@skyramp/mcp 0.2.6 → 0.2.8-rc.1

package/build/prompts/test-recommendation/test-recommendation-prompt.js

@@ -93,10 +93,10 @@ ${isUIOnlyPR ? `\n**UI-only PR** — no backend changes. UI and E2E tests are mo
 Output should be concise and immediately actionable.`
         : `You are in **Repo mode**. Comprehensive test strategy across all endpoints.`;
     // ── UI rec authoring rules ──
-    // Anchors every UI recommendation regardless of whether the agent ran
-    // skyramp_ui_analyze_changes / browser_blueprint before this call. When the
-    // agent has prior captures in its own tool-result history, those serve as
-    // grounding; when it doesn't, recommendations fall back to source-grounded
+    // Anchors every UI recommendation regardless of whether the agent captured
+    // browser_blueprints (from the UI Blueprint Capture section this tool returns).
+    // When the agent has prior captures in its own tool-result history, those serve
+    // as grounding; when it doesn't, recommendations fall back to source-grounded
     // prose. Either way, inventing elements / leaking jargon / duplicating UI
     // tests are wrong, so the rules fire unconditionally.
     const uiRecRulesSection = `

package/build/prompts/test-recommendation/test-recommendation-prompt.test.d.ts

@@ -0,0 +1 @@
+export {};

package/build/prompts/test-recommendation/test-recommendation-prompt.test.js

@@ -1,10 +1,11 @@
-jest.mock("@skyramp/skyramp", () => ({
+import { jest } from "@jest/globals";
+import { TestType } from "../../types/TestTypes.js";
+jest.unstable_mockModule("@skyramp/skyramp", () => ({
     WorkspaceConfigManager: { create: jest.fn() },
 }));
-import { TestType } from "../../types/TestTypes.js";
-import { buildRecommendationPrompt, buildExternalCoverageSet, externalDedupKey } from "./test-recommendation-prompt.js";
-import { buildExecutionPlan } from "./diffExecutionPlan.js";
-import { PATH_PARAM_UUID_GUIDANCE, MAX_TESTS_TO_GENERATE, buildTestQualityCriteria, buildArchitectPreamble, buildContextFetchingGuidance, buildReasoningProtocol, buildFewShotExamples, buildVerificationChecklist, } from "./recommendationSections.js";
+const { buildRecommendationPrompt, buildExternalCoverageSet, externalDedupKey } = await import("./test-recommendation-prompt.js");
+const { buildExecutionPlan } = await import("./diffExecutionPlan.js");
+const { PATH_PARAM_UUID_GUIDANCE, MAX_TESTS_TO_GENERATE, buildTestQualityCriteria, buildArchitectPreamble, buildContextFetchingGuidance, buildReasoningProtocol, buildFewShotExamples, buildVerificationChecklist, } = await import("./recommendationSections.js");
 import { AnalysisScope } from "../../types/RepositoryAnalysis.js";
 // ---------------------------------------------------------------------------
 // Minimal fixtures
@@ -909,6 +910,97 @@ describe("buildRecommendationPrompt — GENERATE slot allocation", () => {
         expect(additionalBlock).toContain("DELETE /api/gadgets/{id} → 401");
     });
 });
+describe("buildExecutionPlan — even-with-spillover test-type distribution", () => {
+    // Backend-only PR (hasFrontendChanges=false) so every GENERATE slot is a backend
+    // scenario drawn from `scored` — this is where contract-vs-integration distribution
+    // applies. baseUrl/auth args mirror the existing buildExecutionPlan tests above.
+    const contract = (name, path, priority = "HIGH") => ({
+        scenario: minimalScenario({
+            scenarioName: name,
+            category: "crud",
+            testType: TestType.CONTRACT,
+            steps: [{ order: 1, method: "POST", path, description: `POST ${path}`, interactionType: "success", expectedStatusCode: 201 }],
+        }),
+        priority: priority,
+        novelty: "new",
+    });
+    const integration = (name, p1, p2, priority = "HIGH") => ({
+        scenario: minimalScenario({
+            scenarioName: name,
+            category: "data_integrity",
+            testType: TestType.INTEGRATION,
+            steps: [
+                { order: 1, method: "POST", path: p1, description: `POST ${p1}`, interactionType: "success", expectedStatusCode: 201 },
+                { order: 2, method: "PATCH", path: p2, description: `PATCH ${p2}`, interactionType: "success", expectedStatusCode: 200 },
+            ],
+        }),
+        priority: priority,
+        novelty: "new",
+    });
+    const plan = (items, maxGen, topN) => buildExecutionPlan(items, maxGen, topN, "http://localhost:3000", "Authorization", ", authScheme: \"Bearer\"", "", "seed", items.length, false);
+    it("distributes GENERATE slots across contract AND integration when budget < candidates", () => {
+        // 4 contract out-rank 4 integration; pure top-3 rank slice would be all contract.
+        const items = [
+            contract("c1", "/api/a"), contract("c2", "/api/b"), contract("c3", "/api/c"), contract("c4", "/api/d"),
+            integration("i1", "/api/x", "/api/x/{id}"), integration("i2", "/api/y", "/api/y/{id}"),
+        ];
+        const prompt = plan(items, 3, 6);
+        const generated = prompt.slice(0, prompt.indexOf("[ADDITIONAL]") >= 0 ? prompt.indexOf("[ADDITIONAL]") : prompt.length);
+        expect(generated).toMatch(/GENERATE\*\* \| contract/);
+        expect(generated).toMatch(/GENERATE\*\* \| integration/);
+    });
+    it("single backend type → unchanged top-N rank slice (no regression)", () => {
+        const items = [contract("c1", "/api/a"), contract("c2", "/api/b"), contract("c3", "/api/c"), contract("c4", "/api/d")];
+        const prompt = plan(items, 3, 6);
+        const generated = prompt.slice(0, prompt.indexOf("[ADDITIONAL]") >= 0 ? prompt.indexOf("[ADDITIONAL]") : prompt.length);
+        // Top 3 contract by rank are generated; the 4th is not.
+        expect(generated).toContain("POST /api/a → 201");
+        expect(generated).toContain("POST /api/b → 201");
+        expect(generated).toContain("POST /api/c → 201");
+        expect(generated).not.toContain("POST /api/d → 201");
+        expect(generated).not.toMatch(/GENERATE\*\* \| integration/);
+    });
+    it("keeps a CRITICAL item in GENERATE even when lower-ranked than other-type candidates", () => {
+        // A CRITICAL integration sits after several HIGH contracts in rank order.
+        const items = [
+            contract("c1", "/api/a"), contract("c2", "/api/b"), contract("c3", "/api/c"),
+            integration("crit", "/api/orders", "/api/orders/{id}", "CRITICAL"),
+        ];
+        const prompt = plan(items, 3, 6);
+        const generated = prompt.slice(0, prompt.indexOf("[ADDITIONAL]") >= 0 ? prompt.indexOf("[ADDITIONAL]") : prompt.length);
+        // The CRITICAL integration must occupy a GENERATE slot (protected-first).
+        expect(generated).toContain("PATCH /api/orders/{id}");
+    });
+    it("spills freed slots to the remaining type when one type runs dry", () => {
+        // 1 integration + 4 contract, budget 3: round 1 = 1 integration + 1 contract,
+        // integration bucket now empty → remaining 1 slot spills to contract → 2 contract total.
+        const items = [
+            integration("i1", "/api/x", "/api/x/{id}"),
+            contract("c1", "/api/a"), contract("c2", "/api/b"), contract("c3", "/api/c"), contract("c4", "/api/d"),
+        ];
+        const prompt = plan(items, 3, 6);
+        const generated = prompt.slice(0, prompt.indexOf("[ADDITIONAL]") >= 0 ? prompt.indexOf("[ADDITIONAL]") : prompt.length);
+        const contractSlots = (generated.match(/GENERATE\*\* \| contract/g) || []).length;
+        const integrationSlots = (generated.match(/GENERATE\*\* \| integration/g) || []).length;
+        expect(integrationSlots).toBe(1);
+        expect(contractSlots).toBe(2);
+    });
+    it("never lists a GENERATE item again under ADDITIONAL (set-difference recompute)", () => {
+        const items = [
+            contract("c1", "/api/a"), contract("c2", "/api/b"), contract("c3", "/api/c"), contract("c4", "/api/d"),
+            integration("i1", "/api/x", "/api/x/{id}"), integration("i2", "/api/y", "/api/y/{id}"),
+        ];
+        const prompt = plan(items, 3, 6);
+        const splitIdx = prompt.indexOf("[ADDITIONAL]");
+        expect(splitIdx).toBeGreaterThan(0);
+        const generated = prompt.slice(0, splitIdx);
+        const additional = prompt.slice(splitIdx);
+        // Each generated endpoint must not reappear in the ADDITIONAL section.
+        for (const m of generated.matchAll(/(POST|PATCH) (\/api\/[^\s]+)/g)) {
+            expect(additional).not.toContain(`${m[1]} ${m[2]}`);
+        }
+    });
+});
 // ---------------------------------------------------------------------------
 // Tests — buildTestQualityCriteria contract-test guidance (regression guard)
 // ---------------------------------------------------------------------------
@@ -1669,100 +1761,6 @@ describe("externalDedupKey", () => {
         expect(externalDedupKey(scenario)).toBe("POST::orders::contract");
     });
 });
-describe("UI grounding guidance (Phase C D-1.a)", () => {
-    const editOrderScenario = minimalScenario({
-        scenarioName: 'edit-order',
-        description: 'edit order',
-        category: 'new_endpoint',
-        priority: 'medium',
-        steps: [
-            { order: 1, method: 'PATCH', path: '/api/v1/orders/{id}', description: 'edit', interactionType: 'success', expectedStatusCode: 200 },
-        ],
-        chainingKeys: [],
-    });
-    const uiGroundingBusinessContext = {
-        mainPurpose: "Test API", userFlows: [], dataFlows: [], integrationPatterns: [],
-        draftedScenarios: [editOrderScenario],
-    };
-    const uiOnlyAnalysis = () => minimalAnalysis({
-        branchDiffContext: {
-            currentBranch: "feature", baseBranch: "main",
-            changedFiles: ['frontend/src/components/EditOrder.tsx'],
-            newEndpoints: [],
-            modifiedEndpoints: [],
-            affectedServices: [],
-        },
-        businessContext: uiGroundingBusinessContext,
-    });
-    const mixedAnalysis = () => minimalAnalysis({
-        branchDiffContext: {
-            currentBranch: "feature", baseBranch: "main",
-            changedFiles: ['frontend/src/components/EditOrder.tsx', 'backend/orders.py'],
-            newEndpoints: [{
-                    path: '/api/v1/orders/{id}',
-                    methods: [{ method: 'PATCH', sourceFile: 'backend/orders.py', interactionCount: 1 }],
-                }],
-            modifiedEndpoints: [],
-            affectedServices: [],
-        },
-        businessContext: uiGroundingBusinessContext,
-    });
-    const backendOnlyAnalysis = () => minimalAnalysis({
-        branchDiffContext: {
-            currentBranch: "feature", baseBranch: "main",
-            changedFiles: ['backend/orders.py'],
-            newEndpoints: [{
-                    path: '/api/v1/orders',
-                    methods: [{ method: 'POST', sourceFile: 'backend/orders.py', interactionCount: 1 }],
-                }],
-            modifiedEndpoints: [],
-            affectedServices: [],
-        },
-        businessContext: uiGroundingBusinessContext,
-    });
-    it("emits UI grounding block on UI-only PRs", () => {
-        const out = buildRecommendationPrompt(uiOnlyAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).toContain("UI recommendation grounding");
-        expect(out).toContain("role=<role>");
-        expect(out).toContain("accessibleName=");
-    });
-    it("emits UI grounding block on mixed PRs", () => {
-        const out = buildRecommendationPrompt(mixedAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).toContain("UI recommendation grounding");
-        expect(out).toContain("contextText=");
-    });
-    it("does not emit UI grounding block on backend-only PRs", () => {
-        const out = buildRecommendationPrompt(backendOnlyAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).not.toContain("UI recommendation grounding");
-    });
-    it("carves out non-UI entries from the grounding format", () => {
-        const out = buildRecommendationPrompt(mixedAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).toMatch(/Contract, integration, e2e.*existing conventions/);
-    });
-    it("includes contextText example for repeating elements", () => {
-        const out = buildRecommendationPrompt(uiOnlyAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).toContain('contextText=["customer@example.com"');
-    });
-    // Phase C D-1.a/7b: directive instruction strengthening + Validates coverage
-    // + marked fallback. Asserts the post-7b language is present.
-    it("uses MUST directive for tuple presence in reasoning field", () => {
-        const out = buildRecommendationPrompt(uiOnlyAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).toContain("MUST contain at least three of");
-        expect(out).toContain("not valid output for UI test types");
-    });
-    it("instructs the agent to ground the Validates line for UI entries", () => {
-        const out = buildRecommendationPrompt(uiOnlyAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).toContain("Validates line — applies to");
-        expect(out).toContain("observable behavior the test verifies");
-        expect(out).toContain("structural facts");
-    });
-    it("requires a [no-blueprint-data] marker on source-grounded fallback entries", () => {
-        const out = buildRecommendationPrompt(uiOnlyAnalysis(), AnalysisScope.CurrentBranchDiff, 10);
-        expect(out).toContain("[no-blueprint-data]");
-        expect(out).toContain("issuesFound");
-        expect(out).toContain("Do NOT silently produce ungrounded reasoning");
-    });
-});
 // ---------------------------------------------------------------------------
 // Tests — UI recommendation authoring rules
 // ---------------------------------------------------------------------------

package/build/prompts/testbot/testbot-prompts.d.ts

@@ -0,0 +1,26 @@
+import { McpServer } from "@modelcontextprotocol/sdk/server/mcp.js";
+import { type Service } from "../../workspace/workspace.js";
+/** One related repository to analyze as shared context in multi-repo mode. */
+export interface RelatedRepository {
+    repositoryPath: string;
+    baseBranch?: string;
+    /** `owner/repo` slug, resolved by the action from the checkout's git remote.
+     *  Used verbatim as the `repository` argument/attribution for this repo so the
+     *  agent never has to infer it. */
+    repo: string;
+}
+/**
+ * Parse the JSON-encoded `relatedRepositories` argument passed via the testbot
+ * prompt/resource. Returns undefined for missing/blank input or malformed JSON so the
+ * caller cleanly falls back to single-repo behavior.
+ */
+export declare function parseRelatedRepositories(raw: string | undefined): RelatedRepository[] | undefined;
+export declare function getTestbotPrompt(prTitle: string, prDescription: string, summaryOutputFile: string, repositoryPath: string, baseBranch?: string, maxRecommendations?: number, maxGenerate?: number, _maxCritical?: number, // Reserved — accepted for API compat but not yet wired into prompt
+prNumber?: number, userPrompt?: string, services?: Service[], uiCredentials?: string, testsRepoDir?: string, relatedRepositories?: RelatedRepository[], primaryRepo?: string): string;
+/**
+ * Read services from .skyramp/workspace.yml. Returns empty array if
+ * the workspace file doesn't exist or can't be parsed.
+ */
+export declare function readWorkspaceServices(repositoryPath: string): Promise<Service[]>;
+export declare function buildWorkspaceRecoveryPrefix(repositoryPath: string): string;
+export declare function registerTestbotPrompt(server: McpServer): void;