commandmate 0.16.0 → 0.18.0

package/dist/cli/utils/verify-runner.d.ts

@@ -0,0 +1,63 @@
+/**
+ * Verification run driver shared by `commandmate verify` and `wait --verify`
+ * (Issue #1544).
+ *
+ * The server answers a verify request with 202 and a run id rather than a
+ * verdict — gates are whole test suites and builds — so every caller has to
+ * start a run, poll it to a terminal status, and translate that status into an
+ * exit code. That sequence lives here once so `verify` and `wait` cannot drift
+ * apart on what "passed" means.
+ *
+ * @module cli/utils/verify-runner
+ */
+import type { VerificationRunStatus, VerificationRunView } from '../types/api-responses';
+import { ApiClient } from './api-client';
+/** Matches the wait command's cadence; runs are minutes long, not seconds. */
+export declare const VERIFY_POLL_INTERVAL_MS = 5000;
+/** Built-in gate id; mirrors WORK_EVIDENCE_GATE_ID in lib/verification/gate-runner.ts. */
+export declare const WORK_EVIDENCE_GATE_ID = "work-evidence";
+export interface VerificationRequest {
+    worktreeId: string;
+    /** 'manual' for the verify command, 'wait' when chained after wait. */
+    trigger: 'manual' | 'wait';
+    instanceId?: string;
+    /**
+     * Task the run judges. Omitted lets the server resolve the worktree's own
+     * task, which is enough while that task is still open — `wait` names one
+     * because the agent may close it before the run starts (#1620).
+     */
+    taskId?: string;
+    /** Omitted means work-evidence plus every gate declared in verify.yaml. */
+    gateIds?: string[];
+    /** Seconds before the CLI stops polling and reports TIMEOUT. */
+    timeoutSec?: number;
+    /** Stream the final `RESULT <status>` line goes to. Defaults to stdout. */
+    resultStream?: 'stdout' | 'stderr';
+    /** Suppress the RESULT line because the caller prints JSON on stdout instead. */
+    suppressResultLine?: boolean;
+}
+export interface VerificationOutcome {
+    exitCode: number;
+    /** Last observed run; on timeout this is still `running`. */
+    run?: VerificationRunView;
+}
+/**
+ * Translate a terminal run status into a process exit code.
+ *
+ * `error` and `cancelled` mean no verdict was reached, so they take the generic
+ * UNEXPECTED_ERROR code rather than VERIFY_FAILED — a caller branching on 20
+ * must be able to trust that gates actually ran and judged the work.
+ */
+export declare function exitCodeForRunStatus(status: VerificationRunStatus): number;
+/**
+ * Parse `--gates a,b` into a gate id list.
+ * @returns The ids, or null when the value names no gate at all.
+ */
+export declare function parseGateIds(value: string | undefined): string[] | null | undefined;
+/**
+ * Start a verification run and poll it to a verdict.
+ *
+ * @throws ApiError when the run cannot be started or the server stops answering
+ */
+export declare function runVerification(client: ApiClient, request: VerificationRequest): Promise<VerificationOutcome>;
+//# sourceMappingURL=verify-runner.d.ts.map

package/dist/cli/utils/verify-runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"verify-runner.d.ts","sourceRoot":"","sources":["../../../src/cli/utils/verify-runner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAGH,OAAO,KAAK,EAEV,qBAAqB,EACrB,mBAAmB,EAGpB,MAAM,wBAAwB,CAAC;AAChC,OAAO,EAAE,SAAS,EAAiC,MAAM,cAAc,CAAC;AAExE,8EAA8E;AAC9E,eAAO,MAAM,uBAAuB,OAAO,CAAC;AAE5C,0FAA0F;AAC1F,eAAO,MAAM,qBAAqB,kBAAkB,CAAC;AAErD,MAAM,WAAW,mBAAmB;IAClC,UAAU,EAAE,MAAM,CAAC;IACnB,uEAAuE;IACvE,OAAO,EAAE,QAAQ,GAAG,MAAM,CAAC;IAC3B,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;;;;OAIG;IACH,MAAM,CAAC,EAAE,MAAM,CAAC;IAChB,2EAA2E;IAC3E,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;IACnB,gEAAgE;IAChE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,2EAA2E;IAC3E,YAAY,CAAC,EAAE,QAAQ,GAAG,QAAQ,CAAC;IACnC,iFAAiF;IACjF,kBAAkB,CAAC,EAAE,OAAO,CAAC;CAC9B;AAED,MAAM,WAAW,mBAAmB;IAClC,QAAQ,EAAE,MAAM,CAAC;IACjB,6DAA6D;IAC7D,GAAG,CAAC,EAAE,mBAAmB,CAAC;CAC3B;AAUD;;;;;;GAMG;AACH,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,qBAAqB,GAAG,MAAM,CAW1E;AAED;;;GAGG;AACH,wBAAgB,YAAY,CAAC,KAAK,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,EAAE,GAAG,IAAI,GAAG,SAAS,CAOnF;AAqED;;;;GAIG;AACH,wBAAsB,eAAe,CACnC,MAAM,EAAE,SAAS,EACjB,OAAO,EAAE,mBAAmB,GAC3B,OAAO,CAAC,mBAAmB,CAAC,CAyD9B"}

package/dist/cli/utils/verify-runner.js

@@ -0,0 +1,175 @@
+"use strict";
+/**
+ * Verification run driver shared by `commandmate verify` and `wait --verify`
+ * (Issue #1544).
+ *
+ * The server answers a verify request with 202 and a run id rather than a
+ * verdict — gates are whole test suites and builds — so every caller has to
+ * start a run, poll it to a terminal status, and translate that status into an
+ * exit code. That sequence lives here once so `verify` and `wait` cannot drift
+ * apart on what "passed" means.
+ *
+ * @module cli/utils/verify-runner
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.WORK_EVIDENCE_GATE_ID = exports.VERIFY_POLL_INTERVAL_MS = void 0;
+exports.exitCodeForRunStatus = exitCodeForRunStatus;
+exports.parseGateIds = parseGateIds;
+exports.runVerification = runVerification;
+const types_1 = require("../types");
+const api_client_1 = require("./api-client");
+/** Matches the wait command's cadence; runs are minutes long, not seconds. */
+exports.VERIFY_POLL_INTERVAL_MS = 5000;
+/** Built-in gate id; mirrors WORK_EVIDENCE_GATE_ID in lib/verification/gate-runner.ts. */
+exports.WORK_EVIDENCE_GATE_ID = 'work-evidence';
+const GATE_LABELS = {
+    passed: 'PASS',
+    failed: 'FAIL',
+    timeout: 'TIMEOUT',
+    skipped: 'SKIP',
+    error: 'ERROR',
+};
+/**
+ * Translate a terminal run status into a process exit code.
+ *
+ * `error` and `cancelled` mean no verdict was reached, so they take the generic
+ * UNEXPECTED_ERROR code rather than VERIFY_FAILED — a caller branching on 20
+ * must be able to trust that gates actually ran and judged the work.
+ */
+function exitCodeForRunStatus(status) {
+    switch (status) {
+        case 'passed':
+            return types_1.VerifyExitCode.SUCCESS;
+        case 'failed':
+            return types_1.VerifyExitCode.VERIFY_FAILED;
+        case 'not_started':
+            return types_1.VerifyExitCode.NOT_STARTED;
+        default:
+            return types_1.ExitCode.UNEXPECTED_ERROR;
+    }
+}
+/**
+ * Parse `--gates a,b` into a gate id list.
+ * @returns The ids, or null when the value names no gate at all.
+ */
+function parseGateIds(value) {
+    if (value === undefined)
+        return undefined;
+    const ids = value
+        .split(',')
+        .map((id) => id.trim())
+        .filter((id) => id !== '');
+    return ids.length > 0 ? ids : null;
+}
+function formatDetail(gate) {
+    // The work-evidence gate's verdict is a count, not an exit status; surfacing
+    // the counts is what tells a caller *why* a run came back not_started.
+    if (gate.gateId === exports.WORK_EVIDENCE_GATE_ID && gate.logTail) {
+        const match = /commits=(\d+)\s+uncommitted=(\d+)/.exec(gate.logTail);
+        if (match)
+            return `commits=${match[1]}, uncommitted=${match[2]}`;
+    }
+    if (gate.status === 'skipped') {
+        const reason = gate.logTail?.split('\n')[0]?.trim();
+        return reason ? reason.slice(0, 160) : 'not run';
+    }
+    const parts = [];
+    if (gate.exitCode !== null && gate.exitCode !== undefined)
+        parts.push(`exit=${gate.exitCode}`);
+    if (gate.durationMs !== null && gate.durationMs !== undefined) {
+        parts.push(`${(gate.durationMs / 1000).toFixed(1)}s`);
+    }
+    return parts.join(', ');
+}
+function formatGateLine(gate) {
+    const label = GATE_LABELS[gate.status] ?? gate.status.toUpperCase();
+    const detail = formatDetail(gate);
+    return detail ? `GATE ${gate.gateId} ${label} (${detail})` : `GATE ${gate.gateId} ${label}`;
+}
+/**
+ * Print each gate exactly once, when it reaches a terminal status.
+ * @param reported - Gate row ids already printed; mutated in place.
+ */
+function reportGates(gates, reported) {
+    for (const gate of gates) {
+        if (gate.status === 'running' || reported.has(gate.id))
+            continue;
+        reported.add(gate.id);
+        console.error(formatGateLine(gate));
+        // A failing gate without its log forces the caller back to the API to learn
+        // anything actionable, which defeats the point of a CLI verdict.
+        if (gate.status !== 'passed' && gate.logTail) {
+            console.error(gate.logTail.replace(/\n+$/, ''));
+        }
+    }
+}
+function sleep(ms) {
+    return new Promise((resolve) => setTimeout(resolve, ms));
+}
+/**
+ * Turn the verify route's 409 into an error that names the blocking run.
+ *
+ * handleApiError() maps 409 through its default branch to "Unexpected HTTP
+ * status: 409", which tells the caller nothing about what to do next.
+ */
+function withConflictMessage(error, worktreeId) {
+    if (!(error instanceof api_client_1.ApiError) || error.statusCode !== 409)
+        return error;
+    const running = error.payload?.runningRunId;
+    const which = typeof running === 'number' ? ` (run ${running})` : '';
+    return new api_client_1.ApiError(`A verification run is already in progress for '${worktreeId}'${which}. ` +
+        'Wait for it to finish, then retry.', error.exitCode, error.statusCode, error.payload);
+}
+/**
+ * Start a verification run and poll it to a verdict.
+ *
+ * @throws ApiError when the run cannot be started or the server stops answering
+ */
+async function runVerification(client, request) {
+    const { worktreeId } = request;
+    let runId;
+    try {
+        const started = await client.post(`/api/worktrees/${worktreeId}/verify`, {
+            trigger: request.trigger,
+            instanceId: request.instanceId,
+            gateIds: request.gateIds,
+            taskId: request.taskId,
+        });
+        runId = (0, api_client_1.assertResponseShape)(started, ['runId'], `POST /api/worktrees/${worktreeId}/verify`).runId;
+    }
+    catch (error) {
+        throw withConflictMessage(error, worktreeId);
+    }
+    console.error(`Verifying: ${worktreeId} (run ${runId})`);
+    const startedAt = Date.now();
+    const reported = new Set();
+    const runPath = `/api/worktrees/${worktreeId}/verify/runs/${runId}`;
+    let lastRun;
+    while (true) {
+        const body = await client.get(runPath);
+        const run = (0, api_client_1.assertResponseShape)(body, ['run'], `GET ${runPath}`).run;
+        lastRun = run;
+        reportGates(run.gates ?? [], reported);
+        if (run.status !== 'running') {
+            if (!request.suppressResultLine) {
+                const line = `RESULT ${run.status}`;
+                if (request.resultStream === 'stderr')
+                    console.error(line);
+                else
+                    console.log(line);
+            }
+            return { exitCode: exitCodeForRunStatus(run.status), run };
+        }
+        // Checked after the terminal-status branch so a run that finishes exactly on
+        // the deadline still reports its verdict instead of a timeout.
+        if (request.timeoutSec) {
+            const elapsed = (Date.now() - startedAt) / 1000;
+            if (elapsed >= request.timeoutSec) {
+                console.error(`Timeout: verification of ${worktreeId} exceeded ${request.timeoutSec}s ` +
+                    `(run ${runId} is still running server-side)`);
+                return { exitCode: types_1.VerifyExitCode.TIMEOUT, run: lastRun };
+            }
+        }
+        await sleep(exports.VERIFY_POLL_INTERVAL_MS);
+    }
+}

package/dist/server/server.js

@@ -226,6 +226,21 @@ app.prepare().then(() => {
             catch (error) {
                 console.error('Error reconciling Skill operations:', error);
             }
+            // Issue #1543: close verification runs that a crash left in `running`.
+            // Gate execution lives in process memory, so none of them survived the
+            // restart; leaving the rows open would keep their worktrees permanently
+            // 409-locked against a new run. Dynamic import for the same reason as the
+            // Skill reconciler above — do not hoist to a static import.
+            try {
+                const { reconcileOrphanVerificationRuns } = await Promise.resolve().then(() => __importStar(require('./src/lib/verification/verification-reconciler')));
+                const verifyReport = reconcileOrphanVerificationRuns(db);
+                if (verifyReport.runs > 0) {
+                    console.log(`Verification runs reconciled: runs=${verifyReport.runs} gates=${verifyReport.gates}`);
+                }
+            }
+            catch (error) {
+                console.error('Error reconciling verification runs:', error);
+            }
             // Get repository paths from environment variables
             const repositoryPaths = (0, worktrees_1.getRepositoryPaths)();
             // Issue #490: Also include DB-registered repositories (e.g. cloned repos)

package/dist/server/src/config/auto-yes-config.js

@@ -16,6 +16,7 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.MAX_STOP_PATTERN_LENGTH = exports.AUTO_STOP_ERROR_THRESHOLD = exports.AUTO_YES_COUNTDOWN_INTERVAL_MS = exports.FULL_CAPTURE_LINES = exports.REDUCED_CAPTURE_LINES = exports.THINKING_POLLING_INTERVAL_MS = exports.DURATION_LABELS = exports.DEFAULT_AUTO_YES_DURATION = exports.ALLOWED_DURATIONS = void 0;
 exports.isAllowedDuration = isAllowedDuration;
 exports.validateStopPattern = validateStopPattern;
+exports.executeRegexWithTimeout = executeRegexWithTimeout;
 exports.formatTimeRemaining = formatTimeRemaining;
 const safe_regex2_1 = __importDefault(require("safe-regex2"));
 const tmux_pane_config_1 = require("../config/tmux-pane-config");
@@ -92,6 +93,27 @@ function validateStopPattern(pattern) {
         return { valid: false, error: 'Invalid regular expression syntax' };
     }
 }
+/**
+ * Execute a regex test with timeout protection.
+ * Uses synchronous execution with safe-regex2 pre-validation as the primary defense.
+ *
+ * Lives here rather than in auto-yes-state.ts (Issue #1547) so the auto-yes
+ * resolver can use it: the resolver is imported by a 'use client' hook, and
+ * auto-yes-state pulls in path-validator, which imports `fs`.
+ *
+ * @param regex - Pre-compiled RegExp to test
+ * @param text - Text to test against
+ * @param _timeoutMs - Reserved for future timeout implementation (default: 100ms)
+ * @returns true/false for match result, null if execution failed
+ */
+function executeRegexWithTimeout(regex, text, _timeoutMs = 100) {
+    try {
+        return regex.test(text);
+    }
+    catch {
+        return null;
+    }
+}
 // =============================================================================
 // Time Formatting
 // =============================================================================

package/dist/server/src/lib/auto-yes-poller.js

@@ -61,6 +61,9 @@ exports.isAnyPollerActiveForWorktree = isAnyPollerActiveForWorktree;
 const cli_session_1 = require("./session/cli-session");
 const prompt_detector_1 = require("./detection/prompt-detector");
 const auto_yes_resolver_1 = require("./polling/auto-yes-resolver");
+const auto_yes_policy_1 = require("./polling/auto-yes-policy");
+const task_transition_service_1 = require("./tasks/task-transition-service");
+const db_instance_1 = require("./db/db-instance");
 const prompt_answer_sender_1 = require("./prompt-answer-sender");
 const manager_1 = require("./cli-tools/manager");
 const cli_patterns_1 = require("./detection/cli-patterns");
@@ -291,8 +294,26 @@ async function detectAndRespondToPrompt(worktreeId, pollerState, cliToolId, clea
         if (isDuplicatePrompt(pollerState, promptKey)) {
             return 'duplicate';
         }
-        // 3. Resolve auto answer
-        const answer = (0, auto_yes_resolver_1.resolveAutoAnswer)(promptDetection.promptData);
+        // 3. Resolve auto answer under the execution contract's policy (Issue #1547).
+        // Auto-Yes calls detectPrompt directly instead of going through
+        // status-detector, so this is the only place the policy can gate an
+        // auto-answer: a prompt the policy withholds is left for a human, and the
+        // response poller's prompt path (WS broadcast + Web Push, see
+        // polling/response-checker.ts) is what tells them it is waiting.
+        const policy = (0, auto_yes_policy_1.getSessionAutoYesPolicy)(worktreeId, cliToolId, instanceId);
+        const resolution = (0, auto_yes_resolver_1.resolveAutoAnswerWithPolicy)(promptDetection.promptData, policy);
+        if (resolution.suppressedBy) {
+            logger.warn('poller:auto-yes-suppressed-by-policy', {
+                worktreeId,
+                cliToolId,
+                instanceId,
+                reason: resolution.suppressedBy,
+                mode: policy?.mode ?? null,
+                pattern: resolution.pattern,
+                promptType: promptDetection.promptData.type,
+            });
+        }
+        const answer = resolution.answer;
         if (answer === null) {
             return 'no_answer';
         }
@@ -318,6 +339,13 @@ async function detectAndRespondToPrompt(worktreeId, pollerState, cliToolId, clea
         pollerState.lastAnsweredPromptKey = promptKey;
         pollerState.lastAnsweredAt = Date.now();
         logger.info('poller:response-sent', { worktreeId, cliToolId, instanceId });
+        // Issue #1548: the prompt was answered without a human. Raised only after
+        // the keys actually reached tmux, so a send that threw is not logged as an
+        // answer. No-ops when this instance is not running a contract.
+        (0, task_transition_service_1.applyEventToActiveTask)((0, db_instance_1.getDbInstance)(), worktreeId, cliToolId, 
+        // Undefined means the primary instance, which the task lookup identifies
+        // by the tool id itself (see getActiveTaskForInstance).
+        instanceId ?? cliToolId, 'prompt_answered_auto', { promptType: promptDetection.promptData.type });
         // Dynamic imports avoid a module cycle through terminal-broadcast ->
         // current-output-builder -> auto-yes-manager -> this poller.
         const [{ startPolling: startResponsePolling }, { broadcastTerminalSnapshotAfterInteraction }] = await Promise.all([
@@ -455,6 +483,8 @@ function startAutoYesPolling(worktreeId, cliToolId, instanceId) {
         stopCheckBaselineLength: -1,
     };
     autoYesPollerStates.set(compositeKey, pollerState);
+    // A task may have been recorded moments ago; start from a fresh policy read.
+    (0, auto_yes_policy_1.invalidateSessionAutoYesPolicy)(compositeKey);
     // Start polling immediately
     pollerState.timerId = setTimeout(() => {
         pollAutoYes(worktreeId, cliToolId, effectiveInstanceId);
@@ -477,6 +507,7 @@ function stopAutoYesPolling(compositeKey) {
         clearTimeout(pollerState.timerId);
     }
     autoYesPollerStates.delete(compositeKey);
+    (0, auto_yes_policy_1.invalidateSessionAutoYesPolicy)(compositeKey);
     logger.info('poller:stopped', { compositeKey });
 }
 /**
@@ -487,6 +518,7 @@ function stopAllAutoYesPolling() {
         if (pollerState.timerId) {
             clearTimeout(pollerState.timerId);
         }
+        (0, auto_yes_policy_1.invalidateSessionAutoYesPolicy)(key);
         logger.info('poller:stopped', { compositeKey: key, reason: 'shutdown' });
     }
     autoYesPollerStates.clear();

package/dist/server/src/lib/auto-yes-state.js

@@ -11,7 +11,7 @@
  * auto-yes-poller.ts -> auto-yes-state.ts -> path-validator.ts
  */
 Object.defineProperty(exports, "__esModule", { value: true });
-exports.THINKING_CHECK_LINE_COUNT = exports.MAX_CONCURRENT_POLLERS = exports.MAX_CONSECUTIVE_ERRORS = exports.MAX_BACKOFF_MS = exports.DUPLICATE_RETRY_EXPIRY_MS = exports.COOLDOWN_INTERVAL_MS = exports.POLLING_INTERVAL_MS = exports.COMPOSITE_KEY_SEPARATOR = void 0;
+exports.THINKING_CHECK_LINE_COUNT = exports.MAX_CONCURRENT_POLLERS = exports.MAX_CONSECUTIVE_ERRORS = exports.MAX_BACKOFF_MS = exports.DUPLICATE_RETRY_EXPIRY_MS = exports.COOLDOWN_INTERVAL_MS = exports.POLLING_INTERVAL_MS = exports.executeRegexWithTimeout = exports.COMPOSITE_KEY_SEPARATOR = void 0;
 exports.buildCompositeKey = buildCompositeKey;
 exports.extractWorktreeId = extractWorktreeId;
 exports.extractCliToolId = extractCliToolId;
@@ -21,7 +21,6 @@ exports.getAutoYesState = getAutoYesState;
 exports.setAutoYesEnabled = setAutoYesEnabled;
 exports.disableAutoYes = disableAutoYes;
 exports.clearAllAutoYesStates = clearAllAutoYesStates;
-exports.executeRegexWithTimeout = executeRegexWithTimeout;
 exports.checkStopCondition = checkStopCondition;
 exports.deleteAutoYesState = deleteAutoYesState;
 exports.getAutoYesStateCompositeKeys = getAutoYesStateCompositeKeys;
@@ -230,23 +229,11 @@ function clearAllAutoYesStates() {
 // Stop Condition (Issue #314)
 // =============================================================================
 /**
- * Execute a regex test with timeout protection.
- * Uses synchronous execution with safe-regex2 pre-validation as the primary defense.
- *
- * @internal Exported for testing purposes only.
- * @param regex - Pre-compiled RegExp to test
- * @param text - Text to test against
- * @param _timeoutMs - Reserved for future timeout implementation (default: 100ms)
- * @returns true/false for match result, null if execution failed
+ * Re-exported from the shared auto-yes config (Issue #1547 moved it there so the
+ * client-safe auto-yes resolver can share the same guarded execution path).
  */
-function executeRegexWithTimeout(regex, text, _timeoutMs = 100) {
-    try {
-        return regex.test(text);
-    }
-    catch {
-        return null;
-    }
-}
+var auto_yes_config_2 = require("../config/auto-yes-config");
+Object.defineProperty(exports, "executeRegexWithTimeout", { enumerable: true, get: function () { return auto_yes_config_2.executeRegexWithTimeout; } });
 /**
  * Check if the terminal output matches the stop condition pattern.
  * If matched, disables auto-yes for the worktree/agent.
@@ -279,7 +266,7 @@ function checkStopCondition(compositeKey, cleanOutput, onStopMatched) {
     }
     try {
         const regex = new RegExp(autoYesState.stopPattern);
-        const matched = executeRegexWithTimeout(regex, cleanOutput);
+        const matched = (0, auto_yes_config_1.executeRegexWithTimeout)(regex, cleanOutput);
         if (matched === null) {
             // Execution failed - disable to prevent future errors
             logger.warn('stop-condition-check', { detail: String({ compositeKey }) });

package/dist/server/src/lib/db/db.js

@@ -12,7 +12,7 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.removeAgentInstance = exports.addAgentInstance = exports.setAgentInstances = exports.countAgentInstances = exports.getAgentInstance = exports.getAgentInstances = exports.deleteSessionState = exports.clearInProgressMessageId = exports.setInProgressMessageId = exports.updateSessionState = exports.getSessionState = exports.getMessagesByDateRange = exports.ACTIVE_FILTER = exports.markPendingPromptsAsAnswered = exports.updatePromptData = exports.getMessageById = exports.recomputeLastUserMessage = exports.clearLastUserMessage = exports.updateLastUserMessage = exports.deleteMessagesByInstance = exports.deleteMessagesByCliTool = exports.deleteMessageById = exports.deleteAllMessages = exports.getLastMessage = exports.getLastUserMessage = exports.getMessages = exports.updateMessageContent = exports.createMessage = exports.getLastAssistantMessageAt = exports.deleteWorktreesByIds = exports.deleteRepositoryWorktrees = exports.migrateWorktreeIdPreservingChildren = exports.getWorktreesByRepository = exports.getWorktreeIdsByRepository = exports.getInitialBranch = exports.saveInitialBranch = exports.updateVibeLocalContextWindow = exports.updateVibeLocalModel = exports.updateSelectedAgents = exports.updateCliToolId = exports.updateStatus = exports.updateFavorite = exports.updateLastViewedAt = exports.updateWorktreeLink = exports.updateWorktreeDescription = exports.upsertWorktree = exports.getWorktreeById = exports.getRepositories = exports.getWorktrees = exports.initDatabase = void 0;
 exports.getAssistantExecutionById = exports.updateAssistantExecution = exports.createAssistantExecution = exports.archiveAssistantMessagesFrom = exports.archiveAllAssistantMessages = exports.getAssistantMessageById = exports.getAssistantMessages = exports.updateAssistantMessageStatus = exports.createAssistantMessage = exports.updateAssistantConversation = exports.createAssistantConversation = exports.getAssistantConversationByRepositoryAndCliTool = exports.getAssistantConversationById = exports.updateDailyReportContent = exports.saveDailyReport = exports.getDailyReport = exports.recoverStuckSendingTimers = exports.clearTimerHistory = exports.cleanupOldTimers = exports.getPendingTimerCountByWorktree = exports.cancelTimersByWorktree = exports.cancelTimer = exports.updateTimerStatus = exports.getPendingTimers = exports.getTimerById = exports.getTimersByWorktree = exports.createTimer = exports.WORKTREE_TODO_STATUSES = exports.isWorktreeTodoStatus = exports.reorderWorktreeTodos = exports.deleteWorktreeTodo = exports.updateWorktreeTodo = exports.createWorktreeTodo = exports.getWorktreeTodoById = exports.getTodosByWorktreeId = exports.deleteTodo = exports.updateTodo = exports.createTodo = exports.getTodoById = exports.getAllTodos = exports.getTodosByRepositoryId = exports.MemoDbError = exports.reorderMemos = exports.deleteMemo = exports.updateMemo = exports.createMemo = exports.getMemoById = exports.getMemosByWorktreeId = exports.InvalidAgentInstanceError = exports.AgentInstanceLimitError = void 0;
-exports.deletePushSubscriptionByEndpoint = exports.updatePushSubscriptionPreferences = exports.getPushSubscriptionsForKind = exports.getAllPushSubscriptions = exports.getPushSubscriptionByEndpoint = exports.upsertPushSubscription = exports.deleteAssistantSessionState = exports.updateAssistantSessionState = exports.getAssistantSessionState = exports.listRunningAssistantExecutions = exports.getRunningAssistantExecutionByConversation = exports.getLatestAssistantExecutionByConversation = void 0;
+exports.listTaskEvents = exports.ACTIVE_TASK_STATUSES = exports.TERMINAL_TASK_STATUSES = exports.TASK_STATUSES = exports.isTerminalTaskStatus = exports.getActiveTaskForInstance = exports.getActiveTask = exports.listTasks = exports.getTask = exports.createTask = exports.VERIFICATION_TRIGGERS = exports.MAX_RUN_HISTORY_DAYS = exports.MAX_RUN_HISTORY_LIMIT = exports.DEFAULT_RUN_HISTORY_LIMIT = exports.listVerificationRunsForPeriod = exports.listRunningVerificationRuns = exports.getRunningVerificationRun = exports.listVerificationRuns = exports.getVerificationRun = exports.finishGateResult = exports.createGateResult = exports.finishVerificationRun = exports.createVerificationRun = exports.deletePushSubscriptionByEndpoint = exports.updatePushSubscriptionPreferences = exports.getPushSubscriptionsForKind = exports.getAllPushSubscriptions = exports.getPushSubscriptionByEndpoint = exports.upsertPushSubscription = exports.deleteAssistantSessionState = exports.updateAssistantSessionState = exports.getAssistantSessionState = exports.listRunningAssistantExecutions = exports.getRunningAssistantExecutionByConversation = exports.getLatestAssistantExecutionByConversation = void 0;
 // init-db
 var init_db_1 = require("./init-db");
 Object.defineProperty(exports, "initDatabase", { enumerable: true, get: function () { return init_db_1.initDatabase; } });
@@ -150,3 +150,36 @@ Object.defineProperty(exports, "getAllPushSubscriptions", { enumerable: true, ge
 Object.defineProperty(exports, "getPushSubscriptionsForKind", { enumerable: true, get: function () { return push_subscriptions_db_1.getPushSubscriptionsForKind; } });
 Object.defineProperty(exports, "updatePushSubscriptionPreferences", { enumerable: true, get: function () { return push_subscriptions_db_1.updatePushSubscriptionPreferences; } });
 Object.defineProperty(exports, "deletePushSubscriptionByEndpoint", { enumerable: true, get: function () { return push_subscriptions_db_1.deletePushSubscriptionByEndpoint; } });
+// verification-db (verification gate runs, Issue #1542)
+var verification_db_1 = require("./verification-db");
+Object.defineProperty(exports, "createVerificationRun", { enumerable: true, get: function () { return verification_db_1.createVerificationRun; } });
+Object.defineProperty(exports, "finishVerificationRun", { enumerable: true, get: function () { return verification_db_1.finishVerificationRun; } });
+Object.defineProperty(exports, "createGateResult", { enumerable: true, get: function () { return verification_db_1.createGateResult; } });
+Object.defineProperty(exports, "finishGateResult", { enumerable: true, get: function () { return verification_db_1.finishGateResult; } });
+Object.defineProperty(exports, "getVerificationRun", { enumerable: true, get: function () { return verification_db_1.getVerificationRun; } });
+Object.defineProperty(exports, "listVerificationRuns", { enumerable: true, get: function () { return verification_db_1.listVerificationRuns; } });
+Object.defineProperty(exports, "getRunningVerificationRun", { enumerable: true, get: function () { return verification_db_1.getRunningVerificationRun; } });
+Object.defineProperty(exports, "listRunningVerificationRuns", { enumerable: true, get: function () { return verification_db_1.listRunningVerificationRuns; } });
+Object.defineProperty(exports, "listVerificationRunsForPeriod", { enumerable: true, get: function () { return verification_db_1.listVerificationRunsForPeriod; } });
+Object.defineProperty(exports, "DEFAULT_RUN_HISTORY_LIMIT", { enumerable: true, get: function () { return verification_db_1.DEFAULT_RUN_HISTORY_LIMIT; } });
+Object.defineProperty(exports, "MAX_RUN_HISTORY_LIMIT", { enumerable: true, get: function () { return verification_db_1.MAX_RUN_HISTORY_LIMIT; } });
+Object.defineProperty(exports, "MAX_RUN_HISTORY_DAYS", { enumerable: true, get: function () { return verification_db_1.MAX_RUN_HISTORY_DAYS; } });
+Object.defineProperty(exports, "VERIFICATION_TRIGGERS", { enumerable: true, get: function () { return verification_db_1.VERIFICATION_TRIGGERS; } });
+// tasks-db (execution contracts, Issue #1545)
+var tasks_db_1 = require("./tasks-db");
+Object.defineProperty(exports, "createTask", { enumerable: true, get: function () { return tasks_db_1.createTask; } });
+Object.defineProperty(exports, "getTask", { enumerable: true, get: function () { return tasks_db_1.getTask; } });
+Object.defineProperty(exports, "listTasks", { enumerable: true, get: function () { return tasks_db_1.listTasks; } });
+Object.defineProperty(exports, "getActiveTask", { enumerable: true, get: function () { return tasks_db_1.getActiveTask; } });
+Object.defineProperty(exports, "getActiveTaskForInstance", { enumerable: true, get: function () { return tasks_db_1.getActiveTaskForInstance; } });
+Object.defineProperty(exports, "isTerminalTaskStatus", { enumerable: true, get: function () { return tasks_db_1.isTerminalTaskStatus; } });
+Object.defineProperty(exports, "TASK_STATUSES", { enumerable: true, get: function () { return tasks_db_1.TASK_STATUSES; } });
+Object.defineProperty(exports, "TERMINAL_TASK_STATUSES", { enumerable: true, get: function () { return tasks_db_1.TERMINAL_TASK_STATUSES; } });
+Object.defineProperty(exports, "ACTIVE_TASK_STATUSES", { enumerable: true, get: function () { return tasks_db_1.ACTIVE_TASK_STATUSES; } });
+// task-events-db (task state machine log, Issue #1548).
+// `updateTaskStatus` and `insertTaskEvent` are intentionally absent from this
+// barrel: `applyTaskEvent` (@/lib/tasks/task-transition-service) is the only
+// writer of either, and reaching around it would put a status in the table with
+// no event explaining it.
+var task_events_db_1 = require("./task-events-db");
+Object.defineProperty(exports, "listTaskEvents", { enumerable: true, get: function () { return task_events_db_1.listTaskEvents; } });

package/dist/server/src/lib/db/migrations/index.js

@@ -32,6 +32,9 @@ const v45_skill_installations_1 = require("./v45-skill-installations");
 const v46_skill_installations_cascade_1 = require("./v46-skill-installations-cascade");
 const v47_skill_installations_roots_1 = require("./v47-skill-installations-roots");
 const v48_skill_operations_audit_index_1 = require("./v48-skill-operations-audit-index");
+const v49_verification_runs_1 = require("./v49-verification-runs");
+const v50_tasks_1 = require("./v50-tasks");
+const v51_task_events_1 = require("./v51-task-events");
 /**
  * Complete ordered list of all migrations.
  * Used internally by the barrel file (db-migrations.ts); NOT exported from the barrel (DR2-004).
@@ -67,4 +70,7 @@ exports.migrations = [
     ...v46_skill_installations_cascade_1.v46_migrations,
     ...v47_skill_installations_roots_1.v47_migrations,
     ...v48_skill_operations_audit_index_1.v48_migrations,
+    ...v49_verification_runs_1.v49_migrations,
+    ...v50_tasks_1.v50_migrations,
+    ...v51_task_events_1.v51_migrations,
 ];

package/dist/server/src/lib/db/migrations/runner.js

@@ -17,7 +17,7 @@ const path_1 = __importDefault(require("path"));
  * Current schema version
  * Increment this when adding new migrations
  */
-exports.CURRENT_SCHEMA_VERSION = 48;
+exports.CURRENT_SCHEMA_VERSION = 51;
 /**
  * Get current schema version from database
  */

package/dist/server/src/lib/db/migrations/v49-verification-runs.js

@@ -0,0 +1,86 @@
+"use strict";
+/**
+ * Migration v49: verification_runs / verification_gate_results (Issue #1542).
+ *
+ * The first place an *interactive* agent's work gets a recorded verdict.
+ * `execution_logs` already stores status + exit_code + result, but only for
+ * scheduled executions; interactive sessions leave nothing behind but
+ * `chat_messages`, so "the agent finished" has never been distinguishable from
+ * "the agent produced something that passes".
+ *
+ * One run per verification attempt, one gate result per command inside it.
+ * Gate results cascade with their run: a run is the only thing that gives a
+ * gate result meaning, so an orphan row is not a record, it is noise.
+ *
+ * Both status vocabularies are CHECK-constrained rather than validated in the
+ * writer alone. The distinctions they encode are the point of the table and a
+ * typo that lands as a new status value would silently create a bucket nothing
+ * queries:
+ *   run.not_started — the work-evidence gate failed (no commits, no diff);
+ *                     distinct from `failed`, which means the work was judged.
+ *   run.error       — the runner broke before any gate ran (bad config, spawn
+ *                     failure); no verdict was reached about the work at all.
+ *   gate.skipped    — deliberately not run (e.g. skipInPrimaryCheckout); the
+ *                     reason belongs in log_tail so a skip is never mistaken
+ *                     for a pass.
+ *
+ * `task_id` is a free column, not a foreign key: the `tasks` table arrives in
+ * Phase 2 (#1545). Declaring a REFERENCES clause against a table that does not
+ * exist yet would make every insert fail once foreign_keys is ON.
+ *
+ * Timestamps are epoch milliseconds (INTEGER), matching v44 onward.
+ */
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.v49_migrations = void 0;
+exports.v49_migrations = [
+    {
+        version: 49,
+        name: 'add-verification-runs',
+        up: (db) => {
+            db.exec(`
+        CREATE TABLE IF NOT EXISTS verification_runs (
+          id INTEGER PRIMARY KEY AUTOINCREMENT,
+          worktree_id TEXT NOT NULL,
+          instance_id TEXT,
+          task_id TEXT,
+          trigger TEXT NOT NULL CHECK (trigger IN ('manual','wait','api','task')),
+          status TEXT NOT NULL CHECK (status IN ('running','passed','failed','not_started','error','cancelled')),
+          base_ref TEXT,
+          started_at INTEGER NOT NULL,
+          finished_at INTEGER
+        );
+      `);
+            db.exec(`
+        CREATE TABLE IF NOT EXISTS verification_gate_results (
+          id INTEGER PRIMARY KEY AUTOINCREMENT,
+          run_id INTEGER NOT NULL REFERENCES verification_runs(id) ON DELETE CASCADE,
+          gate_id TEXT NOT NULL,
+          command TEXT NOT NULL,
+          status TEXT NOT NULL CHECK (status IN ('running','passed','failed','timeout','skipped','error')),
+          exit_code INTEGER,
+          duration_ms INTEGER,
+          log_tail TEXT,
+          started_at INTEGER NOT NULL,
+          finished_at INTEGER
+        );
+      `);
+            db.exec(`
+        CREATE INDEX IF NOT EXISTS idx_verification_runs_worktree
+          ON verification_runs(worktree_id, started_at DESC);
+      `);
+            db.exec(`
+        CREATE INDEX IF NOT EXISTS idx_verification_gate_results_run
+          ON verification_gate_results(run_id);
+      `);
+        },
+        down: (db) => {
+            // Child first, mirroring the dependency direction. Under
+            // PRAGMA foreign_keys = ON (how db-instance.ts opens the database) the
+            // reverse order also works today, but only because the cascade absorbs
+            // the parent's implicit DELETE — that is a property of the cascade, not
+            // of DROP TABLE.
+            db.exec('DROP TABLE IF EXISTS verification_gate_results;');
+            db.exec('DROP TABLE IF EXISTS verification_runs;');
+        },
+    },
+];