@skyramp/mcp 0.4.4-rc.1 → 0.4.4

package/build/tools/submitReportTool.js

@@ -9,13 +9,11 @@ import { ProgrammingLanguage, TESTBOT_SUPPORTED_TEST_TYPES, TestType, } from "..
 import { STEP_METHOD_DESCRIPTION, stepMethodSchema, } from "../types/StepMethod.js";
 import { DriftAction } from "../types/TestAnalysis.js";
 import { isResult, TestExecutionStatus } from "../types/TestExecution.js";
-import { IssueFoundCategory } from "../types/TestbotReport.js";
+import { EnvironmentFamily, FailureCauseKind, IssueFoundCategory, } from "../types/TestbotReport.js";
 import { StateManager, runArtifactDir, getTestsRepoDir, resolveOwnRunStatePath, } from "../utils/AnalysisStateManager.js";
 import { toolError, testFileMatches } from "../utils/utils.js";
 import { isTestbotEnabled, isUtilsReuseEnabled, } from "../utils/featureFlags.js";
-import { BLOCKER_DESCRIPTION, closingAnswerFor, needsBlocker, REFUSED_ANSWER_PREFIX, refusedAnswerObjections, unknownAnswerObjections, } from "../recommendation/answers.js";
-import { blockersIn, probeBlockers } from "../recommendation/blockerProbes.js";
-import { blockerSchema, blockerText, hasBlocker, refutedEvidence, verifyBlocker, } from "../recommendation/blockers.js";
+import { answerFor, unknownAnswerObjections, } from "../recommendation/answers.js";
 import { checkDeliveredMatchesPlan, checkMaintenanceDelivered, } from "../recommendation/verifiers/deliveredMatchesPlan.js";
 import { checkReportedCategoryMatchesPlan, checkRequirementConflictReported, } from "../recommendation/verifiers/reportedCategory.js";
 import { checkDefectsReported, checkIssueTraceability, } from "../recommendation/verifiers/issueTraceability.js";
@@ -117,17 +115,46 @@ const repositoryField = z
     .trim()
     .min(1)
     .describe("The 'owner/repo' this item pertains to (e.g. 'letsramp/api-insight'). Set it on every item — the primary's owner/repo for primary-repo objections, or a related repo's owner/repo for objections derived from that repository's diff.");
+// One home for the family meanings: a `.describe()` on the base enum is replaced
+// by the `.optional().describe()` at each use site, so the meanings go there.
+const ENVIRONMENT_FAMILY_MEANINGS = "auth_token = an API token is missing or expired. " +
+    "secret = another secret or credential is missing. " +
+    "data_connection = the database or data store is not reachable. " +
+    "app_not_started = the app under test did not start or is not reachable. " +
+    "login = the sign-in step failed. " +
+    "external_dependency = an external service is not configured. " +
+    "recording_replay = recording or replay of the trace failed.";
+const environmentFamilyField = z.enum(EnvironmentFamily);
+// SKYR-4555. Why a Fail or Skipped row did not pass; the row-level refinement on
+// testResultSchema decides when it is required and which shapes are consistent.
+const failureCauseSchema = z
+    .object({
+    kind: z
+        .enum(FailureCauseKind)
+        .describe("application = the code under test misbehaved, or a test written to prove a defect failed as intended. " +
+        "environment = the test could not exercise the behaviour because of the test environment: a missing token or secret, no data connection, the app did not start, sign-in failed, an external service is not configured, or recording or replay failed. Name the gap in `family`. " +
+        "test_defect = the generated test itself is wrong, not the app: an ambiguous locator, a strict-mode clash, a bad assertion setup."),
+    family: environmentFamilyField
+        .optional()
+        .describe("The test-environment gap. Required when kind is 'environment'. Not allowed for the other kinds. " +
+        ENVIRONMENT_FAMILY_MEANINGS),
+})
+    .describe("Why this test did not pass. REQUIRED when status is 'Fail' or 'Skipped'. Not allowed when status is 'Pass'. " +
+    "It must agree with reachedBehavior: 'application' needs reachedBehavior true or omitted; 'environment' needs reachedBehavior false or omitted.");
 // videoPath is deliberately absent from this input contract: it is attached server-side
 // from the run's execution records (see attachVideoPath), and zod strips any the model
 // supplies anyway. SKYR-4156 is what happens when the agent owns that field instead.
-const testResultSchema = z.object({
+const testResultSchema = z
+    .object({
     testType: z
-        .nativeEnum(TestType)
+        .enum(TestType)
         .describe("Type of test. Do not include priority or other metadata in this field."),
     endpoint: z
         .string()
         .describe("For contract and integration tests, the HTTP verb and path, e.g. 'GET /api/v1/products'. For ui and e2e tests, the page path the test opens, e.g. '/notifications'."),
-    status: z.enum(["Pass", "Fail", "Skipped"]).describe("Test execution result"),
+    status: z
+        .enum(["Pass", "Fail", "Skipped"])
+        .describe("Test execution result"),
     details: z
         .string()
         .describe("One sentence — no embedded newlines, no markdown. e.g. '10.8s, products_contract_test.py' or 'failed: <one-line error summary>, products_contract_test.py'"),
@@ -147,7 +174,76 @@ const testResultSchema = z.object({
         .describe("For a failing test only. `true` when the request that carries the behaviour under test was sent and answered by the code under test, and the assertion that failed is about that behaviour. " +
         "`false` when the failure came earlier: a 404, 401, 403 or 5xx on a route the change adds; an expired token or a sign-in page; a control or setup step that failed; a generator or SDK error. " +
         "If your own `details` text says the request was not sent or an earlier step tripped, this is `false`. Omit only if you cannot tell."),
+    failureCause: failureCauseSchema.optional(),
     repository: repositoryField,
+})
+    .superRefine((row, ctx) => {
+    // SKYR-4555: the business-user report sorts failures by cause, so every row
+    // that did not pass must name one and a passing row must not. The kind must
+    // agree with reachedBehavior where the agent set it: the two answer the same
+    // question — did the test reach the behaviour under test — from two sides.
+    const cause = row.failureCause;
+    if (row.status === "Pass") {
+        if (cause === undefined)
+            return;
+        ctx.addIssue({
+            code: "custom",
+            path: ["failureCause"],
+            message: `failureCause is not allowed when status is 'Pass'. A passing test has no failure cause. ` +
+                `Remove failureCause from this row.`,
+        });
+        return;
+    }
+    if (cause === undefined) {
+        ctx.addIssue({
+            code: "custom",
+            path: ["failureCause"],
+            message: `failureCause is required when status is '${row.status}'. Set failureCause.kind to one of: ` +
+                `'application' (the code under test misbehaved, or a test written to prove a defect failed as intended); ` +
+                `'environment' (the test environment stopped the test before it reached the behaviour — also set failureCause.family); ` +
+                `'test_defect' (the generated test itself is wrong, not the app).`,
+        });
+        return;
+    }
+    if (cause.kind === FailureCauseKind.Environment) {
+        if (cause.family === undefined) {
+            ctx.addIssue({
+                code: "custom",
+                path: ["failureCause", "family"],
+                message: `failureCause.family is required when failureCause.kind is 'environment'. ` +
+                    `Set it to the gap that stopped the test: ${Object.values(EnvironmentFamily).join(", ")}.`,
+            });
+        }
+    }
+    else if (cause.family !== undefined) {
+        ctx.addIssue({
+            code: "custom",
+            path: ["failureCause", "family"],
+            message: `failureCause.family is allowed only when failureCause.kind is 'environment'. ` +
+                `Remove failureCause.family, or set kind to 'environment' if the test environment stopped the test.`,
+        });
+    }
+    if (cause.kind === FailureCauseKind.Application) {
+        if (row.reachedBehavior === false) {
+            ctx.addIssue({
+                code: "custom",
+                path: ["failureCause", "kind"],
+                message: `failureCause.kind 'application' contradicts reachedBehavior false. 'application' means the test reached the behaviour under test and the code misbehaved. ` +
+                    `If the failure came earlier, set kind to 'environment' or 'test_defect'. If the test did reach the behaviour, set reachedBehavior to true.`,
+            });
+        }
+    }
+    else if (cause.kind === FailureCauseKind.Environment &&
+        row.reachedBehavior === true) {
+        // test_defect is not checked: a wrong assertion can fail after the test
+        // reached the behaviour, or a bad setup step can fail before it.
+        ctx.addIssue({
+            code: "custom",
+            path: ["failureCause", "kind"],
+            message: `failureCause.kind 'environment' contradicts reachedBehavior true. reachedBehavior true means the test reached the behaviour under test, so the environment did not stop it. ` +
+                `Set kind to 'application' or 'test_defect', or set reachedBehavior to false if the failure came earlier.`,
+        });
+    }
 });
 // Structured grounding for UI recommendations. Lifted verbatim from a
 // `browser_blueprint` element captured during this run. Source of truth for
@@ -237,7 +333,7 @@ export const newTestSchema = z.preprocess(stripNullGroundingFields, z
         "Required on EVERY entry, UI included. That is deliberately wider than the approved-plan generation gate, which stays contract and integration only: " +
         "this id is the exact key the report is joined back to the plan on, and an entry without one falls back to guessing from the test name and endpoint."),
     testType: z
-        .nativeEnum(TestType)
+        .enum(TestType)
         .describe("Type of test created. Do not include priority or other metadata in this field."),
     category: z
         .enum(SCENARIO_CATEGORIES)
@@ -284,7 +380,7 @@ export const newTestSchema = z.preprocess(stripNullGroundingFields, z
     // optional above and required here so the message says what to do.
     if ((rec.plannedTestId ?? "").trim().length === 0) {
         ctx.addIssue({
-            code: z.ZodIssueCode.custom,
+            code: "custom",
             path: ["plannedTestId"],
             message: "plannedTestId is required on every newTestsCreated entry. Use the plannedTestId of the plan entry this test was approved under, from skyramp_register_test_plan.",
         });
@@ -294,7 +390,7 @@ export const newTestSchema = z.preprocess(stripNullGroundingFields, z
         for (const field of ["targetElements", "pageContext"]) {
             if (rec[field] !== undefined) {
                 ctx.addIssue({
-                    code: z.ZodIssueCode.custom,
+                    code: "custom",
                     path: [field],
                     message: `${field} is only valid for testType: 'ui'`,
                 });
@@ -305,7 +401,7 @@ export const newTestSchema = z.preprocess(stripNullGroundingFields, z
         // description is required for UI tests (walkthrough of steps/assertions).
         if (!rec.description) {
             ctx.addIssue({
-                code: z.ZodIssueCode.custom,
+                code: "custom",
                 path: ["description"],
                 message: "description is required for testType: 'ui'.",
             });
@@ -313,7 +409,7 @@ export const newTestSchema = z.preprocess(stripNullGroundingFields, z
         // targetElements must be present (array or null — not omitted entirely).
         if (rec.targetElements === undefined) {
             ctx.addIssue({
-                code: z.ZodIssueCode.custom,
+                code: "custom",
                 path: ["targetElements"],
                 message: "targetElements is required for testType: 'ui'. Use null when blueprint capture failed; the null is the signal and needs no marker in description or reasoning.",
             });
@@ -324,7 +420,7 @@ export const newTestSchema = z.preprocess(stripNullGroundingFields, z
         if (Array.isArray(rec.targetElements) &&
             rec.pageContext === undefined) {
             ctx.addIssue({
-                code: z.ZodIssueCode.custom,
+                code: "custom",
                 path: ["pageContext"],
                 message: "pageContext is required for testType: 'ui' when targetElements is an array.",
             });
@@ -355,7 +451,7 @@ const issueFoundSchema = z
         "report an issue here at 'critical' or 'high'. A stated requirement the code does not meet is not a " +
         "medium. The check raises an objection, never a rejection."),
     category: z
-        .nativeEnum(IssueFoundCategory)
+        .enum(IssueFoundCategory)
         .describe("Issue classification. bug = a product/code defect, e.g. found by a test or in the diff. " +
         "lint = a linter or formatter objection (eslint, flake8, prettier). " +
         "type = a type-check failure (tsc, mypy). " +
@@ -374,7 +470,7 @@ const issueFoundSchema = z
         .optional()
         .describe("The `id` of the plan defect this issue reports, from the `defects` list you registered with skyramp_register_test_plan. Spell it as that list spells it: an id the plan does not declare traces nothing. " +
         "Set it on the `category: bug` entry that reports the defect. A `defectId` on a `lint`, `type` or `config` entry does not report the defect — the report renders those in its own Configuration Errors section. " +
-        "CHECKED: every plan defect has a `bug` issue that names it, and a `bug` entry with no `plannedTestId` traces through this defect when a delivered test that expects to fail cites it, or when a `maintains` entry you registered cites it. Leave it out for an issue the plan did not declare."),
+        "CHECKED: every plan defect has a `bug` issue that names it, and a `bug` entry with no `plannedTestId` traces through this defect when a delivered test that expects to fail cites it, when a `maintains` entry you registered cites it, or when you answered `defects:untested:<id>` at plan time. Leave it out for an issue the plan did not declare."),
     sourceFile: citationString.describe("Path of the application file whose code is missing or wrong, relative to the repository root (e.g. 'src/crud/products.py'). " +
         "REQUIRED when category is 'bug'. " +
         "For code that is MISSING — an unmounted router, an unregistered route, an import never added — name the file where the line should be, not the file that defines what is unmounted. " +
@@ -390,8 +486,25 @@ const issueFoundSchema = z
         .describe("1-based line number in sourceFile. Optional and advisory. " +
         "Give the line of the offending statement itself, not the line of the enclosing function's signature — review views place the objection on exactly this line. " +
         "The tool does not check that the line still holds the cited code, because line numbers drift."),
+    environmentFamily: environmentFamilyField
+        .optional()
+        .describe("Only for category 'config'. Not allowed for the other categories. " +
+        "When this config issue is a test-environment gap, name the gap. Omit it for a config issue that is not a test-environment gap. " +
+        ENVIRONMENT_FAMILY_MEANINGS),
 })
     .superRefine((issue, ctx) => {
+    // SKYR-4555: the family names a test-environment gap, which only a config
+    // entry reports. On a bug, lint or type entry it would put the issue in two
+    // places at once.
+    if (issue.environmentFamily !== undefined &&
+        issue.category !== IssueFoundCategory.Config) {
+        ctx.addIssue({
+            code: "custom",
+            path: ["environmentFamily"],
+            message: `environmentFamily is allowed only when category is 'config'. It names the test-environment gap behind a config issue. ` +
+                `Remove environmentFamily from this '${issue.category}' entry, or set category to 'config' if the issue is a test-environment gap.`,
+        });
+    }
     // A `bug` entry asserts that product code is wrong, so it has to name the code:
     // without a file the entry is prose a reviewer cannot act on, and nothing
     // downstream can place it in the diff. Only the file is required — a config file
@@ -408,7 +521,7 @@ const issueFoundSchema = z
     if (issue.sourceFile !== undefined)
         return;
     ctx.addIssue({
-        code: z.ZodIssueCode.custom,
+        code: "custom",
         path: ["sourceFile"],
         message: `sourceFile is required when category is 'bug' — a bug entry must name the code it is about. ` +
             `Open the file, then set sourceFile (path relative to the repository root). Also set sourceSymbol (the function or identifier inside it) when the file has one, and sourceLine (the line of the offending statement) when you have it — both are optional. ` +
@@ -431,11 +544,11 @@ const scenarioStepSchema = z.object({
         .optional()
         .describe("Expected HTTP status code, e.g. 200, 201, 404"),
     requestBody: z
-        .record(z.any())
+        .record(z.string(), z.any())
         .optional()
         .describe("Example request body with realistic field values"),
     responseBody: z
-        .union([z.record(z.any()), z.array(z.any())])
+        .union([z.record(z.string(), z.any()), z.array(z.any())])
         .optional()
         .describe("Key response fields to verify, e.g. { id: 'number', name: 'string', in_stock: 'boolean?' }. An array for a collection/list endpoint that returns a JSON array."),
 });
@@ -445,7 +558,7 @@ export const additionalRecommendationSchema = z.preprocess(stripNullGroundingFie
         .string()
         .describe("Human-readable kebab-case identifier, e.g. 'integration-products-orders-workflow' or 'e2e-checkout-flow'. Format: '<testType>-<scenario-slug>'. Must be unique within the report."),
     testType: z
-        .nativeEnum(TestType)
+        .enum(TestType)
         .describe("Type of test. Do not include priority or other metadata in this field."),
     category: z
         .enum(SCENARIO_CATEGORIES)
@@ -499,7 +612,7 @@ export const additionalRecommendationSchema = z.preprocess(stripNullGroundingFie
     .superRefine((rec, ctx) => {
     if (rec.testType === TestType.CONTRACT && !rec.primaryEndpoint) {
         ctx.addIssue({
-            code: z.ZodIssueCode.custom,
+            code: "custom",
             path: ["primaryEndpoint"],
             message: "primaryEndpoint is required for contract tests — provide the HTTP verb and path, e.g. 'PATCH /api/v1/orders/{order_id}'.",
         });
@@ -509,7 +622,7 @@ export const additionalRecommendationSchema = z.preprocess(stripNullGroundingFie
         for (const field of ["targetElements", "pageContext"]) {
             if (rec[field] !== undefined) {
                 ctx.addIssue({
-                    code: z.ZodIssueCode.custom,
+                    code: "custom",
                     path: [field],
                     message: `${field} is only valid for testType: 'ui'`,
                 });
@@ -519,7 +632,7 @@ export const additionalRecommendationSchema = z.preprocess(stripNullGroundingFie
     else {
         if (rec.targetElements === undefined) {
             ctx.addIssue({
-                code: z.ZodIssueCode.custom,
+                code: "custom",
                 path: ["targetElements"],
                 message: "targetElements is required for testType: 'ui'. Use null when no capture covers the screen.",
             });
@@ -527,7 +640,7 @@ export const additionalRecommendationSchema = z.preprocess(stripNullGroundingFie
         if (Array.isArray(rec.targetElements) &&
             rec.pageContext === undefined) {
             ctx.addIssue({
-                code: z.ZodIssueCode.custom,
+                code: "custom",
                 path: ["pageContext"],
                 message: "pageContext is required for testType: 'ui' when targetElements is an array.",
             });
@@ -537,7 +650,7 @@ export const additionalRecommendationSchema = z.preprocess(stripNullGroundingFie
     }
 }));
 const testMaintenanceSchema = z.object({
-    testType: z.nativeEnum(TestType).describe("Type of test."),
+    testType: z.enum(TestType).describe("Type of test."),
     endpoint: z
         .string()
         .describe("For contract and integration tests, the HTTP verb and path, e.g. 'GET /api/v1/products'. For ui and e2e tests, the page path the test opens, e.g. '/notifications'."),
@@ -548,7 +661,7 @@ const testMaintenanceSchema = z.object({
     })
         .describe("Absolute path of the test file that was maintained, e.g. '/repo/tests/products_smoke_test.py' — the same path you passed to skyramp_execute_test's testFile param. Consumers should basename this for display."),
     action: z
-        .nativeEnum(DriftAction)
+        .enum(DriftAction)
         .describe("The drift action assigned to this test during maintenance triage."),
     description: z.string().describe("What was changed and why"),
     beforeDetails: z
@@ -576,8 +689,8 @@ const testMaintenanceSchema = z.object({
     // "probably the primary" cannot tell a single-repo row from a mis-stamped one.
     repository: z.string(),
     // Server-populated from stateFile execution records — never supplied by the LLM.
-    beforeStatus: z.nativeEnum(TestExecutionStatus),
-    afterStatus: z.nativeEnum(TestExecutionStatus),
+    beforeStatus: z.enum(TestExecutionStatus),
+    afterStatus: z.enum(TestExecutionStatus),
 });
 // LLM input contract: status fields are server-populated from stateFile, and
 // `defects` is attached from the per-file detail the agent drafts, not from the
@@ -1361,7 +1474,7 @@ export function registerSubmitReportTool(server) {
             "This is the ONLY way to submit the report — do NOT write the report to a file manually. " +
             "If a check objects to what you reported, the objection names an id: fix what you can, and answer the rest through `answers` on a second call. " +
             "On an ACCEPTED call this tool also edits the files this run changed: a comment addressed to a Skyramp gate — `reuse-verify: allow …` or `// kept inline: …` — is removed from each of them, because it speaks to a gate and not to the customer who owns the file. Nothing else in those files is touched. A REFUSED call leaves every source file exactly as it found it; it still records its own run state, such as the refusal count and the objections it raised.",
-        inputSchema: {
+        inputSchema: z.object({
             businessCaseAnalysis: z
                 .string()
                 .min(1, "businessCaseAnalysis is required — describe what the PR does, even for no-test runs (config-only, non-application, Testbot onboarding). It is never optional and must not be empty.")
@@ -1412,12 +1525,11 @@ export function registerSubmitReportTool(server) {
                 answer: z
                     .string()
                     .describe("Why the objection does not apply, or what you decided instead. A blank or whitespace-only answer closes nothing and the objection stays open."),
-                blocker: blockerSchema.optional().describe(BLOCKER_DESCRIPTION),
             })
                 .strict())
                 .optional()
                 .default([])
-                .describe("One answer per check that objected to your previous call and that you do not intend to fix. An answered objection is published with your answer beside it and no longer counted open; an unanswered one is published on its own. An objection about a bug with no test is the exception: it closes only when the answer also sets `blocker`, and a sentence on its own leaves it open and published. Answers persist across calls, so an id answered once stays answered — with the blocker it carried. There is no call limit."),
+                .describe("One answer per check that objected to your previous call and that you do not intend to fix. An answered objection is published with your answer beside it and no longer counted open; an unanswered one is published on its own. Answers persist across calls, so an id answered once stays answered. There is no call limit."),
             stateFile: z
                 .string()
                 .optional()
@@ -1426,7 +1538,7 @@ export function registerSubmitReportTool(server) {
                 "another run's execution results, and the call is refused. " +
                 "The report is written beside it, and it holds the execution results, the " +
                 "page captures and the videos this report needs."),
-        },
+        }),
         _meta: {
             keywords: ["report", "summary", "testbot", "submit"],
         },
@@ -2513,14 +2625,6 @@ export function registerSubmitReportTool(server) {
                 typeof entry?.answer === "string" &&
                 entry.answer.trim().length > 0);
             const carriedByKey = new Map(carried.map((entry) => [answerKey(entry.objection), entry]));
-            // Probed here for the same reason the plan tool probes before building its
-            // context: the checks below are synchronous. Both this call's blockers and
-            // the carried ones, so a carried close is re-judged rather than trusted.
-            const blockerChecks = await probeBlockers([
-                ...blockersIn(acknowledged),
-                ...blockersIn(carried),
-            ]);
-            const judge = (blocker) => verifyBlocker(blocker, blockerChecks);
             // What the LAST call raised. Without it there is nothing to notice a
             // fixed objection's absence against.
             const previouslyRaised = (Array.isArray(stateData.reportObjections?.raisedObjections)
@@ -2547,35 +2651,12 @@ export function registerSubmitReportTool(server) {
                 ...raised,
                 ...previouslyRaised.map((entry) => entry.objection),
             ];
-            const raisedKeys = new Set(raised.map(answerKey));
             const reportStageObjections = [
                 ...raised,
                 ...unknownAnswerObjections(answerable, acknowledged, {
                     scope: "this run raised",
                     suggestion: "Answer only the ids this tool has listed on this run. A plan-time objection is answered through skyramp_register_test_plan's own answers, not here.",
                 }),
-                // The same refusal plan time raises: a sentence sent for an objection that
-                // takes a blocker leaves that objection open, and this says so rather than
-                // letting the answer pass in silence. Against this call's answers only,
-                // like `registerPlan` — a carried one that closed nothing simply stays
-                // carried, and its objection stays published open.
-                ...refusedAnswerObjections(raised, acknowledged, judge, (id) => 
-                // The same rule the loop below applies to a carried answer. Without
-                // it a second, weaker answer for an objection an earlier call already
-                // closed raised `answers:refused:` beside the close this call still
-                // publishes as answered. Matched by id AND message, as `carriedByKey`
-                // is: one id can name a different bug on a later call, and a close
-                // carried for the first must not silence a refusal for the second.
-                carried.some((entry) => {
-                    if (entry.objection?.objectionId !== id)
-                        return false;
-                    if (!raisedKeys.has(answerKey(entry.objection)))
-                        return false;
-                    const blocker = hasBlocker(entry.blocker)
-                        ? entry.blocker
-                        : undefined;
-                    return (!needsBlocker(id) || (!!blocker && judge(blocker) !== "refuted"));
-                })),
             ];
             const raisedNow = new Set(reportStageObjections.map(answerKey));
             // A call with no execution result gives these checks nothing to judge, so
@@ -2605,57 +2686,12 @@ export function registerSubmitReportTool(server) {
             const answeredReportObjections = [];
             const openReportObjections = [];
             for (const objection of reportStageObjections) {
-                // `closingAnswerFor` applies the blocker rule to THIS call's answers. A
-                // carried answer was accepted on an earlier call and is re-checked against
-                // the same rule here, because the objection it closed is recomputed every
-                // call. The answer and the blocker come from the ONE entry that closed it:
-                // searching for the blocker apart from the answer published the first
-                // blocker sent, refuted or belonging to no answer at all.
-                const wantsBlocker = needsBlocker(objection.objectionId);
-                const closingNow = closingAnswerFor(objection, acknowledged, judge);
-                const blockerNow = hasBlocker(closingNow?.blocker)
-                    ? closingNow.blocker
-                    : undefined;
-                const carriedEntry = carriedByKey.get(answerKey(objection));
-                const carriedBlocker = hasBlocker(carriedEntry?.blocker)
-                    ? carriedEntry?.blocker
-                    : undefined;
-                const answerNow = closingNow?.answer;
-                // A carried answer meets the same rule as the call that sent it: a
-                // blocker, and one this call's probe does not refute.
-                const carriedCloses = carriedEntry &&
-                    (!wantsBlocker ||
-                        (!!carriedBlocker && judge(carriedBlocker) !== "refuted"));
-                const answer = answerNow ?? (carriedCloses ? carriedEntry?.answer : undefined);
-                if (!answer) {
+                const answer = answerFor(objection, acknowledged) ??
+                    carriedByKey.get(answerKey(objection))?.answer;
+                if (answer)
+                    answeredReportObjections.push({ objection, answer });
+                else
                     openReportObjections.push(objection);
-                    // A carried answer that closed this objection on an earlier call and
-                    // does not now is the probe changing its mind — the service came back
-                    // up. Without this the objection simply reappears with no reason.
-                    if (carriedEntry &&
-                        carriedBlocker &&
-                        judge(carriedBlocker) === "refuted") {
-                        openReportObjections.push({
-                            objectionId: `${REFUSED_ANSWER_PREFIX}${objection.objectionId}`,
-                            verifier: "answers",
-                            message: "The server checked the blocker this answer names and found the opposite, so the objection stays open.",
-                            evidence: `answered "${objection.objectionId}" on an earlier call with \`${carriedBlocker.kind}\`; ${refutedEvidence(carriedBlocker)}`,
-                            suggestion: "Write the test: what you named as the blocker is not in the way. If something else is, send the blocker that names it — the server checks each kind against the run it can see.",
-                        });
-                    }
-                    continue;
-                }
-                // A refuted blocker is not published, as at plan time: on an objection any
-                // answer closes, the sentence closes it and the untrue claim is dropped.
-                const sent = answerNow ? blockerNow : carriedBlocker;
-                const blocker = sent && judge(sent) !== "refuted" ? sent : undefined;
-                answeredReportObjections.push({
-                    objection,
-                    answer,
-                    ...(blocker
-                        ? { blocker, blockerVerified: judge(blocker) === "verified" }
-                        : {}),
-                });
             }
             objections = [
                 ...stillOpen.map((objection) => ({
@@ -2668,16 +2704,6 @@ export function registerSubmitReportTool(server) {
                     ...entry.objection,
                     stage: "plan",
                     answer: typeof entry.answer === "string" ? entry.answer : "",
-                    // A plan-time close that a blocker carried says so here, exactly as
-                    // the post-execution closes below do. Without this a probed close
-                    // was published as an ordinary sentence and the report lost the one
-                    // fact that distinguishes them.
-                    ...(hasBlocker(entry.blocker)
-                        ? {
-                            unverifiedBlocker: blockerText(entry.blocker),
-                            blockerVerified: !!entry.blockerVerified,
-                        }
-                        : {}),
                 })),
                 ...unverified
                     .filter((entry) => entry?.objection)
@@ -2685,29 +2711,16 @@ export function registerSubmitReportTool(server) {
                     ...entry.objection,
                     stage: "plan",
                     answer: typeof entry.answer === "string" ? entry.answer : "",
-                    // `unverifiedCloses[].blocker` is a `PlanBlocker`; reading it as a
-                    // string published an empty one on every plan-time close.
-                    ...(hasBlocker(entry.blocker)
-                        ? { unverifiedBlocker: blockerText(entry.blocker) }
-                        : {}),
+                    unverifiedBlocker: typeof entry.blocker === "string" ? entry.blocker : "",
                 })),
                 ...openReportObjections.map((objection) => ({
                     ...objection,
                     stage: "post-execution",
                 })),
-                // The blocker is published beside the answer exactly as the plan-time
-                // closes above are, and `blockerVerified` says whether a probe backed it
-                // or the server could only take it as given.
-                ...answeredReportObjections.map(({ objection, answer, blocker, blockerVerified }) => ({
+                ...answeredReportObjections.map(({ objection, answer }) => ({
                     ...objection,
                     stage: "post-execution",
                     answer,
-                    ...(blocker
-                        ? {
-                            unverifiedBlocker: blockerText(blocker),
-                            blockerVerified: !!blockerVerified,
-                        }
-                        : {}),
                 })),
                 ...resolvedObjections.map(({ objection, resolvedBy }) => ({
                     ...objection,
@@ -2852,7 +2865,10 @@ export function registerSubmitReportTool(server) {
             // these objects and don't expect it).
             testResults: params.testResults.map((row) => {
                 const { testFilePath: _tfp, ...wire } = attachVideoPath(row, stateData.executionVideos);
-                return wire;
+                // SKYR-4555: the basename is the join key to newTestsCreated and
+                // testMaintenance (consumers compare it with the basename of their
+                // fileName, which may be a relative path); the path stays internal.
+                return { ...wire, fileName: path.basename(row.testFilePath) };
             }),
             issuesFound,
             executionCoverage,

package/build/tools/test-management/actionsTool.d.ts

@@ -1,4 +1,4 @@
-import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
+import { McpServer } from "@modelcontextprotocol/server";
 /**
  * Compute a suggested new filename when an endpoint is renamed.
  */

package/build/tools/test-management/actionsTool.js

@@ -69,19 +69,19 @@ const recommendationSchema = z.object({
     })
         .describe("Absolute path to the page object (POM) file to edit, when the spec delegates selectors to a page object. When set, skyramp_actions edits this file instead of testFilePath. testFilePath (the spec) is still used for catalog validation and the maintenance report."),
     action: z
-        .nativeEnum(DriftAction)
+        .enum(DriftAction)
         .refine((a) => a !== DriftAction.Ignore, {
         message: "Omit IGNORE entries entirely — a discovered test you don't recommend is automatically treated as needing no action.",
     })
         .describe("The maintenance action for this test (UPDATE, REGENERATE, DELETE, or VERIFY). Do not send IGNORE — omit no-action tests from `recommendations` entirely; anything you leave out is recorded as needing no action."),
     testType: z
-        .nativeEnum(TestType)
+        .enum(TestType)
         .describe("Type of this test — carried through to the maintenance report."),
     endpoint: z
         .string()
         .describe("HTTP verb and path this test covers, e.g. 'GET /api/v1/products'."),
     priority: z
-        .nativeEnum(RecommendationPriority)
+        .enum(RecommendationPriority)
         .optional()
         .describe("Update priority"),
     rationale: z
@@ -90,7 +90,7 @@ const recommendationSchema = z.object({
         .min(1)
         .describe("1-2 sentence explanation of why this action is needed — becomes the maintenance report's description, so it must not be blank."),
     estimatedWork: z
-        .nativeEnum(EstimatedWork)
+        .enum(EstimatedWork)
         .optional()
         .describe("Estimated effort to apply the update"),
     updateInstructions: z
@@ -169,15 +169,15 @@ export function registerActionsTool(server) {
         description: `Execute test maintenance actions — final step of the unified Test Health Analysis Flow. Call this once per run even when no tests need action (pass \`recommendations: []\`) — it is where the maintenance verdicts are persisted before the run moves on. Supply only the tests that need an action; any discovered test you omit from \`recommendations\` is treated as needing no action and recorded automatically, so you never enumerate the no-action tests yourself.`,
         // SKYR-4194: the payload is returned via structuredContent, which requires a
         // declared outputSchema — see dualChannelResult.
-        outputSchema: {
+        outputSchema: z.object({
             summary: z
                 .string()
                 .describe("Human-readable report of the maintenance actions to take."),
             instructions: z
                 .string()
                 .describe("JSON instructions for carrying out those actions — file lists, per-file edit context, and the update/regenerate/delete strategies. Read and follow this field."),
-        },
-        inputSchema: actionsSchema,
+        }),
+        inputSchema: z.object(actionsSchema),
     }, async (args) => {
         let errorResult;
         try {