348
+ catch (error) {
349
+ // A regular checkout has a .git directory; tests may use virtual paths.
350
+ const code = error.code;
351
+ if (code === "ENOENT" || code === "EISDIR")
352
+ return undefined;
353
+ throw error;
354
+ }
355
+ const match = /^gitdir: (.+)\s*$/m.exec(dotGit);
356
+ if (!match)
357
+ return undefined;
358
+ const canonical = (root) => {
359
+ try {
360
+ return (0, node_fs_1.realpathSync)(root);
361
+ }
362
+ catch (error) {
363
+ if (error.code === "ENOENT")
364
+ return node_path_1.default.resolve(root);
365
+ throw error;
366
+ }
367
+ };
368
+ const gitdir = canonical(node_path_1.default.resolve(worktreePath, match[1].trim()));
369
+ const repository = canonical(repositoryPath);
370
+ // Never turn a pointer outside the runtime's repository into a new grant.
371
+ return gitdir !== repository && isPathInside(gitdir, repository) ? gitdir : undefined;
372
+ }
336
373
  /**
337
374
  * Mounts only the current session roots into Gemini CLI's native sandbox.
338
375
  * Gemini mounts cwd itself; explicit same-path mounts keep the shared gitdir,
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@sagentlab/navarch-runtime",
3
- "version": "0.1.92",
3
+ "version": "0.1.95",
4
4
  "description": "Navarch machine-side session manager: claims delivery tasks and runs them through local coding agents.",
5
5
  "type": "commonjs",
6
6
  "license": "MIT",