@skyramp/mcp 0.2.4 → 0.2.5-rc.1

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

@@ -358,15 +358,9 @@ export function registerAnalyzeChangesTool(server) {
             idempotentHint: false,
             openWorldHint: true, // may fetch PR comments from GitHub
         },
-        description: `Scan repository API endpoints and discover existing tests — first step of the unified Test Health Analysis Flow.
-
-Combines API endpoint scanning, branch diff computation, and test discovery into a single state file consumed by \`skyramp_analyze_test_health\` and \`skyramp_actions\`.
-
-**Output:** stateFile path + ranked test recommendations + enrichment instructions for calling \`skyramp_recommend_tests\`.`,
-        // TODO: Define outputSchema here instead of embedding structured output format in the
-        // description string — per Archit's review comment. outputSchema reduces token usage
-        // by letting the MCP client understand the response shape structurally rather than
-        // through natural language in the description.
+        description: `Scan repository API endpoints and discover existing tests — first step of the unified Test Health Analysis Flow. Returns a stateFile path and ranked test recommendations. Pass stateFile to skyramp_analyze_test_health and skyramp_actions.`,
+        // TODO: Replace description-embedded output format with outputSchema — structural
+        // output schema reduces token usage vs natural language in description.
         inputSchema: analyzeChangesInputSchema,
     }, async (params, extra) => {
         let errorResult;
@@ -538,6 +532,14 @@ Combines API endpoint scanning, branch diff computation, and test discovery into
             }
             await sendProgress(50, 100, "Discovering existing tests...");
             // ── Step 3: Discover existing tests ──
+            // Always scan from repositoryPath so tests in any subdirectory (e.g.
+            // apps/api/v2/src/modules/) are found regardless of workspace.yml
+            // testDirectory. In PR mode, partitionByRelevance already filters the
+            // results to files relevant to the changed endpoints — no flooding.
+            // testDirectory only controls where generation tools write new files.
+            // testsRepoDir is a cross-repo path override — honour it when set.
+            // Otherwise always scan the full repo root.
+            const testDir = params.testsRepoDir ?? undefined;
             // Compute changedResources from classified endpoints for test discovery filtering.
             // undefined    → full-repo mode (no diff context)
             // []           → PR mode, no endpoints found → skip external tests
@@ -552,21 +554,44 @@ Combines API endpoint scanning, branch diff computation, and test discovery into
                     ...classifiedEndpoints.newEndpoints,
                     ...classifiedEndpoints.removedEndpoints,
                 ];
+                const FRAMEWORK_SUFFIX_RE = /\.(service|controller|transformer|handler|middleware|resolver|repository|module|guard|interceptor|pipe|filter|decorator|input|output|dto|schema)$/i;
                 if (allClassified.length > 0) {
                     // Scanned endpoints always have full paths — extractResourceFromPath
                     // never returns "unknown" for properly resolved paths.
+                    // Exception: NestJS versioned controllers register relative paths (e.g. "GET /")
+                    // which resolve to "unknown". Fall through to file-path extraction in that case.
                     const resolved = allClassified
                         .map((ep) => extractResourceFromPath(ep.path))
                         .filter((r, i, arr) => r !== "unknown" && arr.indexOf(r) === i);
-                    changedResources = resolved.length > 0 ? resolved : ["unknown"];
+                    if (resolved.length > 0) {
+                        changedResources = resolved;
+                    }
+                    else {
+                        // All endpoints resolved to "unknown" (e.g. NestJS relative paths) —
+                        // extract resource names from changed file paths instead.
+                        const candidateFiles = classifiedEndpoints.unmatchedFiles.length > 0
+                            ? classifiedEndpoints.unmatchedFiles
+                            : classifiedEndpoints.changedFiles ?? [];
+                        const fromFiles = candidateFiles
+                            .map((f) => extractResourceFromPath(f.replace(/\.[^./]+$/, "").replace(FRAMEWORK_SUFFIX_RE, "")))
+                            .filter((r, i, arr) => r !== "unknown" && arr.indexOf(r) === i);
+                        changedResources = fromFiles.length > 0 ? fromFiles : ["unknown"];
+                    }
                 }
                 else if (classifiedEndpoints.unmatchedFiles.length > 0) {
-                    // Changed files don't map to any endpoint (e.g. schema, model, or
-                    // migration changes near route files). Use ["unknown"] so external
-                    // tests get name-only entries — enough for the LLM to infer coverage
-                    // from filenames without flooding context with full extraction of
-                    // hundreds of irrelevant test files.
-                    changedResources = ["unknown"];
+                    // Changed files don't map to any endpoint (e.g. DTO, service, output
+                    // formatter, migration). Extract resource names from the file paths so
+                    // external tests whose names overlap with these resources are surfaced
+                    // for drift assessment.
+                    // Strip the file extension and common framework suffixes (.service,
+                    // .controller, .input, .output, etc.) before extracting so that
+                    // "event-types.service.ts" yields "event-types" rather than
+                    // "event-types.service", which would fail relevance scoring against
+                    // test files that contain "event" and "types" but not "service".
+                    const fromFiles = classifiedEndpoints.unmatchedFiles
+                        .map((f) => extractResourceFromPath(f.replace(/\.[^./]+$/, "").replace(FRAMEWORK_SUFFIX_RE, "")))
+                        .filter((r, i, arr) => r !== "unknown" && arr.indexOf(r) === i);
+                    changedResources = fromFiles.length > 0 ? fromFiles : ["unknown"];
                 }
                 else {
                     changedResources = [];
@@ -577,8 +602,7 @@ Combines API endpoint scanning, branch diff computation, and test discovery into
             try {
                 const testDiscoveryService = new TestDiscoveryService();
                 setTestsRepoDir(params.testsRepoDir);
-                const testScanPath = params.testsRepoDir ?? params.repositoryPath;
-                const discoveryResult = await testDiscoveryService.discoverTests(testScanPath, { changedResources });
+                const discoveryResult = await testDiscoveryService.discoverTests(testDir ?? params.repositoryPath, { changedResources });
                 existingTests = discoveryResult.tests.map((test) => ({
                     testFile: test.testFile,
                     testType: test.testType,
@@ -1188,6 +1212,7 @@ Combines API endpoint scanning, branch diff computation, and test discovery into
                     sessionId,
                     routerMountContext,
                     candidateRouteFiles,
+                    relevantExternalTestPaths,
                 },
             };
             // Clean up old state files (>24 hours) before creating new one

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

@@ -15,16 +15,7 @@ export function registerAnalyzeTestHealthTool(server) {
             idempotentHint: true,
             openWorldHint: false,
         },
-        description: `Generate drift and health assessment instructions for existing tests — second step of the unified Test Health Analysis Flow.
-
-**PREREQUISITE:** Call \`skyramp_analyze_changes\` first to get a stateFile.
-
-This tool reads existing tests, the branch diff, and scanned endpoints from the stateFile,
-then returns a structured prompt for the LLM to assess each test for drift and health.
-
-The LLM follows the returned prompt to assign drift details and actions (UPDATE / REGENERATE / VERIFY / DELETE / IGNORE) for each test, then calls \`skyramp_actions\`.
-
-(Optional) Execute tests using \`skyramp_execute_test\` with \`stateFile\` parameter before \`skyramp_actions\` to validate tests live.`,
+        description: `Generate drift assessment instructions for existing tests — second step of the unified Test Health Analysis Flow. Optionally execute tests with skyramp_execute_test before calling skyramp_actions to capture beforeStatus.`,
         inputSchema: {
             stateFile: z
                 .string()
@@ -45,19 +36,24 @@ The LLM follows the returned prompt to assign drift details and actions (UPDATE
             if (!stateData) {
                 return toolError(`State file is empty or invalid: ${args.stateFile}. Call skyramp_analyze_changes first to generate a valid state file.`);
             }
-            // Only Skyramp tests are candidates for drift analysis and maintenance actions.
-            // External (user-written) tests are used only for recommendation deduplication.
-            // Default source to Skyramp for backwards compat with state files created before the source field existed.
-            const existingTests = (stateData.existingTests || []).filter((t) => (t.source ?? TestSource.Skyramp) !== TestSource.External);
-            logger.info(`Loaded ${existingTests.length} existing Skyramp tests from state file (excluded external)`);
             if (!repositoryPath || typeof repositoryPath !== "string") {
                 return toolError(`repositoryPath not found in state file metadata. The state file was likely created by an older version — re-run skyramp_analyze_changes to regenerate it.`);
             }
-            const absoluteRepoPath = path.resolve(repositoryPath);
-            const scannedEndpoints = stateData.repositoryAnalysis?.skeletonEndpoints || [];
-            const routerMountContext = stateData.repositoryAnalysis?.routerMountContext;
-            const candidateRouteFiles = stateData.repositoryAnalysis?.candidateRouteFiles;
-            const diffFilePath = stateData.repositoryAnalysis?.diffFilePath;
+            // Skyramp tests: full drift analysis + all actions permitted.
+            // Relevant external tests (user-written, relevant to this PR's endpoints): drift analysis
+            //   + UPDATE only — REGENERATE and DELETE are report-only (enforced in skyramp_actions).
+            // Other external tests: excluded entirely (deduplication only, not analysed).
+            // relevantExternalTestPaths are stored relative to repositoryPath in the state file.
+            // Re-absolutize here so has() comparisons against t.testFile (absolute) work correctly.
+            const relevantExternalPaths = new Set((stateData.repositoryAnalysis?.relevantExternalTestPaths ?? []).map((p) => path.isAbsolute(p) ? p : path.resolve(repositoryPath, p)));
+            const existingTests = (stateData.existingTests || []).filter((t) => {
+                if ((t.source ?? TestSource.Skyramp) !== TestSource.External)
+                    return true;
+                return relevantExternalPaths.has(t.testFile);
+            });
+            const skyrampCount = existingTests.filter((t) => (t.source ?? TestSource.Skyramp) !== TestSource.External).length;
+            const externalCount = existingTests.length - skyrampCount;
+            logger.info(`Loaded ${skyrampCount} Skyramp + ${externalCount} relevant external tests from state file`);
             // Sweep stale diff files on this natural follow-up call so they don't accumulate.
             // Pass [] for stateTypes so only skyramp-diff-*.diff files are deleted — state files
             // (skyramp-analysis-*, skyramp-recommendation-*) must not be removed here because the
@@ -68,15 +64,7 @@ The LLM follows the returned prompt to assign drift details and actions (UPDATE
             catch (error) {
                 logger.warning(`Failed to cleanup old diff files: ${error.message}`);
             }
-            const promptText = buildDriftAnalysisPrompt({
-                existingTests,
-                scannedEndpoints,
-                repositoryPath: absoluteRepoPath,
-                stateFile: stateManager.getStatePath(),
-                routerMountContext,
-                candidateRouteFiles,
-                diffFilePath,
-            });
+            const promptText = buildDriftAnalysisPrompt(stateManager.getStatePath(), existingTests.map((t) => ({ testFile: t.testFile, source: t.source })));
             return {
                 structuredContent: { prompt: promptText },
                 content: [{ type: "text", text: "Drift analysis prompt generated. Follow the prompt field to assess each test." }],

package/build/utils/docker.test.js

@@ -54,7 +54,7 @@ describe("dockerImageExistsLocally", () => {
     });
 });
 describe("pullDockerImage", () => {
-    const IMAGE = "skyramp/executor:v1.3.26";
+    const IMAGE = "skyramp/executor:v1.3.27";
     beforeEach(() => jest.clearAllMocks());
     describe("on amd64 host", () => {
         const originalArch = process.arch;

package/build/utils/versions.js

@@ -1,3 +1,3 @@
-export const SKYRAMP_IMAGE_VERSION = "v1.3.26";
+export const SKYRAMP_IMAGE_VERSION = "v1.3.27";
 export const EXECUTOR_DOCKER_IMAGE = `skyramp/executor:${SKYRAMP_IMAGE_VERSION}`;
 export const WORKER_DOCKER_IMAGE = `skyramp/worker:${SKYRAMP_IMAGE_VERSION}`;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@skyramp/mcp",
-  "version": "0.2.4",
+  "version": "0.2.5-rc.1",
   "main": "build/index.js",
   "exports": {
     ".": "./build/index.js",
@@ -55,7 +55,7 @@
   "dependencies": {
     "@modelcontextprotocol/sdk": "^1.24.3",
     "@playwright/test": "^1.55.0",
-    "@skyramp/skyramp": "1.3.26",
+    "@skyramp/skyramp": "1.3.27",
     "dockerode": "^5.0.0",
     "fast-glob": "^3.3.3",
     "js-yaml": "^4.1.1",

package/build/prompts/test-maintenance/drift-analysis-prompt.test.js

@@ -1,116 +0,0 @@
-import { buildDriftAnalysisPrompt } from "./drift-analysis-prompt.js";
-import { buildDriftOutputChecklist } from "./driftAnalysisSections.js";
-describe("buildDriftOutputChecklist — final-step recommendations guidance", () => {
-    const STATE_FILE = "/tmp/skyramp-analysis-abc123.json";
-    it("non-inline mode includes recommendations and updateInstructions in final step", () => {
-        const checklist = buildDriftOutputChecklist(3, 0, false, STATE_FILE);
-        // Must instruct the LLM to pass recommendations to skyramp_actions
-        expect(checklist).toContain("recommendations");
-        // Must mention updateInstructions so the LLM knows to populate it
-        expect(checklist).toContain("updateInstructions");
-        // Must reference the stateFile path
-        expect(checklist).toContain(STATE_FILE);
-        // Must call skyramp_actions as the final action
-        expect(checklist).toContain("skyramp_actions");
-    });
-    it("non-inline mode does not contain JSON shape — schema is authoritative", () => {
-        const checklist = buildDriftOutputChecklist(3, 0, false, STATE_FILE);
-        // The JSON shape was moved to inputSchema — prompt must not duplicate it
-        expect(checklist).not.toContain('"testFile":');
-        expect(checklist).not.toContain('"action":');
-    });
-    it("inline mode does not reference skyramp_actions or stateFile", () => {
-        const checklist = buildDriftOutputChecklist(3, 0, true, STATE_FILE);
-        // Inline mode applies changes directly — no skyramp_actions call
-        expect(checklist).not.toContain("skyramp_actions");
-        expect(checklist).not.toContain(STATE_FILE);
-    });
-    it("full prompt (non-inline) includes recommendations guidance", () => {
-        const prompt = buildDriftAnalysisPrompt({
-            existingTests: [],
-            scannedEndpoints: [],
-            repositoryPath: "/repo",
-            stateFile: STATE_FILE,
-        });
-        expect(prompt).toContain("recommendations");
-        expect(prompt).toContain("updateInstructions");
-    });
-});
-describe("buildDriftAnalysisPrompt - inline mode", () => {
-    beforeEach(() => { process.env.SKYRAMP_FEATURE_TESTBOT = "1"; });
-    afterEach(() => { delete process.env.SKYRAMP_FEATURE_TESTBOT; });
-    function inlinePrompt() {
-        return buildDriftAnalysisPrompt({
-            existingTests: [],
-            scannedEndpoints: [],
-            repositoryPath: "/repo",
-            // stateFile omitted → inline mode
-        });
-    }
-    it("wraps inline rules in drift_analysis_rules XML tags", () => {
-        const prompt = inlinePrompt();
-        expect(prompt).toContain("<drift_analysis_rules>");
-        expect(prompt).toContain("</drift_analysis_rules>");
-    });
-    it("does not contain the persona statement", () => {
-        const prompt = inlinePrompt();
-        expect(prompt).not.toContain("You are acting as a Skyramp Integration Architect");
-    });
-    it("does not contain the standalone Test Health Analysis header", () => {
-        const prompt = inlinePrompt();
-        expect(prompt).not.toContain("# Test Health Analysis");
-    });
-    it("does not contain the skyramp_actions CTA (that belongs to standalone mode)", () => {
-        const prompt = inlinePrompt();
-        // Inline mode final step directs applying changes directly, not calling skyramp_actions
-        expect(prompt).not.toContain("call `skyramp_actions`");
-    });
-});
-describe("buildDriftAnalysisPrompt - scanned endpoints rendering", () => {
-    // Reproduces the [object Object] bug: skeletonEndpoints from analyzeChangesTool
-    // stores methods as objects { method: string, ... }, not plain strings.
-    const skeletonMethodObjects = [
-        {
-            path: "/api/v1/",
-            methods: [{ method: "GET", description: "", queryParams: [], authRequired: true, sourceFile: "main.py", interactions: [] }],
-            resourceGroup: "v1",
-            pathParams: [],
-        },
-        {
-            path: "/api/v1/orders",
-            methods: [
-                { method: "GET", description: "", queryParams: [], authRequired: true, sourceFile: "orders.py", interactions: [] },
-                { method: "POST", description: "", queryParams: [], authRequired: true, sourceFile: "orders.py", interactions: [] },
-            ],
-            resourceGroup: "orders",
-            pathParams: [],
-        },
-    ];
-    it("renders HTTP methods as strings, not [object Object]", () => {
-        const prompt = buildDriftAnalysisPrompt({
-            existingTests: [],
-            scannedEndpoints: skeletonMethodObjects,
-            repositoryPath: "/repo",
-            stateFile: "/tmp/state.json",
-        });
-        expect(prompt).not.toContain("[object Object]");
-        expect(prompt).toContain("GET /api/v1/");
-        expect(prompt).toContain("GET|POST /api/v1/orders");
-        // CTA should appear exactly once (not duplicated)
-        const ctaCount = (prompt.match(/call `skyramp_actions`/g) || []).length;
-        expect(ctaCount).toBe(1);
-    });
-    it("also works with plain string methods (ScannedEndpoint format)", () => {
-        const stringMethods = [
-            { path: "/api/v1/products", methods: ["GET", "POST"], sourceFile: "products.py" },
-        ];
-        const prompt = buildDriftAnalysisPrompt({
-            existingTests: [],
-            scannedEndpoints: stringMethods,
-            repositoryPath: "/repo",
-            stateFile: "/tmp/state.json",
-        });
-        expect(prompt).not.toContain("[object Object]");
-        expect(prompt).toContain("GET|POST /api/v1/products");
-    });
-});