@debugg-ai/debugg-ai-mcp 5.0.0 → 6.0.0

package/dist/handlers/testPageChangesHandler.js

@@ -17,7 +17,7 @@ import { randomUUID } from 'node:crypto';
 import { detectRepoName } from '../utils/gitContext.js';
 import { disposeUnhealthyTunnel } from '../utils/tunnelDisposition.js';
 import { probeLocalPort, probeTunnelHealth, extractTunnelErrorCode } from '../utils/localReachability.js';
-import { extractLocalhostPort } from '../utils/urlParser.js';
+import { extractLocalhostPort, rewriteLocalhostInText } from '../utils/urlParser.js';
 import { getCachedTemplateUuid, getCachedProjectUuid, invalidateTemplateCache, invalidateProjectCache, } from '../utils/handlerCaches.js';
 import { isTransientWorkflowError, transientReasonTag } from '../utils/transientErrors.js';
 import { Telemetry, TelemetryEvents } from '../utils/telemetry.js';
@@ -211,7 +211,7 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                 if (!probe.reachable) {
                     const payload = {
                         error: 'LocalServerUnreachable',
-                        message: `No server listening on 127.0.0.1:${localPort}. Start your dev server on that port before running check_app_in_browser. Probe result: ${probe.code} (${probe.detail ?? 'no detail'}).`,
+                        message: `No server listening on 127.0.0.1:${localPort}.`,
                         detail: {
                             port: localPort,
                             probeCode: probe.code,
@@ -244,10 +244,7 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                     catch (provisionError) {
                         const msg = provisionError instanceof Error ? provisionError.message : String(provisionError);
                         const diag = provisionError instanceof TunnelProvisionError ? ` ${provisionError.diagnosticSuffix()}` : '';
-                        throw new Error(`Failed to provision tunnel for ${ctx.originalUrl}. ` +
-                            `The remote browser needs a secure tunnel to reach your local dev server. ` +
-                            `Make sure your dev server is running on the specified port and try again. ` +
-                            `(Detail: ${msg})${diag}`);
+                        throw new Error(`Failed to provision tunnel for ${ctx.originalUrl}: ${msg}${diag}`);
                     }
                     provision = tunnel;
                     try {
@@ -261,10 +258,7 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                     }
                     catch (tunnelError) {
                         const msg = tunnelError instanceof Error ? tunnelError.message : String(tunnelError);
-                        throw new Error(`Tunnel creation failed for ${ctx.originalUrl}. ` +
-                            `Could not establish a secure connection between the remote browser and your local port. ` +
-                            `Verify your dev server is running and the port is accessible. ` +
-                            `(Detail: ${msg})`);
+                        throw new Error(`Tunnel creation failed for ${ctx.originalUrl}: ${msg}`);
                     }
                     logger.info(`Tunnel ready: ${ctx.targetUrl} (id: ${ctx.tunnelId})`);
                 }
@@ -296,7 +290,7 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                     if (!health.healthy) {
                         const payload = {
                             error: 'TunnelTrafficBlocked',
-                            message: `Tunnel was established but traffic isn't reaching the dev server. ${health.detail ?? ''} Common causes: dev server binds to 0.0.0.0 or ::1 but not 127.0.0.1; dev server crashed; firewall.`,
+                            message: `Tunnel established; a request through it to 127.0.0.1:${extractLocalhostPort(ctx.originalUrl)} failed: ${health.detail ?? health.code}.`,
                             detail: {
                                 code: health.code,
                                 status: health.status,
@@ -346,22 +340,18 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                 : Promise.resolve(undefined),
         ]);
         if (!templateUuid) {
-            throw new Error('App Evaluation Workflow Template not found. ' +
-                'Ensure the template is seeded in the backend (GET /api/v1/workflows/?is_template=true).');
+            throw new Error(`App Evaluation workflow template not found (slug "${getEvalTemplateSlug()}").`);
         }
         // Fail fast + actionable when project_id can't be resolved (pinned backend
         // semantics: project_id is required). Surfacing "link this repo to a
         // project" now — before executeWorkflow — beats letting a backend workflow
         // node fail mid-run several minutes into the evaluation.
         if (!projectUuid) {
-            const detail = repoName
-                ? `Repo "${repoName}" isn't linked to a DebuggAI project.`
-                : 'No git repository was detected to resolve a project from.';
             const payload = {
                 error: 'ProjectRequired',
-                message: `project_id is required but could not be resolved. ${detail} ` +
-                    'Link this repo to a project at https://debugg.ai (Projects → connect your repository), then retry. ' +
-                    'To evaluate a different repo than the current one, pass repoName.',
+                message: repoName
+                    ? `No DebuggAI project found for repo "${repoName}".`
+                    : 'No git repository detected and no repoName passed.',
             };
             logger.warn(`check_app_in_browser: ${payload.message}`);
             return { content: [{ type: 'text', text: JSON.stringify(payload, null, 2) }], isError: true };
@@ -369,7 +359,12 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
         // --- Build context data (camelCase here — axiosTransport auto-converts to snake_case) ---
         const contextData = {
             targetUrl: ctx.targetUrl ?? originalUrl,
-            question: input.description,
+            // The goal is read by a browser in OUR cloud. A goal that still says
+            // "go to http://localhost:3017" sends it to its own loopback, where it
+            // gets connection refused and the run is recorded as the app failing —
+            // with the tunnel working the whole time. Rewrite it onto the same
+            // tunnel the targetUrl above uses.
+            question: rewriteLocalhostInText(input.description, ctx.targetUrl),
         };
         if (projectUuid) {
             contextData.projectId = projectUuid;
@@ -386,7 +381,8 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
             if (input.auth.precondition)
                 auth.precondition = input.auth.precondition;
             // Bead go1m: entryUrl/deepUrl are URLs the run NAVIGATES, so they need the
-            // same localhost→tunnel rewrite `url` gets (contextData.targetUrl above).
+            // same localhost→tunnel rewrite `url` and the goal text get
+            // (contextData.targetUrl and contextData.question above).
             // Forwarded verbatim they reached the remote browser as literal localhost
             // and it dialled its own loopback: offscope_host + ERR_CONNECTION_REFUSED.
             for (const field of ['entryUrl', 'deepUrl']) {
@@ -397,11 +393,8 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                 if (!rewrite.ok) {
                     const payload = {
                         error: 'AuthUrlPortMismatch',
-                        message: `auth.${field} points at localhost:${rewrite.port} but url points at ` +
-                            `localhost:${rewrite.primaryPort}. A single call tunnels exactly one local ` +
-                            `port, so the remote browser cannot reach both. Put the login page and the ` +
-                            `target page on the same port, or drop auth.${field} and pass username/password ` +
-                            `at the top level so the agent signs in on the page it already reached.`,
+                        message: `auth.${field} points at localhost:${rewrite.port}; url points at ` +
+                            `localhost:${rewrite.primaryPort}. One call tunnels one local port.`,
                         detail: { field, authPort: rewrite.port, urlPort: rewrite.primaryPort, url: originalUrl },
                     };
                     logger.warn(`check_app_in_browser: ${payload.message}`);
@@ -546,11 +539,12 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                 // next line (line 183 in services/workflows.ts), so there's no
                 // post-pollExecution progress emission that could race the response.
                 if (TERMINAL_STATUSES.has(exec.status)) {
-                    const terminalOutcome = exec.state?.outcome ?? exec.status;
+                    // Status only: state.outcome is an internal value and can be one the
+                    // user-facing verdict retired ('inconclusive', 'unverified').
                     await progressCallback({
                         progress: TOTAL_STEPS,
                         total: TOTAL_STEPS,
-                        message: `Complete: ${terminalOutcome}`,
+                        message: `Complete: ${exec.status}`,
                     });
                     return;
                 }
@@ -622,89 +616,32 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
         const duration = Date.now() - startTime;
         // --- Format result ---
         const nodes = finalExecution.nodeExecutions ?? [];
-        // subworkflow.run is the current graph shape — carries outcome, actionHistory, screenshot
+        // subworkflow.run carries an inline base64 screenshot on some graph shapes;
+        // only used for the image block below.
         const subworkflowNode = nodes.find(n => n.nodeType === 'subworkflow.run');
-        // surfer.execute_task and brain.step/brain.evaluate are older graph shapes
-        const surferNode = nodes.find(n => n.nodeType === 'surfer.execute_task');
-        // Action trace: brain.step nodes (old) → subworkflow.run actionHistory (new)
-        const brainSteps = nodes
-            .filter(n => n.nodeType === 'brain.step' && n.outputData)
-            .sort((a, b) => a.executionOrder - b.executionOrder);
-        const actionTrace = brainSteps.map((n, i) => {
-            const d = n.outputData.decision ?? n.outputData;
-            return {
-                step: i + 1,
-                action: d.actionType ?? d.action_type,
-                intent: d.intent,
-                target: d.target,
-                value: d.value ?? undefined,
-                success: n.outputData.success ?? n.status === 'success',
-                durationMs: n.executionTimeMs,
-            };
-        });
-        const subworkflowHistory = subworkflowNode?.outputData?.actionHistory;
-        if (actionTrace.length === 0 && Array.isArray(subworkflowHistory) && subworkflowHistory.length > 0) {
-            subworkflowHistory.forEach((step, i) => {
-                actionTrace.push({
-                    step: i + 1,
-                    action: step.actionType ?? step.action_type ?? step.action,
-                    intent: step.intent,
-                    target: step.target,
-                    value: step.value ?? undefined,
-                    success: step.success ?? true,
-                    durationMs: step.durationMs ?? step.duration_ms ?? undefined,
-                });
-            });
-        }
-        // Evaluation: RELAY the backend's block; derive one only for older backends.
-        //
-        // The backend computes `evaluation` from the SAME verdict as the headline
-        // outcome (sentinal-sk5sl.1) precisely so one payload cannot say
-        // outcome='inconclusive' at the top and evaluation.outcome='unknown'
-        // underneath. Rebuilding it here from raw subworkflow node output
-        // reintroduced that contradiction client-side, and worse: the node's
-        // `success` is a bare boolean, so "could not determine" arrived as
-        // passed:false. The backend deliberately sends passed:null for
-        // inconclusive — an undetermined run is not a failed one.
-        //
-        // Deriving from nodes stays ONLY as the pre-contract fallback.
-        const evalNode = nodes.find(n => n.nodeType === 'brain.evaluate');
-        let evaluation;
-        if (finalExecution.evaluation && typeof finalExecution.evaluation === 'object') {
-            evaluation = finalExecution.evaluation;
-        }
-        else if (evalNode?.outputData) {
-            evaluation = {
-                passed: evalNode.outputData.passed,
-                outcome: evalNode.outputData.outcome,
-                reason: evalNode.outputData.reason,
-                verifications: evalNode.outputData.verifications,
-            };
-        }
-        else if (subworkflowNode?.outputData) {
-            const sw = subworkflowNode.outputData;
-            evaluation = {
-                passed: sw.success,
-                outcome: sw.outcome,
-                reason: sw.error || undefined,
-            };
-        }
         // --- Relay the backend's explicit verdict (bead 56kd.2) ---
         // ONE adapter owns the backend-field → MCP mapping (services/verdictAdapter).
-        // We consume the verdict/budget/evidence VERBATIM — no fabricated outcome,
-        // no success default-false, no synthesized 'assertion-mismatch'. A
-        // missing/unknown verdict surfaces as 'inconclusive', never as a failure.
         // On a poll-deadline timeout (bead 56kd.3) pollExecution returns the last
         // observed execution flagged `timedOut`; force outcome 'timeout' since there
-        // is no terminal backend verdict, and shape its partial evidence below.
+        // is no terminal backend verdict.
         const timedOut = finalExecution.timedOut === true;
         const verdict = adaptVerdict(finalExecution, {
-            fallbackBudget: MAX_EXEC_STEPS,
             outcomeOverride: timedOut ? 'timeout' : undefined,
+            // real numbers from the deadline path, so the reason is a record of what
+            // we observed rather than a fixed sentence
+            pollTimeout: timedOut ? finalExecution.pollTimeout : undefined,
         });
-        // Evidence: prefer the backend's contract evidence.actionTrace; fall back to
-        // the legacy node-extracted trace while the backend contract deploys.
-        const relayActionTrace = verdict.actionTrace ?? actionTrace;
+        const relayActionTrace = verdict.actionTrace;
+        // Evaluation: the backend derives it from the SAME verdict as the headline
+        // (sentinal-sk5sl.1), so its passed/outcome/reason are that verdict a second
+        // time — and passed is just outcome === 'pass'. Relay only what the verdict
+        // does not already carry (e.g. verifications); omit it when that is nothing.
+        let evaluation;
+        if (finalExecution.evaluation && typeof finalExecution.evaluation === 'object') {
+            const { passed: _p, outcome: _o, reason: _r, ...rest } = finalExecution.evaluation;
+            if (Object.keys(rest).length > 0)
+                evaluation = rest;
+        }
         // --- Post-hoc tunnel reclassification (bugs z15n, 4bui) ---
         // The pre-flight probe above proves the tunnel was alive when we handed it to
         // the remote browser — it can still die mid-run. When it does, the browser
@@ -769,28 +706,17 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
         }
         const responsePayload = {
             outcome: verdict.outcome,
-            success: verdict.success,
+            ...(verdict.reason ? { reason: verdict.reason } : {}),
+            // platform-98fv.16: never attempted a test — no verdict, the backend's reason verbatim
+            ...(verdict.skipped ? { skipped: true } : {}),
+            ...(verdict.skipReason ? { skipReason: verdict.skipReason } : {}),
             status: finalExecution.status,
-            stepsTaken: verdict.stepsTaken,
-            stepsBudget: verdict.stepsBudget, // from the response (bead 56kd.2)
-            stepsRemaining: verdict.stepsRemaining,
-            targetUrl: originalUrl,
             executionId: executionUuid,
+            targetUrl: originalUrl,
             durationMs: finalExecution.durationMs ?? duration,
         };
-        // failureCategory = the outcome verbatim (fail | inconclusive | error |
-        // timeout); OMITTED on success. No inference, no 'assertion-mismatch'.
-        if (verdict.failureCategory)
-            responsePayload.failureCategory = verdict.failureCategory;
-        if (verdict.reason)
-            responsePayload.reason = verdict.reason;
-        // --- Which identity did the run actually sign in as? ---
-        // A login as the wrong account and an app rejecting the right one produce
-        // the SAME symptom ("login failed") and used to be indistinguishable from
-        // the result, so an environment-default substitution read as an application
-        // bug. Relay the logins verbatim, and when the caller named an account but
-        // an environment default was used anyway, say so explicitly rather than
-        // leaving the caller to infer it from a failed check.
+        if (verdict.budget)
+            responsePayload.budget = verdict.budget;
         // For "navigate and describe what you see", the answer IS the deliverable —
         // and it reached callers only incidentally, buried in actionTrace[0].intent,
         // which the verify-gate rewrites when it disagrees with the page. Surface it.
@@ -800,10 +726,12 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
             responsePayload.logins = verdict.logins;
         if (verdict.loginError)
             responsePayload.loginError = verdict.loginError;
-        // Bead b5x6: only a SUBMITTED env-default login under a DIFFERENT account
-        // counts. A refused/skipped login (offscope_host, no form) typed nothing,
-        // and the env's stored credential may be the very account that was named —
-        // warning on either states something that did not happen.
+        // Bead b5x6: the caller named an account and the run SUBMITTED an
+        // environment default for a different one. Stated as the two facts — what
+        // was asked for, what was used — and nothing about what to conclude.
+        // Only a SUBMITTED env-default login under a DIFFERENT account counts: a
+        // refused/skipped login typed nothing, and the env's stored credential may
+        // be the very account that was named.
         const requestedIdentity = input.username
             ?? input.auth?.username
             ?? input.loginCredentials?.[0]?.username;
@@ -816,43 +744,26 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
             responsePayload.credentialWarning = {
                 requested: requestedIdentity,
                 used: [...new Set(substituted.map(l => l.username))],
-                message: `This run signed in with an environment default credential even though ` +
-                    `'${requestedIdentity}' was specified. Treat a login failure here as a ` +
-                    `credential-resolution problem, not an application failure.`,
             };
             logger.warn(`check_app_in_browser: requested identity '${requestedIdentity}' but the run used ` +
                 `environment-default credential(s): ${responsePayload.credentialWarning.used.join(', ')}`);
         }
-        // Bug z15n: OUR tunnel died mid-run, so this 'fail' describes our error page,
-        // not the user's app. Relay it as a distinct, retryable infrastructure class
-        // so an automated caller doesn't record a UI regression that never happened.
-        // The backend's original verdict is preserved verbatim, never swallowed.
-        //
-        // Bug 4bui: the stated cause leads with the MARKER, because the marker is the
-        // evidence that justified getting here — the run itself recorded the tunnel's
-        // interstitial. The probe only corroborates (and may legitimately say the
-        // tunnel has recovered by now), so it rides in `detail`, never as the cause.
+        // Bug z15n: the run itself recorded our tunnel's DEBUGG_TUNNEL_* interstitial,
+        // so its 'fail' describes our error page, not the user's app. That is an
+        // MCP-observed fact, so the MCP may state it: outcome 'error', a reason
+        // built from the marker and the probe result, the backend's own verdict kept
+        // verbatim under `backendVerdict`. No commentary on top.
         if (tunnelFault) {
             const { probe, tunnelErrorCode } = tunnelFault;
             responsePayload.backendVerdict = { outcome: verdict.outcome, reason: verdict.reason };
             responsePayload.outcome = 'error';
-            responsePayload.success = false;
-            responsePayload.failureCategory = 'infrastructure';
             responsePayload.error = 'TunnelOfflineDuringRun';
-            responsePayload.message =
-                'During this run the remote browser landed on our tunnel\'s error page instead of your ' +
-                    'app, so the check evaluated our error page rather than your UI. This is an infrastructure ' +
-                    'fault on our side, not a failed check — retry it. The backend\'s original verdict is ' +
-                    'preserved under `backendVerdict`.';
-            responsePayload.reason =
-                `The run recorded our tunnel's ${tunnelErrorCode} interstitial, so the remote browser reached ` +
-                    'our tunnel error page instead of your app.' +
-                    (probe
-                        ? probe.healthy
-                            ? ' (Our re-probe after the run found the tunnel reachable again, so it has since recovered.)'
-                            : ` (Our re-probe after the run also failed: ${probe.code}` +
-                                `${probe.status ? ` (HTTP ${probe.status})` : ''}.)`
-                        : '');
+            const probeFact = probe
+                ? probe.healthy
+                    ? 'post-run tunnel probe: reachable'
+                    : `post-run tunnel probe: ${probe.code}${probe.status ? ` (HTTP ${probe.status})` : ''}`
+                : 'post-run tunnel probe: not run';
+            responsePayload.reason = `The run recorded the tunnel error page ${tunnelErrorCode}; ${probeFact}.`;
             responsePayload.detail = {
                 tunnelErrorCode,
                 probeCode: probe?.code,
@@ -869,19 +780,14 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
             responsePayload.agentError = finalExecution.state.error;
         if (finalExecution.errorMessage)
             responsePayload.errorMessage = finalExecution.errorMessage;
-        if (finalExecution.errorInfo?.failedNodeId)
-            responsePayload.failedNode = finalExecution.errorInfo.failedNodeId;
+        if (finalExecution.errorInfo)
+            responsePayload.errorInfo = finalExecution.errorInfo;
         if (executeResponse.resolvedEnvironmentId)
             responsePayload.resolvedEnvironmentId = executeResponse.resolvedEnvironmentId;
         if (executeResponse.resolvedCredentialId)
             responsePayload.resolvedCredentialId = executeResponse.resolvedCredentialId;
-        if (surferNode?.outputData) {
-            responsePayload.surferOutput = sanitizeResponseUrls(surferNode.outputData, ctx);
-        }
-        // Backend release 2026-04-25: browser_session block on execution detail
-        // carries presigned S3 URLs for HAR + console log + recording. Pass through
-        // verbatim — sanitizeResponseUrls below only strips tunnel hosts so S3 URLs
-        // are preserved. Resolves client-feedback items #1 (network) + #7 (console).
+        // browser_session block: presigned S3 URLs for HAR + console log + recording,
+        // passed through verbatim.
         if (finalExecution.browserSession) {
             responsePayload.browserSession = finalExecution.browserSession;
         }
@@ -962,7 +868,6 @@ async function testPageChangesHandlerInner(input, context, rawProgressCallback)
                 ? [resourceLinkBlock(gifUrl, 'run-recording.gif', {
                         mimeType: 'image/gif',
                         title: 'Run recording',
-                        description: 'Animated recording of the run (presigned URL — open or fetch on demand).',
                     })]
                 : []),
             ...artifactResourceLinks(sanitizedPayload.browserSession),

package/dist/handlers/triggerCrawlHandler.js

@@ -11,6 +11,7 @@
  * and reports back what happened.
  */
 import { config } from '../config/index.js';
+import { toUserFacingOutcome, noVerdictReason } from '../services/verdictAdapter.js';
 import { Logger } from '../utils/logger.js';
 import { handleExternalServiceError } from '../utils/errors.js';
 import { DebuggAIServerClient } from '../services/index.js';
@@ -92,7 +93,7 @@ export async function triggerCrawlHandler(input, context, rawProgressCallback) {
                 if (!probe.reachable) {
                     const payload = {
                         error: 'LocalServerUnreachable',
-                        message: `No server listening on 127.0.0.1:${localPort}. Start your dev server on that port before running trigger_crawl. Probe result: ${probe.code} (${probe.detail ?? 'no detail'}).`,
+                        message: `No server listening on 127.0.0.1:${localPort}.`,
                         detail: { port: localPort, probeCode: probe.code, probeDetail: probe.detail, elapsedMs: probe.elapsedMs },
                     };
                     logger.warn(`Pre-flight port probe failed for ${ctx.originalUrl}: ${probe.code} in ${probe.elapsedMs}ms`);
@@ -119,9 +120,7 @@ export async function triggerCrawlHandler(input, context, rawProgressCallback) {
                     catch (provisionError) {
                         const msg = provisionError instanceof Error ? provisionError.message : String(provisionError);
                         const diag = provisionError instanceof TunnelProvisionError ? ` ${provisionError.diagnosticSuffix()}` : '';
-                        throw new Error(`Failed to provision tunnel for ${ctx.originalUrl}. ` +
-                            `The remote browser needs a secure tunnel to reach your local dev server. ` +
-                            `(Detail: ${msg})${diag}`);
+                        throw new Error(`Failed to provision tunnel for ${ctx.originalUrl}: ${msg}${diag}`);
                     }
                     provision = tunnel;
                     ctx = await ensureTunnel(ctx, tunnel.tunnelKey, tunnel.tunnelId, tunnel.keyId, () => client.tunnels.revoke(tunnel), tunnel);
@@ -151,7 +150,7 @@ export async function triggerCrawlHandler(input, context, rawProgressCallback) {
                     if (!health.healthy) {
                         const payload = {
                             error: 'TunnelTrafficBlocked',
-                            message: `Tunnel was established but traffic isn't reaching the dev server. ${health.detail ?? ''} Common causes: dev server binds to 0.0.0.0 or ::1 but not 127.0.0.1; dev server crashed; firewall.`,
+                            message: `Tunnel established; a request through it to 127.0.0.1:${extractLocalhostPort(ctx.originalUrl)} failed: ${health.detail ?? health.code}.`,
                             detail: {
                                 code: health.code,
                                 status: health.status,
@@ -186,9 +185,7 @@ export async function triggerCrawlHandler(input, context, rawProgressCallback) {
             return client.workflows.findTemplateBySlug(templateSlug);
         });
         if (!templateUuid) {
-            throw new Error(`Crawl Workflow Template not found (slug "${templateSlug}"). ` +
-                `Ensure the backend has that template seeded and accessible ` +
-                `(GET /api/v1/workflows/?slug=${templateSlug}).`);
+            throw new Error(`Crawl workflow template not found (slug "${templateSlug}").`);
         }
         // --- Build contextData + env ---
         const contextData = {
@@ -293,13 +290,29 @@ export async function triggerCrawlHandler(input, context, rawProgressCallback) {
             targetUrl: ctx.originalUrl,
             durationMs: finalExecution.durationMs ?? duration,
         };
-        const outcome = finalExecution.state?.outcome;
-        if (outcome !== undefined && outcome !== null)
-            responsePayload.outcome = outcome;
+        // Map onto the user-facing enum instead of relaying raw. This path never
+        // went through adaptVerdict, so it would otherwise emit whatever the
+        // backend sent — including retired non-answers like 'inconclusive'.
+        const rawOutcome = finalExecution.state?.outcome;
+        if (finalExecution.verdict?.skipped === true && !finalExecution.verdict.outcome) {
+            // platform-98fv.16: the run never attempted a test — no verdict, the
+            // backend's recorded reason verbatim. Never the raw run outcome.
+            responsePayload.outcome = null;
+            responsePayload.skipped = true;
+            const skipReason = finalExecution.verdict.reason;
+            if (typeof skipReason === 'string' && skipReason.trim() !== '')
+                responsePayload.skipReason = skipReason;
+        }
+        else if (rawOutcome !== undefined && rawOutcome !== null) {
+            responsePayload.outcome = toUserFacingOutcome(rawOutcome);
+            if (responsePayload.outcome === 'error' && !finalExecution.errorMessage) {
+                responsePayload.reason = noVerdictReason(rawOutcome);
+            }
+        }
         if (finalExecution.errorMessage)
             responsePayload.errorMessage = finalExecution.errorMessage;
-        if (finalExecution.errorInfo?.failedNodeId)
-            responsePayload.failedNode = finalExecution.errorInfo.failedNodeId;
+        if (finalExecution.errorInfo)
+            responsePayload.errorInfo = finalExecution.errorInfo;
         if (executeResponse.resolvedEnvironmentId)
             responsePayload.resolvedEnvironmentId = executeResponse.resolvedEnvironmentId;
         if (executeResponse.resolvedCredentialId)
@@ -310,34 +323,29 @@ export async function triggerCrawlHandler(input, context, rawProgressCallback) {
         if (finalExecution.browserSession) {
             responsePayload.browserSession = finalExecution.browserSession;
         }
-        // Extract crawl metrics from surfer.crawl node (absent in older graph shapes)
-        const crawlNode = nodes.find(n => n.nodeType === 'surfer.crawl');
-        if (crawlNode?.outputData) {
-            const d = crawlNode.outputData;
-            responsePayload.crawlSummary = {
-                pagesDiscovered: d.pagesDiscovered,
-                actionsExecuted: d.actionsExecuted,
-                stepsTaken: d.stepsTaken,
-                transitionsRecorded: d.transitionsRecorded,
-                knowledgeGraphStates: d.knowledgeGraphStates,
-                success: d.success,
-                ...(d.error ? { error: d.error } : {}),
-            };
-        }
-        // Extract KG import result from knowledge_graph.import node (absent in older graph shapes)
-        const kgNode = nodes.find(n => n.nodeType === 'knowledge_graph.import');
-        if (kgNode?.outputData) {
-            const d = kgNode.outputData;
-            responsePayload.knowledgeGraph = {
-                imported: !d.skipped,
-                skipped: d.skipped ?? false,
-                reason: d.reason ?? '',
-                edgesImported: d.edgesImported ?? 0,
-                statesImported: d.statesImported ?? 0,
-                knowledgeGraphId: d.knowledgeGraphId ?? '',
-                ...(Array.isArray(d.importErrors) && d.importErrors.length > 0 ? { importErrors: d.importErrors } : {}),
-            };
-        }
+        // Node output, relayed under the backend's own keys. Only the listed keys,
+        // and only when the node reported them: a missing count is absent, never a
+        // fabricated 0 / '' that reads as "the backend said zero".
+        const pick = (src, keys) => {
+            if (!src)
+                return undefined;
+            const out = {};
+            for (const k of keys)
+                if (src[k] !== undefined && src[k] !== null)
+                    out[k] = src[k];
+            return out;
+        };
+        const crawlSummary = pick(nodes.find(n => n.nodeType === 'surfer.crawl')?.outputData, [
+            'pagesDiscovered', 'actionsExecuted', 'stepsTaken', 'transitionsRecorded',
+            'knowledgeGraphStates', 'success', 'error',
+        ]);
+        if (crawlSummary)
+            responsePayload.crawlSummary = crawlSummary;
+        const knowledgeGraph = pick(nodes.find(n => n.nodeType === 'knowledge_graph.import')?.outputData, [
+            'skipped', 'reason', 'edgesImported', 'statesImported', 'knowledgeGraphId', 'importErrors',
+        ]);
+        if (knowledgeGraph)
+            responsePayload.knowledgeGraph = knowledgeGraph;
         logger.toolComplete('trigger_crawl', duration);
         // Bead 0bq: final progress is emitted INSIDE pollExecution's onUpdate when
         // terminal status is detected. Emitting it here would race the response

package/dist/handlers/updateEnvironmentHandler.js

@@ -134,10 +134,10 @@ export async function updateEnvironmentHandler(input, _context) {
         if (warnings.length > 0)
             payload.credentialWarnings = warnings;
         if (input.authorizedCredentialHosts !== undefined) {
-            const hostsWarning = checkAuthorizedCredentialHostsEcho(input.authorizedCredentialHosts, environment?.authorizedCredentialHosts, 'update');
+            const hostsWarning = checkAuthorizedCredentialHostsEcho(input.authorizedCredentialHosts, environment?.authorizedCredentialHosts);
             if (hostsWarning) {
                 payload.authorizedCredentialHostsWarning = hostsWarning;
-                logger.warn(`update_environment: ${hostsWarning.message}`);
+                logger.warn(`update_environment: authorizedCredentialHosts not persisted`, hostsWarning);
             }
         }
         logger.toolComplete('update_environment', Date.now() - start);

package/dist/services/index.js

@@ -1,3 +1,4 @@
+import { toUserFacingOutcome } from './verdictAdapter.js';
 import { createWorkflowsService } from "./workflows.js";
 import { createTunnelsService } from "./tunnels.js";
 import { AxiosTransport } from "../utils/axiosTransport.js";
@@ -298,7 +299,8 @@ export class DebuggAIServerClient {
         if (!this.tx)
             throw new Error('Client not initialized — call init() first');
         const response = await this.tx.delete(`api/v1/environments/${envUuid}/sessions/`, { params: filters });
-        return { invalidated: response?.invalidated ?? 0 };
+        // absent → null: a count the backend did not report is not a reported 0
+        return { invalidated: response?.invalidated ?? null };
     }
     /**
      * Fetch a single environment by UUID. Throws AxiosError with status 404 if not found.
@@ -500,8 +502,8 @@ export class DebuggAIServerClient {
                 uuid: s.uuid,
                 name: s.name,
                 description: s.description ?? null,
-                runStatus: s.runStatus ?? s.run_status ?? 'NEVER_RUN',
-                testsCount: s.testsCount ?? s.tests_count ?? 0,
+                runStatus: s.runStatus ?? s.run_status ?? null,
+                testsCount: s.testsCount ?? s.tests_count ?? null,
                 passRate: s.passRate ?? s.pass_rate ?? null,
                 lastRunAt: s.lastRunAt ?? s.last_run_at ?? null,
             })),
@@ -536,7 +538,7 @@ export class DebuggAIServerClient {
             agentTaskDescription: t.agentTaskDescription ?? t.agent_task_description ?? '',
             suite: t.suite ?? input.suiteUuid,
             project: t.project ?? input.projectUuid,
-            runCount: t.runCount ?? t.run_count ?? 0,
+            runCount: t.runCount ?? t.run_count ?? null,
         };
     }
     async updateTestCase(testUuid, patch) {
@@ -572,7 +574,7 @@ export class DebuggAIServerClient {
         const s = await this.tx.post(`api/v1/test-suites/${suiteUuid}/run/`, body);
         return {
             suiteUuid,
-            runStatus: s?.runStatus ?? s?.run_status ?? 'PENDING',
+            runStatus: s?.runStatus ?? s?.run_status ?? null,
             testsTriggered: (s?.tests ?? []).length,
         };
     }
@@ -584,7 +586,7 @@ export class DebuggAIServerClient {
         return {
             uuid: s.uuid,
             name: s.name,
-            runStatus: s.runStatus ?? s.run_status ?? 'NEVER_RUN',
+            runStatus: s.runStatus ?? s.run_status ?? null,
             testsCount: tests.length,
             passRate: s.passRate ?? s.pass_rate ?? null,
             lastRunAt: s.lastRunAt ?? s.last_run_at ?? null,
@@ -594,14 +596,18 @@ export class DebuggAIServerClient {
                 return {
                     uuid: t.uuid,
                     name: t.name,
-                    runCount: t.runCount ?? t.run_count ?? 0,
-                    passedRunsCount: t.passedRunsCount ?? t.passed_runs_count ?? 0,
-                    failedRunsCount: t.failedRunsCount ?? t.failed_runs_count ?? 0,
+                    runCount: t.runCount ?? t.run_count ?? null,
+                    passedRunsCount: t.passedRunsCount ?? t.passed_runs_count ?? null,
+                    failedRunsCount: t.failedRunsCount ?? t.failed_runs_count ?? null,
                     passRate: t.passRate ?? t.pass_rate ?? null,
                     lastRun: lastRun ? {
                         uuid: lastRun.uuid,
                         status: lastRun.status,
-                        outcome: lastRun.outcome,
+                        // the user-facing verdict allowlist, same as every other surface;
+                        // a run with no outcome yet keeps its null
+                        outcome: typeof lastRun.outcome === 'string' && lastRun.outcome.trim() !== ''
+                            ? toUserFacingOutcome(lastRun.outcome)
+                            : lastRun.outcome ?? null,
                         executionTime: lastRun.executionTime ?? lastRun.execution_time ?? null,
                         timestamp: lastRun.timestamp,
                     } : null,
@@ -614,8 +620,8 @@ export class DebuggAIServerClient {
             uuid: s.uuid,
             name: s.name,
             description: s.description ?? null,
-            runStatus: s.runStatus ?? s.run_status ?? 'NEVER_RUN',
-            testsCount: s.testsCount ?? s.tests_count ?? 0,
+            runStatus: s.runStatus ?? s.run_status ?? null,
+            testsCount: s.testsCount ?? s.tests_count ?? null,
         };
     }
 }