@bridge4dev/runner 0.49.0 → 0.51.0

package/dist/agent-versions.d.ts

@@ -58,8 +58,39 @@ export interface ProbeResult {
     stdout: string;
     stderr: string;
 }
-/** Run one probe. Rejects on a non-zero exit, a timeout or a missing binary. */
-export type ProbeRunner = (file: string, argv: readonly string[], timeoutMs: number) => Promise<ProbeResult>;
+export interface ProbeOptions {
+    /**
+     * Keep the output of a command that exited non-zero.
+     *
+     * **This is the bug the flag exists for, and it is worth stating in full.**
+     * `codex doctor --json` exits 1 whenever ANY of its checks fails — and three
+     * of them are network checks (`network.provider_reachability`,
+     * `network.websocket_reachability`, `updates.status`) that have nothing to
+     * do with the question we are asking. Measured on a live machine
+     * (codex-cli 0.153.4): with the network reachable, exit 0; with it blocked,
+     * **exit 1 while `runtime.provenance` still reports `status: ok` and
+     * `install method: npm (…)` on stdout**.
+     *
+     * So a passing hiccup on somebody's dev server used to throw away an answer
+     * we already held, the machine reported «install method unknown», and the
+     * «Update» button disappeared from that agent's row for the hour the
+     * measurement is cached. It came back on its own later, which is why it read
+     * as a permanent property of three machines rather than as a fault.
+     *
+     * A verdict about the machine's health is not a verdict about whether the
+     * command told us what we asked. Timeouts and «no such binary» are still
+     * failures — those really are «no answer».
+     */
+    tolerateExitCode?: boolean;
+}
+/** Run one probe. Rejects on a timeout, a missing binary, or (by default) a non-zero exit. */
+export type ProbeRunner = (file: string, argv: readonly string[], timeoutMs: number, options?: ProbeOptions) => Promise<ProbeResult>;
+/**
+ * Exported for its own tests: the interesting behaviour of this function is
+ * entirely in how it reads `execFile`'s error, and a stub `ProbeRunner` — which
+ * is what every other test here uses — cannot reproduce that by construction.
+ */
+export declare const runProbe: ProbeRunner;
 /**
  * Fold whatever an agent calls its install method into the five values the
  * product reasons about.

package/dist/agent-versions.js

@@ -26,12 +26,30 @@ let cached = null;
 export function invalidateAgentVersions() {
     cached = null;
 }
-const runProbe = (file, argv, timeoutMs) => new Promise((resolve, reject) => {
+/**
+ * Exported for its own tests: the interesting behaviour of this function is
+ * entirely in how it reads `execFile`'s error, and a stub `ProbeRunner` — which
+ * is what every other test here uses — cannot reproduce that by construction.
+ */
+export const runProbe = (file, argv, timeoutMs, options) => new Promise((resolve, reject) => {
     const child = execFile(file, [...argv], { timeout: timeoutMs, maxBuffer: PROBE_MAX_BUFFER }, (error, stdout, stderr) => {
-        if (error)
-            reject(error);
-        else
+        if (!error) {
+            resolve({ stdout, stderr });
+            return;
+        }
+        /*
+         * `killed` is how Node reports the timeout it was given, and a process
+         * we killed produced a TRUNCATED page of JSON — parsing that would be
+         * reading half an answer. A missing binary never runs at all and has no
+         * numeric exit code, so it falls through to `reject` too.
+         */
+        const exited = typeof error.code === 'number';
+        const killed = error.killed === true;
+        if (options?.tolerateExitCode && exited && !killed) {
             resolve({ stdout, stderr });
+            return;
+        }
+        reject(error);
     });
     // `claude doctor` reads stdin. An `execFile` child inherits an open pipe
     // nobody ever writes to, so without this EOF the probe would sit there
@@ -117,10 +135,16 @@ export async function measureAgent(runtime, run = runProbe) {
     }
     // The install method is a nice-to-have: it decides whether a BUTTON appears,
     // never whether the version is shown. A doctor that failed leaves `unknown`.
+    //
+    // `tolerateExitCode` — see `ProbeOptions`. The diagnostics command grades the
+    // whole machine and fails its own exit code over an unreachable endpoint,
+    // while answering our question perfectly on stdout.
     let managedBy = 'unknown';
     try {
         const probe = runtime.managedByProbe;
-        const { stdout, stderr } = await run(binary, probe.argv, PROBE_TIMEOUT_MS);
+        const { stdout, stderr } = await run(binary, probe.argv, PROBE_TIMEOUT_MS, {
+            tolerateExitCode: true,
+        });
         const output = `${stdout}\n${stderr}`;
         const source = probe.jsonPath ? readJsonPath(stdout, probe.jsonPath) : output;
         managedBy = normalizeManagedBy(source ? matchProbe(probe, source) : null);

package/dist/index.js

@@ -465,6 +465,12 @@ function runnerCapabilities(apiUrlOverride) {
                 ? ['verify_start', 'verify_status', 'verify_cancel', 'preview_checkout', 'preview_stop']
                 : []),
             ...(agentInstallEnabled ? ['agent_install'] : []),
+            // «Перемерить версии агентов сейчас» — единственный путь, которым машину
+            // можно СПРОСИТЬ про версии; всё остальное она присылает сама и по своему
+            // расписанию. Объявлено отдельно от `agent_install` и БЕЗ его условия:
+            // это чтение, а не установка, и машину, чей владелец запретил ставить из
+            // дашборда, спросить о том, что на ней стоит, по-прежнему можно.
+            'agent_versions_refresh',
         ],
     };
 }

package/dist/supervisor.d.ts

@@ -244,7 +244,17 @@ export declare class Supervisor {
         agent: string;
         from: string | null;
         to: string;
-    }): Promise<void>;
+    }, 
+    /**
+     * Force a fresh probe on a path the reason alone does not describe.
+     *
+     * «Перемерить сейчас» sends `measured` — it IS a plain measurement — but it
+     * is also the one caller that must not be answered out of the hour-long
+     * cache, because a person pressed a button precisely to get past it.
+     */
+    options?: {
+        force?: boolean;
+    }): Promise<AgentVersionsMeasurement | null>;
     private queueVersionChange;
     /**
      * Р13: the agent of a starting session, moved forward in the background.

package/dist/supervisor.js

@@ -259,14 +259,22 @@ export class Supervisor {
      * missing audit lines are not the problem worth solving.
      */
     static MAX_PENDING_VERSION_CHANGES = 8;
-    async publishAgentVersions(reason, changed) {
+    async publishAgentVersions(reason, changed, 
+    /**
+     * Force a fresh probe on a path the reason alone does not describe.
+     *
+     * «Перемерить сейчас» sends `measured` — it IS a plain measurement — but it
+     * is also the one caller that must not be answered out of the hour-long
+     * cache, because a person pressed a button precisely to get past it.
+     */
+    options) {
         try {
             const measure = this.opts.measureAgentVersions ?? measureAgentVersions;
             // After an install — ours or the auto one — the cached number is a lie,
             // and a stale `at` with it would be dropped by the API's ordering guard
             // together with the audit line the install owes. Forcing it here means a
             // caller cannot forget to invalidate the cache first.
-            const measurement = await measure({ force: reason !== 'measured' });
+            const measurement = await measure({ force: options?.force ?? reason !== 'measured' });
             // A plain tick carries an earlier install's news rather than replacing it.
             // The API keys the audit line by the measurement's `at`, so the same
             // change arriving under a later `at` is one line, not two.
@@ -294,11 +302,15 @@ export class Supervisor {
                     void this.publishAgentVersions('measured');
                 }
             }
+            // Handed back so a caller can answer with EXACTLY what was published,
+            // rather than measuring a second time and hoping the two agree.
+            return measurement;
         }
         catch (error) {
             if (changed && reason !== 'measured')
                 this.queueVersionChange({ reason, changed });
             log.warn('supervisor: could not report agent versions', { error: String(error) });
+            return null;
         }
     }
     queueVersionChange(entry) {
@@ -5119,6 +5131,43 @@ export class Supervisor {
                     void this.publishAgentVersions('manual', moved);
                     return;
                 }
+                /**
+                 * «Перемерить версии агентов прямо сейчас» — за кнопкой «Check again».
+                 *
+                 * Единственный путь, которым машину можно СПРОСИТЬ про версии. Всё
+                 * остальное присылает она сама и по своему расписанию: кадр уходит на
+                 * каждом переподключении (из кэша), раз в час по таймеру, и свежий —
+                 * только после установки, которую сделал сам DevBridge.
+                 *
+                 * Отсюда и дыра, которую эта команда закрывает: человек обновил агента
+                 * руками через `claude update` на машине, раннеру об этом никто не
+                 * сказал, и карточка держала старое число до часа. «Новая сессия
+                 * чинила» его побочным эффектом — автообновление шло ставить, честно
+                 * мерило перед установкой и сбрасывало кэш, — а на машине с
+                 * ВЫКЛЮЧЕННЫМ автообновлением не происходило и этого.
+                 *
+                 * Меряем один раз и отдаём измеренное в ответе, а не только кадром:
+                 * кадр и ответ идут по одному сокету, но обрабатываются на той стороне
+                 * независимо, и запрос, который вернулся раньше, чем доехала запись,
+                 * оставил бы на экране ровно то устаревшее число, ради которого кнопку
+                 * и нажали. Кадр при этом всё равно уходит — у него свой путь с
+                 * аудитом, и повторная запись с тем же `at` отбрасывается сторожем
+                 * порядка на стороне API.
+                 */
+                case 'agent_versions_refresh': {
+                    // Один замер на нажатие, и это структура, а не совпадение: публикация
+                    // отдаёт то, что измерила, и ответ собирается из НЕГО. Мерить второй
+                    // раз «для ответа» значило бы держаться за кэш, который вызывающий не
+                    // видит, — а кэш здесь и есть то, мимо чего нажимают кнопку.
+                    const snapshot = await this.publishAgentVersions('measured', undefined, { force: true });
+                    if (!snapshot) {
+                        return void reply({ ok: false, error: 'Could not measure the agents' });
+                    }
+                    return void reply({
+                        ok: true,
+                        result: { at: snapshot.at, reason: 'measured', agents: snapshot.agents },
+                    });
+                }
                 /**
                  * «Would this file reach the agent, and how big is it» (ticket #192).
                  *

package/dist/version.d.ts

@@ -1,2 +1,2 @@
-export declare const RUNNER_VERSION = "0.49.0";
+export declare const RUNNER_VERSION = "0.51.0";
 //# sourceMappingURL=version.d.ts.map

package/dist/version.js

@@ -1,3 +1,3 @@
 // Kept in sync with package.json by the release script (manual for now).
-export const RUNNER_VERSION = '0.49.0';
+export const RUNNER_VERSION = '0.51.0';
 //# sourceMappingURL=version.js.map

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@bridge4dev/runner",
-  "version": "0.49.0",
+  "version": "0.51.0",
   "description": "DevBridge dev runner — connects a dev server to DevBridge and runs agent sessions (Claude Code / Codex)",
   "homepage": "https://bridge4.dev",
   "license": "MIT",