autokap 1.9.2 → 1.9.3

package/dist/cli-runner.js

@@ -217,12 +217,51 @@ export async function runCapture(options) {
     // badge on the preset while it captures. Best-effort + local-only: skip dry
     // runs and cloud runs (AUTOKAP_RUN_ID is set on cloud runners, which own
     // their own capture_runs row). A failure here must never block the capture.
-    if (!options.dryRun && !process.env.AUTOKAP_RUN_ID) {
+    const isLocalRegisteredRun = !options.dryRun && !process.env.AUTOKAP_RUN_ID;
+    if (isLocalRegisteredRun) {
         await postRunStart(config, runId, program.presetId, program.variants.length, options.env);
     }
+    // Ctrl-C handling: a local run that registered itself server-side must never
+    // leave the preset stuck on the "En cours" badge. On SIGINT/SIGTERM we abort
+    // the in-flight capture (so the finally-block cleanup runs) and best-effort
+    // tell the server the run was interrupted — otherwise the capture_runs row
+    // sits in `running` and the preset stays "capturing" until the 15-min stale
+    // cutoff. A caller-supplied abortSignal is forwarded into the same controller
+    // so both paths converge on one cancellation.
+    const abortController = new AbortController();
+    if (options.abortSignal) {
+        if (options.abortSignal.aborted) {
+            abortController.abort(options.abortSignal.reason);
+        }
+        else {
+            options.abortSignal.addEventListener('abort', () => abortController.abort(options.abortSignal.reason), { once: true });
+        }
+    }
+    let interrupted = false;
+    let abortNotified = false;
+    const notifyAborted = async () => {
+        if (abortNotified || !isLocalRegisteredRun)
+            return;
+        abortNotified = true;
+        await postRunAborted(config, runId, program.presetId, options.env);
+    };
+    const onInterrupt = (signal) => {
+        if (interrupted) {
+            // Second interrupt: the user wants out now. Hard-exit with the
+            // conventional 128+SIGINT(2) code; the best-effort notify already fired.
+            process.exit(130);
+        }
+        interrupted = true;
+        logger.warn(`[capture] ${signal} received — stopping the run and notifying AutoKap…`);
+        abortController.abort(new Error('User interrupted (Ctrl-C)'));
+    };
+    if (isLocalRegisteredRun) {
+        process.on('SIGINT', onInterrupt);
+        process.on('SIGTERM', onInterrupt);
+    }
     const runOptions = {
         recoveryChain,
-        abortSignal: options.abortSignal,
+        abortSignal: abortController.signal,
         maxParallelVariants,
         llmConfig,
         presetName: program.presetId,
@@ -292,15 +331,30 @@ export async function runCapture(options) {
     };
     let runResult;
     let cliResult;
+    let runAborted = false;
     try {
         runResult = await executeProgram(program, createAdapter, runOptions);
+        runAborted = interrupted || runResult.error === 'aborted';
         if (runResult.success) {
             logger.info(`[capture] Run completed successfully — ${runResult.telemetry.totalOpcodes} opcodes, ${runResult.telemetry.recoveredOpcodes} recovered, ${runResult.totalDurationMs}ms`);
         }
+        else if (runAborted) {
+            logger.warn('[capture] Run interrupted by the user — skipping artifact upload.');
+        }
         else {
             logger.error(`[capture] Run failed: ${runResult.error}`);
         }
-        if (options.dryRun) {
+        if (runAborted) {
+            // Don't upload partial artifacts for a cancelled run; the server is told
+            // separately (notifyAborted) so the preset leaves the "capturing" state.
+            cliResult = {
+                success: false,
+                runId,
+                runResult,
+                error: 'Capture interrupted (Ctrl-C)',
+            };
+        }
+        else if (options.dryRun) {
             logger.info(`[capture] DRY RUN complete — ${runResult.telemetry.totalOpcodes} opcodes executed, 0 captures, 0 credits charged`);
             cliResult = { success: runResult.success, runId, runResult };
         }
@@ -354,6 +408,16 @@ export async function runCapture(options) {
         }
     }
     finally {
+        if (isLocalRegisteredRun) {
+            process.off('SIGINT', onInterrupt);
+            process.off('SIGTERM', onInterrupt);
+        }
+        // On interruption, mark the run terminal server-side regardless of the
+        // user's debug-log preference, so the preset never stays stuck "capturing".
+        // Idempotent with the error-log flush below (same capture_failed dedupeKey).
+        if (runAborted) {
+            await notifyAborted();
+        }
         // AUT-149: export structured debug logs to AutoKap on capture failure.
         // Best-effort — the LogCollector swallows network errors.
         const shouldExport = options.exportDebugLogs !== false
@@ -693,6 +757,29 @@ async function postRunStart(config, runId, presetId, variantCount, env) {
         logger.warn(`[capture] Run registration error: ${message}`);
     }
 }
+// Best-effort terminal notification when a local run is interrupted (Ctrl-C).
+// Marks the run failed server-side and clears the preset's "capturing" badge,
+// so an aborted capture can't leave the preset stuck "En cours". Never throws.
+async function postRunAborted(config, runId, presetId, env) {
+    try {
+        const response = await fetch(`${config.apiBaseUrl}/api/cli/runs`, {
+            method: 'PATCH',
+            headers: {
+                'Authorization': `Bearer ${config.apiKey}`,
+                'Content-Type': 'application/json',
+            },
+            body: JSON.stringify({ runId, presetId, status: 'aborted', env }),
+            signal: AbortSignal.timeout(10_000),
+        });
+        if (!response.ok) {
+            logger.warn(`[capture] Failed to report interruption (HTTP ${response.status}); the preset may show "in progress" until the stale cutoff`);
+        }
+    }
+    catch (err) {
+        const message = err instanceof Error ? err.message : String(err);
+        logger.warn(`[capture] Interruption report error: ${message}`);
+    }
+}
 async function uploadResults(config, program, result, runId, sessionId, provenance) {
     const artifactJobs = result.variantResults.flatMap((variant) => {
         const variantSpec = program.variants.find((entry) => entry.id === variant.variantId);

package/dist/execution-types.d.ts

@@ -600,10 +600,11 @@ export interface ExecutionProgram {
      */
     deviceConfigs?: Record<string, DeviceConfig>;
     /**
-     * Project-level public URL used to decorate browser mockups. The CLI
+     * Project-level decorative URL used to decorate browser mockups. The CLI
      * substitutes the captured origin (typically a local dev server) with this
      * value via `transformBrowserUrl` before baking it into the browser bar.
-     * Server-resolved from `projects.public_url`.
+     * AUT-269: derived automatically from the project's prod environment base
+     * URL (absent when no prod environment is configured), not a separate field.
      */
     publicUrl?: string;
     /**

package/dist/opcode-runner.js

@@ -156,8 +156,31 @@ export async function executeProgram(program, createAdapter, options = {}) {
     });
     await Promise.all(workers);
     const completedVariantResults = variantResults.filter((result) => Boolean(result));
-    const aborted = options.abortSignal?.aborted && completedVariantResults.length < program.variants.length;
-    const success = !aborted && completedVariantResults.length > 0 && completedVariantResults.every(v => v.success);
+    const aborted = Boolean(options.abortSignal?.aborted) && completedVariantResults.length < program.variants.length;
+    // Fail-closed on incomplete delivery. Previously `success` was computed purely
+    // from the SURVIVING variant results, so a run that silently dropped variants
+    // (a variant finishing "ok" while persisting no artifact) was recorded as a
+    // clean success — a 4-of-24 capture looked identical to a full run, got billed,
+    // was never retried, and surfaced nothing. Require that every variant ran,
+    // succeeded, AND produced its expected number of artifacts; name the deficient
+    // variants so the failure is actionable and the runtime cause is diagnosable
+    // straight from the run log. Dry runs intentionally skip capture opcodes (and
+    // programs with no capture points have nothing to enforce), so they are exempt
+    // from the artifact check but still require every variant to succeed.
+    const expectedArtifactsPerVariant = program.steps.filter((step) => isArtifactProducingOpcode(step.kind)).length;
+    const enforceArtifactCompleteness = !options.dryRun && expectedArtifactsPerVariant > 0;
+    const deficientVariants = program.variants
+        .map((variant, index) => ({ variant, result: variantResults[index] }))
+        .filter(({ result }) => !result
+        || !result.success
+        || (enforceArtifactCompleteness && result.artifacts.length < expectedArtifactsPerVariant))
+        .map(({ variant, result }) => !result
+        ? `${variant.id} (not executed)`
+        : !result.success
+            ? `${variant.id} (failed: ${result.error ?? 'unknown error'})`
+            : `${variant.id} (${result.artifacts.length}/${expectedArtifactsPerVariant} artifacts)`);
+    const incompleteDelivery = deficientVariants.length > 0;
+    const success = !aborted && completedVariantResults.length > 0 && !incompleteDelivery;
     const detectedAppVersion = completedVariantResults.reduce((acc, variantResult) => acc ?? (variantResult.detectedAppVersion ?? null), null);
     // AUT-241 — surface (don't mask) cuts: aggregate every recording warning from
     // each variant's clip/video artifacts. Diagnostic only; never affects success.
@@ -172,7 +195,11 @@ export async function executeProgram(program, createAdapter, options = {}) {
         totalDurationMs: Date.now() - startTime,
         detectedAppVersion,
         warnings: aggregatedWarnings.length ? aggregatedWarnings : undefined,
-        error: aborted ? 'aborted' : (success ? undefined : completedVariantResults.find(v => !v.success)?.error),
+        error: aborted
+            ? 'aborted'
+            : success
+                ? undefined
+                : `incomplete run: ${deficientVariants.length}/${program.variants.length} variant(s) did not deliver expected artifacts — ${deficientVariants.join('; ')}`,
         failureKind: success ? undefined : completedVariantResults.find(v => v.failureKind)?.failureKind,
     };
 }
@@ -320,6 +347,13 @@ function softSkipResult(opcode, index, startTime, reason, telemetry) {
         error: reason,
     };
 }
+/** Opcodes whose ACTION produces a persisted artifact (a screenshot or a finalized
+ * video clip). A passing postcondition does NOT imply the artifact exists, so these
+ * get special handling on the recovery path (executeOpcode → failWithRecovery) and
+ * in the run-level completeness gate (executeProgram). */
+function isArtifactProducingOpcode(kind) {
+    return kind === 'CAPTURE_SCREENSHOT' || kind === 'END_CLIP';
+}
 async function executeOpcode(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, opcodeTimings, artifacts, options, variantId, executionState, artifactPlan, mockDataGroups, currentVariant, credentials) {
     const startTime = Date.now();
     const effectiveTimeoutMs = resolveOpcodeTimeoutMs(opcode);
@@ -333,6 +367,47 @@ async function executeOpcode(opcode, index, adapter, verifier, breaker, recovery
     const getProgress = makeProgressGetter(adapter);
     const actionEffectPolicy = getOpcodeActionEffectPolicy(opcode);
     const isSoft = isSoftOpcodeKind(opcode.kind);
+    // Snapshot so we can tell whether THIS opcode produced its artifact. A recovered
+    // CAPTURE_SCREENSHOT must not pass as a phantom success when no screenshot was
+    // taken — the postcondition (e.g. element_visible) can pass without a capture.
+    const artifactCountAtStart = artifacts.length;
+    // Wraps handleFailure: when recovery succeeds for an artifact-producing opcode
+    // that pushed NO artifact for itself, re-run the capture once so the artifact
+    // truly exists; hard-fail if it still can't. END_CLIP finalization is stateful,
+    // so it hard-fails without re-running. Without this a "recovered"
+    // CAPTURE_SCREENSHOT finishes ok with no screenshot — the silent partial loss.
+    const failWithRecovery = async (reason) => {
+        const failureResult = await handleFailure(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, options, executionState, variantId, currentVariant, startTime, deadlineMs, globalDeadlineMs, effectiveTimeoutMs, reason);
+        if (failureResult.status !== 'recovered'
+            || !isArtifactProducingOpcode(opcode.kind)
+            || artifacts.length > artifactCountAtStart) {
+            return failureResult;
+        }
+        if (opcode.kind === 'CAPTURE_SCREENSHOT') {
+            const recaptureBudgetMs = getRemainingTimeMs(globalDeadlineMs);
+            if (recaptureBudgetMs > 0) {
+                let recapture;
+                try {
+                    recapture = await withTimeout(() => executeOpcodeAction(opcode, index, adapter, artifacts, telemetry, currentVariant, executionState, artifactPlan, mockDataGroups, options, credentials), recaptureBudgetMs);
+                }
+                catch (err) {
+                    recapture = { success: false, error: err instanceof Error ? err.message : String(err) };
+                }
+                if (recapture.success && artifacts.length > artifactCountAtStart) {
+                    logger.debug(`[opcode ${index}] re-captured screenshot after recovery (${reason})`);
+                    return failureResult;
+                }
+            }
+        }
+        return {
+            opcodeIndex: index,
+            kind: opcode.kind,
+            status: 'failed',
+            durationMs: Date.now() - startTime,
+            recoveryAttempts: failureResult.recoveryAttempts ?? 1,
+            error: `recovery succeeded but produced no ${opcode.kind === 'END_CLIP' ? 'clip' : 'screenshot'} artifact: ${reason}`,
+        };
+    };
     // Track page context for circuit breaker
     try {
         const url = await adapter.getCurrentUrl();
@@ -358,7 +433,7 @@ async function executeOpcode(opcode, index, adapter, verifier, breaker, recovery
             logger.debug(`[opcode ${index}] no budget left after captureBeforeState (deadline=${actionDeadlineMs}, now=${Date.now()})`);
             if (isSoft)
                 return softSkipResult(opcode, index, startTime, reason, telemetry);
-            return handleFailure(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, options, executionState, variantId, currentVariant, startTime, deadlineMs, globalDeadlineMs, effectiveTimeoutMs, reason);
+            return failWithRecovery(reason);
         }
         // For mediaMode='video', capture pre-action timing + bbox metadata inside
         // the active clip window only. Opcodes outside a clip are not part of the
@@ -407,7 +482,7 @@ async function executeOpcode(opcode, index, adapter, verifier, breaker, recovery
             const reason = result.error ?? 'action failed';
             if (isSoft)
                 return softSkipResult(opcode, index, startTime, reason, telemetry);
-            return handleFailure(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, options, executionState, variantId, currentVariant, startTime, deadlineMs, globalDeadlineMs, effectiveTimeoutMs, reason);
+            return failWithRecovery(reason);
         }
         // Verify postcondition — extend-on-progress up to the global deadline so a
         // slow action no longer starves it (failure mode #3: clamped to ~1ms).
@@ -417,7 +492,7 @@ async function executeOpcode(opcode, index, adapter, verifier, breaker, recovery
             logger.debug(`[opcode ${index}] no budget left for postcondition check`);
             if (isSoft)
                 return softSkipResult(opcode, index, startTime, reason, telemetry);
-            return handleFailure(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, options, executionState, variantId, currentVariant, startTime, deadlineMs, globalDeadlineMs, effectiveTimeoutMs, reason);
+            return failWithRecovery(reason);
         }
         const runtimePostcondition = resolveRuntimePostcondition(opcode);
         const postStart = Date.now();
@@ -430,7 +505,7 @@ async function executeOpcode(opcode, index, adapter, verifier, breaker, recovery
             const reason = `postcondition failed: ${postcondition.reason}`;
             if (isSoft)
                 return softSkipResult(opcode, index, startTime, reason, telemetry);
-            return handleFailure(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, options, executionState, variantId, currentVariant, startTime, deadlineMs, globalDeadlineMs, effectiveTimeoutMs, reason);
+            return failWithRecovery(reason);
         }
         // Verify action effects through the shared policy. Weak `any_change`
         // postconditions are only meaningful if this verifier observes a real
@@ -446,7 +521,7 @@ async function executeOpcode(opcode, index, adapter, verifier, breaker, recovery
                         `postcondition passed, treating as redundant-but-successful`);
                 }
                 else {
-                    return handleFailure(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, options, executionState, variantId, currentVariant, startTime, deadlineMs, globalDeadlineMs, effectiveTimeoutMs, `action had no effect: ${verification.summary}`);
+                    return failWithRecovery(`action had no effect: ${verification.summary}`);
                 }
             }
         }
@@ -480,7 +555,7 @@ async function executeOpcode(opcode, index, adapter, verifier, breaker, recovery
         const errorMsg = err instanceof Error ? err.message : String(err);
         if (isSoft)
             return softSkipResult(opcode, index, startTime, errorMsg, telemetry);
-        return handleFailure(opcode, index, adapter, verifier, breaker, recoveryChain, telemetry, healerPatches, options, executionState, variantId, currentVariant, startTime, deadlineMs, globalDeadlineMs, effectiveTimeoutMs, errorMsg);
+        return failWithRecovery(errorMsg);
     }
 }
 /** Post-action breathing room (ms) injected between visible interactions

package/dist/postcondition.js

@@ -100,8 +100,8 @@ function compileRoutePattern(pattern) {
     // Support glob-like patterns: ** matches anything (incl. slashes / empty),
     // * matches a single path segment, ? matches one non-slash char.
     // Tokenize in one pass so the `*` rewrite doesn't clobber the `*` produced
-    // by the `**` rewrite (e.g. `/home**` must compile to `^/home.*$`, not
-    // `^/home.[^/]*$` which would reject `/home` itself).
+    // by the `**` rewrite (e.g. `/home**` must compile to `/home.*`, not
+    // `/home.[^/]*` which would reject `/home` itself).
     let regexStr = '';
     for (let i = 0; i < pattern.length; i++) {
         const ch = pattern[i];
@@ -122,7 +122,14 @@ function compileRoutePattern(pattern) {
             regexStr += ch;
         }
     }
-    return new RegExp(`^${regexStr}$`);
+    // Substring (contains) match — NOT anchored. Generated programs author bare
+    // patterns that are either a prefix of the real path (`/projects/` ⊂
+    // `/projects/<id>`) or a nested segment (`/tracking` ⊂
+    // `/projects/<id>/tracking`). An anchored `^…$` could match neither, which
+    // surfaced as a misleading "page stuck, no progress" failure after the
+    // navigation had actually succeeded. Callers needing strict matching pass an
+    // anchored regex (handled above).
+    return new RegExp(regexStr);
 }
 async function checkElementVisible(adapter, selector) {
     // Primary check: use Playwright waitFor (fast, reliable)

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "autokap",
-  "version": "1.9.2",
+  "version": "1.9.3",
   "description": "AI-powered CLI tool for capturing clean screenshots of websites",
   "type": "module",
   "main": "./dist/index.js",