@songsid/agend 2.3.3-beta.1 → 2.3.3-beta.3

package/dist/fleet-manager.js

@@ -67,7 +67,7 @@ const __dirname = dirname(__filename);
 const DEFAULT_OPEN_ACCESS = { mode: "open", allowed_users: [], max_pending_codes: 0, code_expiry_minutes: 0 };
 import { STATUS_EMOJI_CONFIG_KEYS, STATUS_EMOJI_KEYS, STATUS_EMOJI_SUGGESTIONS, TELEGRAM_REACTION_EMOJIS, builtinStatusEmojis, customEmojiValue, emojiImageUrl, normalizeEmoji, previewStatusEmojis, reactionForm, reactionMatchKey, resolveStatusEmojis, statusAvoidList, statusEmojiProblem, statusMatchKey, statusMatchKeys, textForm, } from "./status-emojis.js";
 import { isProbeableRouteTarget } from "./fleet-context.js";
-import { getEffectiveInstanceDefaults, loadFleetConfig, loadRawFleetConfig, DEFAULT_COST_GUARD, DEFAULT_DAILY_SUMMARY, DEFAULT_INSTANCE_CONFIG, autoGeneralInstanceConfig } from "./config.js";
+import { getEffectiveInstanceDefaults, loadFleetConfig, loadRawFleetConfig, loadFleetConfigSource, loadRawFleetConfigSource, DEFAULT_COST_GUARD, DEFAULT_DAILY_SUMMARY, DEFAULT_INSTANCE_CONFIG, autoGeneralInstanceConfig } from "./config.js";
 import { EventLog } from "./event-log.js";
 import { binaryProbe } from "./binary-probe.js";
 import { AVAILABILITY_STALE_MS, BackendAvailabilityStore, hostAvailabilityProbes } from "./backend-availability.js";
@@ -100,7 +100,6 @@ import { escapeTelegramHtml, scheduleClock, scheduleRetryLabel } from "./schedul
 import { DEFAULT_SCHEDULER_CONFIG } from "./scheduler/index.js";
 import { TopicCommands, saveCommandForBackend, parseSaveFilename, parsePauseWakeCommand, parseCompactCommand, SAVE_FILENAME_RE, resolveInstanceContext, forgetInstanceContext, readStatuslineJson, statuslineModel } from "./topic-commands.js";
 import { DailySummary } from "./daily-summary.js";
-import { TaskBoardSummaryClock, buildSummary, isQuiet, renderSummary, summaryMarkFile, summaryTime } from "./task-board-summary.js";
 import { WebhookEmitter } from "./webhook-emitter.js";
 import { TmuxControlClient } from "./tmux-control.js";
 import { safeHandler } from "./safe-async.js";
@@ -235,6 +234,7 @@ import { remoteCardAction, remoteCardChoiceLabel, remoteCardOutcomeText, remoteC
 import { recentChatContext } from "./log-tail.js";
 import { describeSignalSource, recordInternalRequest, withOrigin } from "./fleet-control-audit.js";
 import { handleAgentRequest, ToolNotPermittedError } from "./agent-endpoint.js";
+import { ConfigHistory, ConfigSyntaxError, configIssueText } from "./config-history.js";
 import { ClassicChannelManager, classicWebEchoOn, getClassicBackendChoices, inferClassicChannelType, isSelectableClassicBackend, readClassicLastActivityAt, readClassicRoomNames } from "./classic-channel-manager.js";
 import { assertExplicitInstanceRemoval } from "./instance-removal.js";
 import { KNOWN_BACKENDS, validateFleetConfig } from "./config-validator.js";
@@ -797,8 +797,6 @@ export class FleetManager {
     costGuard = null;
     statuslineWatcher;
     dailySummary = null;
-    /** #2246 step 3: the task board's daily summary for the person (task-board-summary.ts). */
-    taskBoardSummary = null;
     dailyTipScheduler = null;
     /** The task board's wakes (step 2): an instance hears when someone else changes a card it owns or handed out. */
     taskBoardWaker = null;
@@ -909,6 +907,12 @@ export class FleetManager {
     kiroMigrationRuns = new Map();
     /** #1386 "Needs you": the list, its live chat messages and Acknowledge. Started with the fleet (finishStartup). */
     needsYou = null;
+    configHistory = null;
+    configHistoryTimer = null;
+    loadedFleetSource = null;
+    getConfigHistory() {
+        return this.configHistory ??= new ConfigHistory(this.dataDir, () => this.needsYou?.poke(), message => this.logger.warn(message));
+    }
     /** When the fleet first saw an instance crashed, for its "Needs you" age; cleared when it is not. */
     needsCrashedAt = new Map();
     /** Pause reason/time read from the marker, per paused instance; dropped on any attention change of it, and at most
@@ -2408,12 +2412,10 @@ export class FleetManager {
                     const notice = silent
                         ? t(`delivery.fleet_${row.state}_silent_schedule`, row.targetInstance, row.operationId, row.deliveryId, row.lastError ?? "unknown")
                         : t(`delivery.fleet_${row.state}`, row.targetInstance, row.operationId, row.deliveryId);
+                    // #2804: log-only. The person sees a failed or unconfirmed delivery as a Needs-you item (with Acknowledge),
+                    // not as a separate General post; the sending agent still gets its delivery-outcome notice.
                     this.logger.error({ deliveryId: row.deliveryId, operationId: row.operationId, target: row.targetInstance, state: row.state,
                         cause: row.lastError, ...(silent ? { silentSchedule: true } : {}) }, notice);
-                    // A self-scheduled silent heartbeat stays log-only, including the
-                    // operator alert; another source still receives the normal notice.
-                    if (!silent || row.sourceInstance !== row.targetInstance)
-                        this.notifyFleetError(notice);
                 }
             }
         });
@@ -2445,8 +2447,9 @@ export class FleetManager {
             if (event.count)
                 this.needsYou?.poke(); // #1386: rows that expired failed — they wait on someone now
             this.wakeCoordinator?.kick();
+            // #2804: the expired rows are Needs-you items now; the count is logged, not posted to General.
             if (event.count)
-                this.notifyFleetError(t("delivery.expired", event.count, event.uncertain ?? 0));
+                this.logger.warn({ count: event.count, uncertain: event.uncertain ?? 0 }, t("delivery.expired", event.count, event.uncertain ?? 0));
         });
         this.scheduleDeliveryOutboxPump();
     }
@@ -2931,7 +2934,9 @@ export class FleetManager {
                 if ((current?.state === "delivering" || current?.state === "submission_started")
                     && current.targetDaemonBootId === claimed.targetDaemonBootId
                     && current.managerBootId === claimed.managerBootId) {
-                    this.notifyFleetError(t("delivery.lane_waiting", claimed.targetInstance, Math.round(DURABLE_DELIVERY_LANE_ALERT_MS / 60_000), deliveryId));
+                    // #2804: logged, not posted. A delivery still in flight needs nobody yet; if it ends failed or unconfirmed it
+                    // becomes a Needs-you item.
+                    this.logger.warn({ deliveryId, target: claimed.targetInstance }, t("delivery.lane_waiting", claimed.targetInstance, Math.round(DURABLE_DELIVERY_LANE_ALERT_MS / 60_000), deliveryId));
                 }
             }, DURABLE_DELIVERY_LANE_ALERT_MS);
             alertTimer.unref?.();
@@ -3103,7 +3108,8 @@ export class FleetManager {
             // fleet.yaml, sent SIGHUP, and got no reaction and no explanation.
             this.logger.error({ err }, "SIGHUP config reload failed");
             const message = err instanceof Error ? err.message : String(err);
-            this.notifyFleetError(t("fleet.reload_failed", message));
+            if (!(err instanceof ConfigSyntaxError))
+                this.notifyFleetError(t("fleet.reload_failed", message));
         });
     }
     /**
@@ -3339,19 +3345,19 @@ export class FleetManager {
         };
     }
     /** Load fleet.yaml and build routing table */
-    loadConfig(configPath) {
+    loadConfig(configPath, options = {}) {
         this.configPath = configPath;
-        const source = existsSync(configPath) ? readFileSync(configPath, "utf-8") : "{}\n";
-        this.rawFleetDocument = parseDocument(source, { keepSourceTokens: true });
-        if (this.rawFleetDocument.errors.length > 0) {
-            throw new Error(`Invalid fleet.yaml: ${this.rawFleetDocument.errors[0].message}`);
-        }
-        const raw = loadRawFleetConfig(configPath);
-        const loaded = loadFleetConfig(configPath);
+        const history = this.getConfigHistory();
+        history.track("fleet.yaml", configPath);
+        const selected = existsSync(configPath) ? history.read("fleet.yaml", configPath, options.startup) : { source: "{}\n", fallback: false };
+        this.rawFleetDocument = parseDocument(selected.source, { keepSourceTokens: true });
+        const raw = loadRawFleetConfigSource(selected.source);
+        const loaded = loadFleetConfigSource(selected.source);
         this.assertProviderSecretEnvKeys(loaded);
         this.rawFleetConfig = raw;
         this.settingsRuntimeRevision++;
         this.fleetConfig = loaded;
+        this.loadedFleetSource = selected;
         this.savedFleetConfigSnapshot = structuredClone(this.fleetConfig);
         this.warnAboutRemovedLoginMode(loaded);
         this.enforceGuestSwitch();
@@ -3360,6 +3366,8 @@ export class FleetManager {
         // Scope the path to its owning data directory so another inherited home's override cannot enable CW.
         process.env.AGEND_FLEET_DATA_DIR = resolvePath(this.dataDir);
         process.env.AGEND_FLEET_CONFIG_PATH = resolvePath(configPath);
+        if (options.remember !== false && existsSync(configPath))
+            history.accepted("fleet.yaml", configPath, selected);
         return this.fleetConfig;
     }
     relayLoginModeWarned = false;
@@ -8169,7 +8177,7 @@ export class FleetManager {
         this.tmuxExecHelperTimer ??= setInterval(() => tmuxExecHelper.maintain(), 5_000);
         this.tmuxExecHelperTimer.unref?.();
         this.rotateFleetLogs();
-        const fleet = this.loadConfig(configPath);
+        const fleet = this.loadConfig(configPath, { startup: true });
         this.slimFleetConfigAtStartup();
         this.logConfigWarningsAtStartup();
         setLocale(detectLocale(fleet)); // user-facing text language (fleet.yaml defaults.locale / timezone)
@@ -8247,7 +8255,7 @@ export class FleetManager {
         // migrates legacy single-bot entries and names without a suffix. Classic
         // routing does NOT go through the routing engine (single-key, can't hold two
         // bots in one channel) — it resolves per-bot via getInstanceByChannel.
-        this.classicChannels = new ClassicChannelManager(this.dataDir, this.logger);
+        this.classicChannels = new ClassicChannelManager(this.dataDir, this.logger, this.getConfigHistory());
         const classicAdapters = fleet.channels?.length ? fleet.channels : (fleet.channel ? [fleet.channel] : []);
         this.classicChannels.configureAdapters(classicAdapters);
         // The unrecoverable-id report is deliberately NOT sent here: no adapter
@@ -8265,6 +8273,9 @@ export class FleetManager {
         // rooms), before the dashboard serves anything; from now on every change is written back (off the loop).
         this.restoreWebChat(fleet);
         this.startSearchIndex(fleet);
+        // Parse-only source monitoring; it does not apply hand-edited Fleet settings.
+        this.configHistoryTimer = setInterval(() => this.configHistory?.check(), 1_000);
+        this.configHistoryTimer.unref?.();
         // Poll classicBot.yaml for external changes every 30s
         this.classicReloadTimer = setInterval(() => {
             // Attributes the synchronous part, up to the reload's first await (#1235).
@@ -8296,7 +8307,8 @@ export class FleetManager {
             ...fleet.defaults.daily_summary,
         };
         this.dailySummary = new DailySummary(summaryConfig, costGuardConfig.timezone, (text) => {
-            this.postDailySummary(text);
+            if (text !== null)
+                this.postDailySummary(text);
             // Rotate classic channel chat logs daily
             this.classicChannels?.rotateLogs();
             this.rotateInboxes();
@@ -8314,17 +8326,6 @@ export class FleetManager {
             return DailySummary.generateText(this.eventLog, instances, costMap, this.costGuard?.getFleetTotalCents() ?? 0);
         });
         this.dailySummary.start();
-        // #2246 step 3: the task board's daily summary — its own time (defaults.task_board.summary_at, read at every look),
-        // in the fleet's zone; a summary missed while the fleet was down goes out when it is back.
-        this.taskBoardSummary = new TaskBoardSummaryClock({
-            at: () => summaryTime(this.fleetConfig?.defaults?.task_board?.summary_at) ?? null,
-            zone: () => this.usageZone(),
-            now: () => new Date(),
-            mark: summaryMarkFile(join(this.dataDir, "task-board-summary.json")),
-            send: (since, until) => this.postTaskBoardSummary(since, until),
-            failed: (err) => this.logger.warn({ err }, "The task board's daily summary failed; the next minute tries again"),
-        });
-        this.taskBoardSummary.start();
         // Rotate classic channel chat logs daily (piggyback on daily summary timer)
         this.classicChannels?.rotateLogs();
         this.rotateInboxes();
@@ -8413,7 +8414,7 @@ export class FleetManager {
             fleet.instances[name] = autoGeneralInstanceConfig(fleet.defaults, generalDir, need.adapterId); // #2598: its kiro_ui is the fleet's
             generalsCreated = true;
         }
-        if (generalsCreated)
+        if (generalsCreated && !this.configHistory?.isFallback(this.configPath))
             this.saveFleetConfig();
         if (topicMode && (fleet.channel || fleet.channels?.length)) {
             const schedulerConfig = {
@@ -8589,7 +8590,7 @@ export class FleetManager {
                     fixedGeneral = true;
                 }
             }
-            if (fixedGeneral)
+            if (fixedGeneral && !this.configHistory?.isFallback(this.configPath))
                 this.saveFleetConfig();
             // Auto-create topics AFTER adapter is ready (needs adapter.createTopic)
             await this.topicCommands.autoCreateTopics();
@@ -12616,6 +12617,7 @@ export class FleetManager {
             },
         });
         this.rawFleetConfig = loadRawFleetConfig(this.configPath);
+        this.configHistory?.accepted("fleet.yaml", this.configPath, { source: output, fallback: false });
         // The instance map is defaults-expanded. A defaults edit changes that view even when it writes no instance
         // leaves. Re-resolve from the committed YAML so inherited values cannot become stale overrides on the next save.
         // Keep the FleetConfig object itself: Settings and other request contexts hold it.
@@ -12694,6 +12696,8 @@ export class FleetManager {
     }
     /** One-time upgrade migration; invalid YAML is never rewritten. */
     migrateNoticeDeliveryStorage() {
+        if (this.configHistory?.isFallback(this.configPath))
+            return;
         if (this.fleetConfig && this.configPath && (this.fleetConfig.web?.notify_login === "dm" || this.fleetConfig.defaults.tips === "dm"
             || this.fleetConfig.web?.notice_delivery !== undefined)
             && validateFleetConfig(this.rawFleetConfig).valid) {
@@ -12708,6 +12712,8 @@ export class FleetManager {
         }
     }
     migrateOfficeStorage() {
+        if (this.configHistory?.isFallback(this.configPath))
+            return;
         if (!this.fleetConfig || !this.configPath || !hasLegacyOffice(this.rawFleetConfig) || !validateFleetConfig(this.rawFleetConfig).valid)
             return;
         try {
@@ -12731,6 +12737,8 @@ export class FleetManager {
         }
     }
     slimFleetConfigAtStartup() {
+        if (this.configHistory?.isFallback(this.configPath))
+            return;
         this.migrateOfficeStorage();
         this.migrateNoticeDeliveryStorage();
         const redundantPaths = collectRedundantInstanceDefaultPaths(this.rawFleetConfig);
@@ -14312,51 +14320,6 @@ export class FleetManager {
         }
         return { chatId, opts: {} };
     }
-    /**
-     * #2246 step 3: the task board's daily summary of [since, until) (task-board-summary.ts). Each chat world's General
-     * gets its own instances' cards — a card that no world's instance holds (nobody's, a person's) goes to the primary
-     * one — and each of that world's Generals shows the same text in the web chat. A world with nothing to say gets
-     * nothing. A web-only fleet: everything, in the web chat.
-     */
-    postTaskBoardSummary(since, until) {
-        const board = this.scheduler?.db.board;
-        if (!board)
-            return;
-        const read = board.summaryRead(since.toISOString(), until.toISOString(), this.dispatcherGone);
-        const projects = board.listProjects(true);
-        const stats = board.projectStats(this.dispatcherGone);
-        const names = new Map(projects.map((p) => [p.id, p.name]));
-        const configured = this.getPrimaryAdapterId();
-        const primary = configured !== undefined && this.worlds.has(configured) ? configured : this.worlds.keys().next().value;
-        const worldOf = (assignee) => {
-            const owner = assignee ? this.getInstanceAdapterId(assignee) : undefined;
-            return owner !== undefined && this.worlds.has(owner) ? owner : primary;
-        };
-        const zone = this.usageZone();
-        for (const world of this.worlds.size ? [...this.worlds.keys()] : [undefined]) {
-            const summary = buildSummary(read, since, until, { inChat: (assignee) => worldOf(assignee) === world, projects, stats });
-            if (isQuiet(summary))
-                continue;
-            const text = renderSummary(summary, { zone, t: (key, ...args) => t(key, ...args), projectOf: (c) => (c.projectId === null ? null : names.get(c.projectId) ?? null) });
-            const cards = [...new Set([...summary.review, ...summary.blocked, ...summary.done, ...summary.created].map((c) => c.code))];
-            if (world !== undefined) {
-                const adapter = this.worlds.get(world)?.adapter;
-                const target = this.fleetNoticeTarget(world);
-                if (adapter && target) {
-                    adapter.sendText(target.chatId, text, target.opts).catch(e => this.logger.warn({ err: e, world }, "Failed to send the task board's daily summary"));
-                }
-                else {
-                    this.logger.warn({ world }, "The task board's daily summary has no postable target — set a General topic or channel.options.general_channel_id");
-                }
-            }
-            for (const [name, cfg] of Object.entries(this.fleetConfig?.instances ?? {})) {
-                if (cfg.general_topic !== true || (world !== undefined && this.getInstanceAdapterId(name) !== world))
-                    continue;
-                // #2341 review: with the cards it names, so the public link withholds it while one of them is hidden there.
-                this.emitSseEvent("message", { instance: name, sender: name, role: "status", text, ts: new Date().toISOString(), cards });
-            }
-        }
-    }
     /**
      * Post the daily summary where fleet-wide notices go.
      *
@@ -15066,6 +15029,7 @@ export class FleetManager {
                 pauseReason, pausedAt,
                 crashedAt: crashed ? (this.needsCrashedAt.get(name) ?? null) : null,
                 signIn: pauseReason === "auth" ? this.signInFor(name) : null,
+                usageHold: this.daemons.get(name)?.usageHold?.() ?? null,
             });
         }
         return out;
@@ -15139,6 +15103,9 @@ export class FleetManager {
             },
             tasks: () => this.needsYouTaskItems(),
             backendSignIns: () => this.needsYouBackendSignIns(),
+            configIssues: () => (this.configHistory?.currentIssues() ?? []).map(issue => ({
+                ...issue, detail: configIssueText(issue, t), owner: this.worlds.keys().next().value,
+            })),
             taskAction: action => this.taskActionFromChat(action),
             settings: () => ({
                 liveMessage: this.fleetConfig?.needs_you?.live_message !== false,
@@ -21278,6 +21245,9 @@ Plus the operational skills (fleet-health, instance-lifecycle, scheduling, sessi
         const sharesStopped = this.officeSharesInstance?.close(); // #2636: every Office share ends, and its tunnel
         // #1386: every live "Needs you" message says the fleet stopped (capabilities revoked first) — while the
         // adapters can still edit. Bounded: a platform that does not answer must not hold the shutdown.
+        if (this.configHistoryTimer)
+            clearInterval(this.configHistoryTimer);
+        this.configHistoryTimer = null;
         const needsStopped = this.stopNeedsYou();
         // #1490 P3: stop() now returns a Promise; await it so the final
         // ledger flush completes before process.exit(0) in doStopAll's caller.
@@ -21333,7 +21303,6 @@ Plus the operational skills (fleet-health, instance-lifecycle, scheduling, sessi
         this.clearStatuslineWatchers();
         this.costGuard?.stop();
         this.dailySummary?.stop();
-        this.taskBoardSummary?.stop();
         this.dailyTipScheduler?.stop();
         this.dailyTipScheduler = null;
         if (this.updateCheckTimer) {
@@ -21624,14 +21593,16 @@ Plus the operational skills (fleet-health, instance-lifecycle, scheduling, sessi
         const previousRawConfig = this.rawFleetConfig;
         const previousRawDocument = this.rawFleetDocument;
         const previousSavedSnapshot = this.savedFleetConfigSnapshot;
+        const previousSource = this.loadedFleetSource;
         try {
-            this.loadConfig(this.configPath);
+            this.loadConfig(this.configPath, { remember: false });
         }
         catch (err) {
             this.fleetConfig = oldConfig;
             this.rawFleetConfig = previousRawConfig;
             this.rawFleetDocument = previousRawDocument;
             this.savedFleetConfigSnapshot = previousSavedSnapshot;
+            this.loadedFleetSource = previousSource;
             throw err;
         }
         const validation = validateFleetConfig(this.rawFleetConfig);
@@ -21647,6 +21618,7 @@ Plus the operational skills (fleet-health, instance-lifecycle, scheduling, sessi
             this.rawFleetConfig = previousRawConfig;
             this.rawFleetDocument = previousRawDocument;
             this.savedFleetConfigSnapshot = previousSavedSnapshot;
+            this.loadedFleetSource = previousSource;
             this.observeWebAuthority(); // #2453 review: the configuration back in force is the one seen
             this.logger.error({
                 oldCount,
@@ -21667,6 +21639,8 @@ Plus the operational skills (fleet-health, instance-lifecycle, scheduling, sessi
             // mark every row done and tell the user "changes applied".
             return { rejected: why };
         }
+        if (this.loadedFleetSource)
+            this.getConfigHistory().accepted("fleet.yaml", this.configPath, this.loadedFleetSource);
         this.migrateOfficeStorage();
         this.migrateNoticeDeliveryStorage();
         if (canvasWorkerEnabled(oldConfig?.web) !== this.canvasWorkerEnabled()) {
@@ -21881,7 +21855,7 @@ Plus the operational skills (fleet-health, instance-lifecycle, scheduling, sessi
      * the panel shows the mismatch instead of a button.
      */
     checkStartupSignatureConsistency() {
-        if (!this.configPath) {
+        if (!this.configPath || this.configHistory?.isFallback(this.configPath)) {
             this.fleetSignatureMismatch = null;
             return;
         }