@skyramp/mcp 0.3.8 → 0.4.0-rc.1

package/build/utils/repoScanner.js

@@ -1,300 +0,0 @@
-import * as fs from "fs";
-import * as path from "path";
-import { nextjsFileToApiPath, parseFileEndpoints, } from "./routeParsers.js";
-import { resolvePythonMountPrefixes } from "./pythonMountPrefixes.js";
-import { walkSourceFiles } from "./fileWalk.js";
-import { logger } from "./logger.js";
-const GLOB_SKIP_DIRS = new Set(["node_modules", ".git", ".next", "dist", "build"]);
-function globRecursive(dir, extensions) {
-    return walkSourceFiles(dir, { extensions, skipDirs: GLOB_SKIP_DIRS });
-}
-function safeReadFile(filePath) {
-    try {
-        return fs.readFileSync(filePath, "utf-8");
-    }
-    catch {
-        return null;
-    }
-}
-/**
- * Returns the relative paths of entry-point files that define the routing
- * hierarchy for this repository. The LLM reads these files in Step 1.5 to
- * trace router/module nesting and build an authoritative path resolution table.
- *
- * Intentionally returns only file paths — no extracted lines. Static regex
- * extraction is fragile and framework-specific; the LLM reads the files directly
- * and understands any framework's routing syntax.
- */
-export function grepRouterMountingContext(repositoryPath) {
-    // NestJS root module (app.module.ts) is included alongside conventional
-    // entry-point names because NestJS routing is implicit through module
-    // composition rather than explicit mount calls.
-    const entryCandidates = new Set([
-        "main.py", "app.py", "server.py", "asgi.py", "wsgi.py",
-        "app.ts", "app.js", "server.ts", "server.js", "index.ts", "index.js",
-        "urls.py",
-        "main.go", "server.go", "router.go", "routes.go",
-        "routes.rb", "web.php", "api.php",
-        "app.module.ts", "app.module.js",
-        "router.ex", // Phoenix/Elixir — defines all scopes and pipelines
-        "main.rs", "lib.rs", // Rust (Axum, Actix, Rocket)
-    ]);
-    // Content signals that indicate a file wires routers/modules together.
-    const mountPatterns = [
-        /include_router/i,
-        /register_blueprint/i,
-        /\.use\s*\(\s*["']\//,
-        /urlpatterns/,
-        /path\s*\(.+include\s*\(/,
-        /\.Group\s*\(/,
-        /Route::prefix/,
-        /namespace\s+(?::[A-Za-z0-9_]+|["'][^"']+["'])/, // Rails: namespace :api or namespace "/api"
-        /scope\s+["']\//,
-        /@Module\s*\(/,
-        /scope\s+["'].*,\s*do/, // Phoenix: scope "/api", MyApp.Router, do
-    ];
-    const found = [];
-    const seenAbsPaths = new Set();
-    // Single glob from repo root — globRecursive already recurses so scanning
-    // each subdir separately would redundantly re-traverse those subtrees.
-    const allFiles = globRecursive(repositoryPath, [".py", ".ts", ".js", ".go", ".rb", ".php", ".rs", ".ex"]);
-    for (const f of allFiles) {
-        if (/test/i.test(f))
-            continue;
-        if (seenAbsPaths.has(f))
-            continue;
-        const basename = path.basename(f);
-        const isNamedEntry = entryCandidates.has(basename);
-        const isNestModule = /\.module\.(ts|js)$/.test(f);
-        if (!isNamedEntry && !isNestModule)
-            continue;
-        const fc = safeReadFile(f);
-        if (fc === null)
-            continue;
-        if (isNestModule) {
-            if (!/@Module\s*\(/.test(fc) || !/imports\s*:\s*\[/.test(fc))
-                continue;
-        }
-        else {
-            if (!mountPatterns.some((p) => p.test(fc)))
-                continue;
-        }
-        seenAbsPaths.add(f);
-        found.push(f.startsWith(repositoryPath) ? f.slice(repositoryPath.length + 1) : f);
-    }
-    return found;
-}
-function addEndpointToMap(endpointMap, apiPath, method, sourceFile, repositoryPath, line) {
-    const relative = sourceFile.startsWith(repositoryPath)
-        ? sourceFile.slice(repositoryPath.length + 1) : sourceFile;
-    const normalizedPath = apiPath.startsWith("/") ? apiPath : `/${apiPath}`;
-    const key = `${relative}::${normalizedPath}`;
-    const existing = endpointMap.get(key);
-    if (existing) {
-        existing.methods.add(method);
-        // Keep the earliest declaration line so ownership-range scoping starts there.
-        if (typeof line === "number" && (existing.line === undefined || line < existing.line)) {
-            existing.line = line;
-        }
-    }
-    else {
-        endpointMap.set(key, { path: normalizedPath, methods: new Set([method]), sourceFile: relative, line });
-    }
-}
-function scanNextjsFile(file, repositoryPath, endpointMap) {
-    const relative = file.startsWith(repositoryPath)
-        ? file.slice(repositoryPath.length + 1) : file;
-    const apiPath = nextjsFileToApiPath(relative);
-    if (!apiPath)
-        return false;
-    const content = safeReadFile(file);
-    if (content === null) {
-        addEndpointToMap(endpointMap, apiPath, "MULTI", file, repositoryPath);
-        return true;
-    }
-    const detectedMethods = new Set();
-    for (const line of content.split("\n")) {
-        const mc = line.match(/req\.method\s*===?\s*["'](GET|POST|PUT|PATCH|DELETE)["']/i);
-        if (mc)
-            detectedMethods.add(mc[1].toUpperCase());
-        const em = line.match(/export\s+(?:async\s+)?function\s+(GET|POST|PUT|PATCH|DELETE)\s*\(/i);
-        if (em)
-            detectedMethods.add(em[1].toUpperCase());
-    }
-    if (detectedMethods.size > 0) {
-        for (const m of detectedMethods)
-            addEndpointToMap(endpointMap, apiPath, m, file, repositoryPath);
-    }
-    else {
-        addEndpointToMap(endpointMap, apiPath, "MULTI", file, repositoryPath);
-    }
-    return true;
-}
-/** Filename pattern used to identify candidate route/handler files. */
-// Extended to cover NestJS (service, gateway, resolver) and other frameworks.
-// Tested against the full relative path — terms here are specific enough not to over-match.
-const ROUTE_FILE_PATTERN = /route|controller|endpoint|handler|view|urls|api|router|service|gateway|resolver|\bserver\b/i;
-// Generic terms like "app" and "main" must only match the basename (filename),
-// not directory names — otherwise every file under src/app/ would match by-name
-// and fill MAX_CANDIDATE_FILES, potentially skipping the content pass.
-const ROUTE_FILE_BASENAME_PATTERN = /\bapp\b|\bmain\b/i;
-/**
- * Content-based routing signature: a file is a route file if it contains BOTH
- * a URL-path-like string literal AND an HTTP method registration, regardless of
- * framework or language. This catches any framework whose files don't have
- * route/controller/handler-like names (e.g. Rust `main.rs`, .NET `Startup.cs`,
- * Axum, Ktor, Hono, Elysia, etc.).
- *
- * Two separate patterns are intentional — requiring both suppresses false
- * positives from files that merely reference HTTP methods or paths in other
- * contexts (test fixtures, config, documentation).
- */
-const ROUTING_URL_PATTERN = /["'`](\/[a-zA-Z0-9_{}:*-][a-zA-Z0-9_{}/:*-]*)/;
-const ROUTING_METHOD_PATTERN = /\.(get|post|put|patch|delete|head|options)\s*\(|@(Get|Post|Put|Patch|Delete|Route|Controller|Handler|RequestMapping)\b|route\s*\(["']|path\s*\(["']|HandleFunc|ServeHTTP|\.(Map|map)(Get|Post|Put|Patch|Delete)\s*\(|\brouting\s*\{/i;
-/** Maximum files to return — keeps the LLM prompt size bounded. */
-const MAX_CANDIDATE_FILES = 60;
-/**
- * Return relative paths of all source files that are likely to define API
- * routes. Two-pass strategy:
- *   1. Fast: filename matches ROUTE_FILE_PATTERN (no file I/O needed).
- *   2. Content: files not matched by name are read and checked for both a
- *      URL-path literal AND an HTTP method registration. This covers any
- *      framework the filename filter misses — Axum, Ktor, .NET, Hono, etc.
- * Result is capped at MAX_CANDIDATE_FILES to keep the LLM prompt bounded.
- */
-export function findCandidateRouteFiles(repositoryPath) {
-    const sourceFiles = globRecursive(repositoryPath, [
-        ".ts", ".tsx", ".js", ".jsx",
-        ".py", ".java", ".kt",
-        ".go", ".rb", ".php",
-        ".rs", ".cs", ".ex", ".exs",
-    ]);
-    const byName = [];
-    const unchecked = [];
-    for (const f of sourceFiles) {
-        if (/test/i.test(f))
-            continue;
-        const relative = f.startsWith(repositoryPath) ? f.slice(repositoryPath.length + 1) : f;
-        if (ROUTE_FILE_PATTERN.test(relative) || ROUTE_FILE_BASENAME_PATTERN.test(path.basename(relative))) {
-            byName.push(relative);
-        }
-        else {
-            unchecked.push(f); // store absolute path for readFileSync
-        }
-    }
-    // Content pass — only run if name pass left budget remaining
-    const byContent = [];
-    if (byName.length < MAX_CANDIDATE_FILES) {
-        const budget = MAX_CANDIDATE_FILES - byName.length;
-        let scanned = 0;
-        const MAX_SCAN = 2000; // cap file reads; the unchecked array can be much larger
-        for (const f of unchecked) {
-            if (byContent.length >= budget || scanned++ >= MAX_SCAN)
-                break;
-            const content = safeReadFile(f);
-            if (content === null)
-                continue;
-            if (ROUTING_URL_PATTERN.test(content) && ROUTING_METHOD_PATTERN.test(content)) {
-                const relative = f.startsWith(repositoryPath) ? f.slice(repositoryPath.length + 1) : f;
-                byContent.push(relative);
-            }
-        }
-    }
-    const combined = [...byName, ...byContent];
-    const contentPassSkipped = byName.length >= MAX_CANDIDATE_FILES;
-    const result = combined.slice(0, MAX_CANDIDATE_FILES);
-    if (combined.length > MAX_CANDIDATE_FILES) {
-        logger.warning("findCandidateRouteFiles: hit MAX_CANDIDATE_FILES cap — some route files may be excluded from LLM analysis", {
-            cap: MAX_CANDIDATE_FILES,
-            byName: byName.length,
-            byContent: byContent.length,
-            contentPassSkipped,
-            totalFound: combined.length,
-        });
-    }
-    return result;
-}
-/**
- * Cross-file FastAPI mount-prefix lookup, resolved lazily: `resolvePythonMountPrefixes`
- * walks the whole repo's Python, so it's only invoked the first time a `.py` file is
- * actually scanned — non-Python repos pay nothing. Returns undefined for non-`.py`
- * files (they never carry Python mount prefixes).
- */
-function makeMountPrefixLookup(repositoryPath) {
-    let resolved = null;
-    return (relative) => {
-        if (!relative.endsWith(".py"))
-            return undefined;
-        if (resolved === null)
-            resolved = resolvePythonMountPrefixes(repositoryPath);
-        return resolved.get(relative.replace(/\\/g, "/"));
-    };
-}
-export function scanAllRepoEndpoints(repositoryPath) {
-    // Normalize up-front so relative-path derivation (and mount-prefix lookups) don't
-    // depend on a caller-supplied trailing separator.
-    repositoryPath = path.resolve(repositoryPath);
-    const endpointMap = new Map();
-    const mountPrefixFor = makeMountPrefixLookup(repositoryPath);
-    const sourceFiles = globRecursive(repositoryPath, [".ts", ".tsx", ".js", ".jsx", ".py", ".java", ".kt", ".go", ".rb", ".php", ".rs", ".cs"]);
-    for (const file of sourceFiles) {
-        if (scanNextjsFile(file, repositoryPath, endpointMap))
-            continue;
-        const relative = file.startsWith(repositoryPath)
-            ? file.slice(repositoryPath.length + 1) : file;
-        if (!ROUTE_FILE_PATTERN.test(relative) && !ROUTE_FILE_BASENAME_PATTERN.test(path.basename(relative)))
-            continue;
-        const content = safeReadFile(file);
-        if (content === null)
-            continue;
-        for (const ep of parseFileEndpoints(content, relative, mountPrefixFor(relative))) {
-            addEndpointToMap(endpointMap, ep.path, ep.method, file, repositoryPath, ep.line);
-        }
-    }
-    return Array.from(endpointMap.values()).map((data) => ({
-        path: data.path,
-        methods: Array.from(data.methods),
-        sourceFile: data.sourceFile,
-        ...(data.line !== undefined ? { line: data.line } : {}),
-    }));
-}
-export function scanRelatedEndpoints(repositoryPath, changedFiles) {
-    repositoryPath = path.resolve(repositoryPath);
-    const relatedDirs = new Set();
-    for (const f of changedFiles) {
-        const absDir = path.dirname(path.join(repositoryPath, f));
-        relatedDirs.add(absDir);
-        const parentDir = path.dirname(absDir);
-        if (parentDir !== repositoryPath)
-            relatedDirs.add(parentDir);
-    }
-    const endpointMap = new Map();
-    // Mount prefixes are resolved over the whole repo (aggregators may live outside
-    // the related dirs) but lazily — only once a Python file is actually scanned.
-    const mountPrefixFor = makeMountPrefixLookup(repositoryPath);
-    for (const dir of relatedDirs) {
-        if (!fs.existsSync(dir))
-            continue;
-        const files = globRecursive(dir, [".ts", ".tsx", ".js", ".jsx", ".py", ".java", ".kt", ".go", ".rb", ".php", ".rs", ".cs"]);
-        for (const file of files) {
-            if (scanNextjsFile(file, repositoryPath, endpointMap))
-                continue;
-            const relative = file.startsWith(repositoryPath)
-                ? file.slice(repositoryPath.length + 1) : file;
-            if (!ROUTE_FILE_PATTERN.test(relative) && !ROUTE_FILE_BASENAME_PATTERN.test(path.basename(relative)))
-                continue;
-            const fileContent = safeReadFile(file);
-            if (fileContent === null)
-                continue;
-            for (const ep of parseFileEndpoints(fileContent, relative, mountPrefixFor(relative))) {
-                addEndpointToMap(endpointMap, ep.path, ep.method, file, repositoryPath, ep.line);
-            }
-        }
-    }
-    return Array.from(endpointMap.values()).map((data) => ({
-        path: data.path,
-        methods: Array.from(data.methods),
-        sourceFile: data.sourceFile,
-        ...(data.line !== undefined ? { line: data.line } : {}),
-    }));
-}

package/build/utils/routeParsers.d.ts

@@ -1,95 +0,0 @@
-import { BranchDiffData } from "./branchDiff.js";
-import type { ScannedEndpoint } from "./repoScanner.js";
-export interface ParsedDiffEndpoint {
-    method: string;
-    path: string;
-    sourceFile: string;
-    /** 1-based line where the route declaration starts, when parsed from full file content. */
-    line?: number;
-}
-export declare function nextjsFileToApiPath(filePath: string): string | null;
-export declare const UI_FILE_EXTENSIONS: readonly ["tsx", "jsx", "vue", "svelte", "html", "xml"];
-export declare const UI_FILE_GIT_PATHSPEC: string;
-export declare const UI_FILE_EXTENSIONS_PIPED: string;
-export declare function parseRouteLine(line: string, sourceFile: string): ParsedDiffEndpoint | null;
-/**
- * Scan all route definitions in a single source file, combining prefix declarations
- * with method-level annotations. Two orthogonal state machines run in parallel:
- *
- * 1. Class-prefix (Spring, NestJS, ASP.NET, Kotlin):
- *    @Controller/@RestController/@RequestMapping (class level) + class declaration →
- *    activates classPrefix, applied to all @Get/@Post/@GetMapping etc. below.
- *
- * 2. Router-variable prefix (FastAPI, Flask Blueprint):
- *    varName = APIRouter(prefix=...) or Blueprint(url_prefix=...) — handles both
- *    single-line and multi-line declarations — maps the variable name to its prefix.
- *    Applied when @varName.method(...) is seen.
- *
- * Falls back to pure parseRouteLine for all other frameworks (Express inline, Gin,
- * Rails, Laravel, Actix, Hapi, Go stdlib, etc.), which remain unaffected.
- *
- * `mountPrefixes` optionally supplies, per router variable, the **ancestor**
- * prefix that router inherits from cross-file `include_router(...)` mount points
- * (see `resolvePythonMountPrefixes`). It is composed *ahead* of the router's own
- * in-file prefix, so a `router = APIRouter(prefix="/recipes")` mounted under an
- * aggregator's `APIRouter(prefix="/api")` yields `/api/recipes`, not `/recipes`.
- */
-export declare function parseFileEndpoints(content: string, sourceFile: string, mountPrefixes?: Map<string, string>): ParsedDiffEndpoint[];
-export declare function isParamSegment(segment: string): boolean;
-export declare function isOpaqueIdSegment(segment: string): boolean;
-export declare const SKIP_PATH_SEGMENTS: Set<string>;
-/**
- * True for version/prefix path segments — "v1".."v99" and the literal "api".
- * `SKIP_PATH_SEGMENTS` enumerates only v1-v3, so resource-token derivation uses
- * this predicate as well; without it any endpoint versioned beyond v3 would
- * derive a nonsense token ("v4") as its resource (SKYR-3857).
- */
-export declare function isVersionLikeSegment(segment: string): boolean;
-/**
- * Extract the primary resource key from an endpoint path for use as a dedup key.
- *
- * For top-level resources (e.g. "/api/v1/orders/{id}") returns the resource name ("orders").
- * For sub-resources (e.g. "/orders/{id}/items") returns "orders_items" — including the
- * parent prevents "/orders/{id}/items" and "/products/{id}/items" from colliding on "items".
- */
-export declare function extractResourceFromPath(endpointPath: string): string;
-export declare function isRouteLikeFile(filePath: string): boolean;
-/**
- * Files that may contain endpoints removed in this branch: every changed file
- * except newly-created ones. Both whole-file deletions AND line-edits that drop a
- * route from a surviving file belong here — newly-created files cannot contain a
- * route that existed in the base branch.
- */
-export declare function selectRemovalCandidateFiles(diffData: BranchDiffData): string[];
-export declare function recoverRemovedEndpointsFromBase(candidateFiles: string[], fetchBaseContent: (file: string) => Promise<string>, repositoryPath?: string, baseBranch?: string): Promise<ScannedEndpoint[]>;
-/** Strip git's `a/` / `b/` diff-path prefix and normalize separators. */
-export declare function normalizeDiffPath(spec: string): string;
-/**
- * Parse a unified diff into the set of new-side (post-change) line numbers that
- * were touched, keyed by normalized file path.
- *
- * Context and added (`+`) lines advance the new-side cursor; added lines are
- * recorded. Removed (`-`) lines don't advance the cursor but record the current
- * new-side position, so deletion-only hunks still map to a location (otherwise a
- * pure deletion would scope to nothing and fall back to whole-file matching).
- * Feeds hunk-aware scoping in `classifyEndpointsByChangedFiles` so a
- * change can be attributed to the specific route whose declaration owns those
- * lines — instead of flagging every route in a touched file (the failure mode
- * on centralized-routing SUTs that register the whole API in one hub file).
- */
-export declare function parseChangedLinesByFile(diffContent: string): Map<string, Set<number>>;
-export interface ClassifiedEndpoints {
-    /** Endpoints in scanned catalog whose sourceFile is in changedFiles (not new/deleted). */
-    changedEndpoints: ScannedEndpoint[];
-    /** Endpoints whose sourceFile is in a newly-created file. */
-    newEndpoints: ScannedEndpoint[];
-    /** Endpoints whose sourceFile is in a deleted file (recovered from base branch). */
-    removedEndpoints: ScannedEndpoint[];
-    /** Changed files that don't map to any scanned endpoint. */
-    unmatchedFiles: string[];
-    changedFiles: string[];
-    currentBranch: string;
-    baseBranch: string;
-    affectedServices: string[];
-}
-export declare function classifyEndpointsByChangedFiles(diffData: BranchDiffData, scannedEndpoints: ScannedEndpoint[], deletedFileEndpoints?: ScannedEndpoint[]): ClassifiedEndpoints;