@pauly4010/evalai-sdk 1.8.0 → 1.9.0

package/dist/cli/index.js

@@ -10,11 +10,17 @@
 Object.defineProperty(exports, "__esModule", { value: true });
 const baseline_1 = require("./baseline");
 const check_1 = require("./check");
+const ci_1 = require("./ci");
+const diff_1 = require("./diff");
+const discover_1 = require("./discover");
 const doctor_1 = require("./doctor");
 const explain_1 = require("./explain");
+const impact_analysis_1 = require("./impact-analysis");
 const init_1 = require("./init");
+const migrate_1 = require("./migrate");
 const print_config_1 = require("./print-config");
 const regression_gate_1 = require("./regression-gate");
+const run_1 = require("./run");
 const share_1 = require("./share");
 const upgrade_1 = require("./upgrade");
 const argv = process.argv.slice(2);
@@ -32,6 +38,60 @@ else if (subcommand === "gate") {
     const code = (0, regression_gate_1.runGate)(argv.slice(1));
     process.exit(code);
 }
+else if (subcommand === "migrate") {
+    // Handle migrate subcommand
+    const migrateSubcommand = argv[1];
+    if (migrateSubcommand === "config") {
+        // Parse migrate config arguments
+        let inputPath = "";
+        let outputPath = "";
+        let verbose = false;
+        let helpers = true;
+        let preserveIds = true;
+        let provenance = true;
+        for (let i = 2; i < argv.length; i++) {
+            const arg = argv[i];
+            if (arg === "--in" || arg === "-i") {
+                inputPath = argv[++i];
+            }
+            else if (arg === "--out" || arg === "-o") {
+                outputPath = argv[++i];
+            }
+            else if (arg === "--verbose" || arg === "-v") {
+                verbose = true;
+            }
+            else if (arg === "--no-helpers") {
+                helpers = false;
+            }
+            else if (arg === "--no-preserve-ids") {
+                preserveIds = false;
+            }
+            else if (arg === "--no-provenance") {
+                provenance = false;
+            }
+        }
+        if (!inputPath || !outputPath) {
+            console.error("Error: Both --in and --out options are required");
+            console.error("Usage: evalai migrate config --in <input> --out <output> [options]");
+            process.exit(1);
+        }
+        (0, migrate_1.migrateConfig)({
+            input: inputPath,
+            output: outputPath,
+            verbose,
+            helpers,
+            preserveIds,
+            provenance,
+        }).catch((err) => {
+            console.error(`Migration failed: ${err instanceof Error ? err.message : String(err)}`);
+            process.exit(1);
+        });
+    }
+    else {
+        console.error("Error: Unknown migrate subcommand. Use 'evalai migrate config'");
+        process.exit(1);
+    }
+}
 else if (subcommand === "upgrade") {
     const code = (0, upgrade_1.runUpgrade)(argv.slice(1));
     process.exit(code);
@@ -82,11 +142,111 @@ else if (subcommand === "share") {
         process.exit(1);
     });
 }
+else if (subcommand === "discover") {
+    // Parse arguments for discover command
+    const args = argv.slice(1);
+    const manifestFlag = args.includes("--manifest");
+    (0, discover_1.discoverSpecs)({ manifest: manifestFlag })
+        .then(() => process.exit(0))
+        .catch((err) => {
+        console.error(`EvalAI ERROR: ${err instanceof Error ? err.message : String(err)}`);
+        process.exit(1);
+    });
+}
+else if (subcommand === "impact-analysis") {
+    // Parse arguments for impact-analysis command
+    const args = argv.slice(1);
+    const baseIndex = args.indexOf("--base");
+    const changedFilesIndex = args.indexOf("--changed-files");
+    const formatIndex = args.indexOf("--format");
+    const baseBranch = baseIndex !== -1 ? args[baseIndex + 1] : "main";
+    const changedFiles = changedFilesIndex !== -1 ? args[changedFilesIndex + 1]?.split(",") : undefined;
+    const format = formatIndex !== -1 ? args[formatIndex + 1] : "human";
+    (0, impact_analysis_1.runImpactAnalysisCLI)({ baseBranch, changedFiles, format }).catch((err) => {
+        console.error(`EvalAI ERROR: ${err instanceof Error ? err.message : String(err)}`);
+        process.exit(2);
+    });
+}
+else if (subcommand === "run") {
+    // Parse arguments for run command
+    const args = argv.slice(1);
+    const specIdsIndex = args.indexOf("--spec-ids");
+    const impactedOnlyIndex = args.indexOf("--impacted-only");
+    const baseIndex = args.indexOf("--base");
+    const formatIndex = args.indexOf("--format");
+    const writeResultsIndex = args.indexOf("--write-results");
+    const specIds = specIdsIndex !== -1 ? args[specIdsIndex + 1]?.split(",") : undefined;
+    const impactedOnly = impactedOnlyIndex !== -1;
+    const baseBranch = baseIndex !== -1 ? args[baseIndex + 1] : undefined;
+    const format = formatIndex !== -1 ? args[formatIndex + 1] : "human";
+    const writeResults = writeResultsIndex !== -1;
+    (0, run_1.runEvaluationsCLI)({
+        specIds,
+        impactedOnly: impactedOnly ? !!baseBranch : false,
+        baseBranch,
+        format,
+        writeResults,
+    }).catch((err) => {
+        console.error(`EvalAI ERROR: ${err instanceof Error ? err.message : String(err)}`);
+        process.exit(2);
+    });
+}
+else if (subcommand === "diff") {
+    // Parse arguments for diff command
+    const args = argv.slice(1);
+    const baseIndex = args.indexOf("--base");
+    const headIndex = args.indexOf("--head");
+    const formatIndex = args.indexOf("--format");
+    const base = baseIndex !== -1 ? args[baseIndex + 1] : undefined;
+    const head = headIndex !== -1 ? args[headIndex + 1] : undefined;
+    const format = formatIndex !== -1 ? args[formatIndex + 1] : "human";
+    (0, diff_1.runDiffCLI)({ base, head, format }).catch((err) => {
+        console.error(`EvalAI ERROR: ${err instanceof Error ? err.message : String(err)}`);
+        process.exit(2);
+    });
+}
+else if (subcommand === "ci") {
+    // Parse arguments for ci command
+    const args = argv.slice(1);
+    const baseIndex = args.indexOf("--base");
+    const impactedOnlyIndex = args.indexOf("--impacted-only");
+    const formatIndex = args.indexOf("--format");
+    const writeResultsIndex = args.indexOf("--write-results");
+    const base = baseIndex !== -1 ? args[baseIndex + 1] : undefined;
+    const impactedOnly = impactedOnlyIndex !== -1;
+    const format = formatIndex !== -1 ? args[formatIndex + 1] : "human";
+    const writeResults = writeResultsIndex !== -1;
+    (0, ci_1.runCICLI)({ base, impactedOnly, format, writeResults }).catch((err) => {
+        console.error(`EvalAI ERROR: ${err instanceof Error ? err.message : String(err)}`);
+        process.exit(2);
+    });
+}
 else {
     console.log(`EvalAI CLI
 
 Usage:
   evalai init                Create evalai.config.json + baseline + CI workflow
+  evalai discover            Discover behavioral specs in project and show statistics
+  evalai discover --manifest  Generate evaluation manifest for incremental analysis
+  evalai impact-analysis     Analyze impact of changes and suggest targeted tests
+    --base <branch>          Base branch to compare against (default: main)
+    --changed-files <files>  Comma-separated list of changed files (for CI)
+    --format <fmt>           Output format: human (default), json
+  evalai ci                  One-command CI loop (manifest → impact → run → diff)
+    --base <ref>            Base reference for diff (baseline|last|<runId>|<path>|<gitref>)
+    --impacted-only          Run only specs impacted by changes
+    --format <fmt>           Output format: human (default), json, github
+    --write-results          Write run results to .evalai/last-run.json
+  evalai run                 Run evaluation specifications
+    --spec-ids <ids>         Comma-separated list of spec IDs to run
+    --impacted-only          Run only specs impacted by changes (requires --base)
+    --base <branch>          Base branch for impact analysis (with --impacted-only)
+    --format <fmt>           Output format: human (default), json
+    --write-results          Write results to .evalai/last-run.json
+  evalai diff                Compare two run reports and show behavioral changes
+    --base <branch>          Base branch or report path (default: main)
+    --head <path>            Head report path (default: .evalai/last-run.json)
+    --format <fmt>           Output format: human (default), json
   evalai gate [options]      Run regression gate (local test-based, no API needed)
   evalai check [options]     CI/CD evaluation gate (API-based)
   evalai explain [options]   Explain last gate/check failure with root causes + fixes
@@ -133,6 +293,19 @@ Options for doctor:
 
 Examples:
   evalai init
+  evalai discover
+  evalai discover --manifest
+  evalai impact-analysis --base main
+  evalai impact-analysis --base main --format json
+  evalai impact-analysis --changed-files src/utils.ts,datasets/test.json
+  evalai run
+  evalai run --spec-ids spec1,spec2
+  evalai run --impacted-only --base main
+  evalai run --format json --write-results
+  evalai diff
+  evalai diff --base main
+  evalai diff --base main --format json
+  evalai diff --a .evalai/runs/base.json --b .evalai/last-run.json
   evalai gate
   evalai gate --format json
   evalai explain

package/dist/cli/manifest.d.ts

@@ -0,0 +1,105 @@
+/**
+ * TICKET 2 — Evaluation Manifest Generation
+ *
+ * Goal: turn discovery output into a stable, versioned, machine-consumable artifact
+ * that becomes the input to run / impact / diff.
+ *
+ * This is the compiler output that everything else consumes.
+ */
+import type { SpecAnalysis } from "./discover";
+import type { ExecutionModeConfig } from "../runtime/execution-mode";
+/**
+ * Manifest schema version
+ */
+export declare const MANIFEST_SCHEMA_VERSION = 1;
+/**
+ * SDK version from package.json
+ */
+export declare const SDK_VERSION = "1.8.0";
+/**
+ * Evaluation Manifest Schema
+ */
+export interface EvaluationManifest {
+    /** Schema version for compatibility */
+    schemaVersion: number;
+    /** When this manifest was generated */
+    generatedAt: number;
+    /** Project metadata */
+    project: {
+        name: string;
+        root: string;
+        namespace: string;
+    };
+    /** Runtime information */
+    runtime: {
+        mode: "spec" | "legacy";
+        sdkVersion: string;
+    };
+    /** Spec files with hashes */
+    specFiles: SpecFile[];
+    /** Individual specifications */
+    specs: Spec[];
+}
+/**
+ * Spec file information
+ */
+export interface SpecFile {
+    /** POSIX-relative file path */
+    filePath: string;
+    /** SHA-256 hash of file content */
+    fileHash: string;
+    /** Number of specs in this file */
+    specCount: number;
+}
+/**
+ * Individual specification
+ */
+export interface Spec {
+    /** Stable canonical ID */
+    id: string;
+    /** Spec name */
+    name: string;
+    /** Suite path from tags or file structure */
+    suitePath: string[];
+    /** POSIX-relative file path */
+    filePath: string;
+    /** Position in file */
+    position: {
+        line: number;
+        column: number;
+    };
+    /** Tags/categories */
+    tags: string[];
+    /** Dependencies */
+    dependsOn: {
+        prompts: string[];
+        datasets: string[];
+        tools: string[];
+        code: string[];
+    };
+}
+/**
+ * Lock file for caching
+ */
+export interface ManifestLock {
+    /** When lock was generated */
+    generatedAt: number;
+    /** File hashes for incremental updates */
+    fileHashes: Record<string, string>;
+}
+/**
+ * Generate evaluation manifest from discovery results
+ */
+export declare function generateManifest(specs: SpecAnalysis[], projectRoot: string, projectName: string, executionMode: ExecutionModeConfig): Promise<EvaluationManifest>;
+/**
+ * Write manifest to disk
+ */
+export declare function writeManifest(manifest: EvaluationManifest, projectRoot: string): Promise<void>;
+/**
+ * Read existing manifest
+ */
+export declare function readManifest(projectRoot: string): Promise<EvaluationManifest | null>;
+/**
+ * Read existing lock file
+ */
+export declare function readLock(projectRoot: string): Promise<ManifestLock | null>;

package/dist/cli/manifest.js

@@ -0,0 +1,275 @@
+"use strict";
+/**
+ * TICKET 2 — Evaluation Manifest Generation
+ *
+ * Goal: turn discovery output into a stable, versioned, machine-consumable artifact
+ * that becomes the input to run / impact / diff.
+ *
+ * This is the compiler output that everything else consumes.
+ */
+var __createBinding = (this && this.__createBinding) || (Object.create ? (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    var desc = Object.getOwnPropertyDescriptor(m, k);
+    if (!desc || ("get" in desc ? !m.__esModule : desc.writable || desc.configurable)) {
+      desc = { enumerable: true, get: function() { return m[k]; } };
+    }
+    Object.defineProperty(o, k2, desc);
+}) : (function(o, m, k, k2) {
+    if (k2 === undefined) k2 = k;
+    o[k2] = m[k];
+}));
+var __setModuleDefault = (this && this.__setModuleDefault) || (Object.create ? (function(o, v) {
+    Object.defineProperty(o, "default", { enumerable: true, value: v });
+}) : function(o, v) {
+    o["default"] = v;
+});
+var __importStar = (this && this.__importStar) || (function () {
+    var ownKeys = function(o) {
+        ownKeys = Object.getOwnPropertyNames || function (o) {
+            var ar = [];
+            for (var k in o) if (Object.prototype.hasOwnProperty.call(o, k)) ar[ar.length] = k;
+            return ar;
+        };
+        return ownKeys(o);
+    };
+    return function (mod) {
+        if (mod && mod.__esModule) return mod;
+        var result = {};
+        if (mod != null) for (var k = ownKeys(mod), i = 0; i < k.length; i++) if (k[i] !== "default") __createBinding(result, mod, k[i]);
+        __setModuleDefault(result, mod);
+        return result;
+    };
+})();
+Object.defineProperty(exports, "__esModule", { value: true });
+exports.SDK_VERSION = exports.MANIFEST_SCHEMA_VERSION = void 0;
+exports.generateManifest = generateManifest;
+exports.writeManifest = writeManifest;
+exports.readManifest = readManifest;
+exports.readLock = readLock;
+const fs = __importStar(require("node:fs/promises"));
+const path = __importStar(require("node:path"));
+const crypto = __importStar(require("node:crypto"));
+/**
+ * Manifest schema version
+ */
+exports.MANIFEST_SCHEMA_VERSION = 1;
+/**
+ * SDK version from package.json
+ */
+exports.SDK_VERSION = "1.8.0";
+/**
+ * Generate evaluation manifest from discovery results
+ */
+async function generateManifest(specs, projectRoot, projectName, executionMode) {
+    const generatedAt = Math.floor(Date.now() / 1000);
+    const namespace = generateNamespace(projectRoot);
+    // Process spec files and specs
+    const specFiles = [];
+    const processedSpecs = [];
+    // Group specs by file
+    const specsByFile = new Map();
+    for (const spec of specs) {
+        const normalizedPath = normalizePath(spec.file, projectRoot);
+        if (!specsByFile.has(normalizedPath)) {
+            specsByFile.set(normalizedPath, []);
+        }
+        specsByFile.get(normalizedPath).push(spec);
+    }
+    // Process each file
+    for (const [filePath, fileSpecs] of specsByFile) {
+        const absolutePath = path.join(projectRoot, filePath);
+        const fileHash = await hashFile(absolutePath);
+        specFiles.push({
+            filePath,
+            fileHash,
+            specCount: fileSpecs.length,
+        });
+        // Process individual specs
+        for (const spec of fileSpecs) {
+            const processedSpec = await processSpec(spec, filePath, projectRoot);
+            processedSpecs.push(processedSpec);
+        }
+    }
+    return {
+        schemaVersion: exports.MANIFEST_SCHEMA_VERSION,
+        generatedAt,
+        project: {
+            name: projectName,
+            root: ".",
+            namespace,
+        },
+        runtime: {
+            mode: executionMode.mode,
+            sdkVersion: exports.SDK_VERSION,
+        },
+        specFiles,
+        specs: processedSpecs,
+    };
+}
+/**
+ * Process individual specification
+ */
+async function processSpec(spec, filePath, projectRoot) {
+    const absolutePath = path.join(projectRoot, filePath);
+    const content = await fs.readFile(absolutePath, "utf-8");
+    // Extract position from AST analysis (simplified for now)
+    const position = extractPosition(content, spec.name);
+    // Extract dependencies from content
+    const dependsOn = extractDependencies(content);
+    // Generate suite path from tags or file structure
+    const suitePath = generateSuitePath(spec.tags, filePath);
+    return {
+        id: spec.id,
+        name: spec.name,
+        suitePath,
+        filePath: normalizePath(spec.file, projectRoot),
+        position,
+        tags: spec.tags,
+        dependsOn,
+    };
+}
+/**
+ * Extract position from content (simplified implementation)
+ */
+function extractPosition(content, specName) {
+    const lines = content.split("\n");
+    const specPattern = new RegExp(`defineEval\\s*\\(\\s*["'\`]${specName}["'\`]`, "g");
+    let match = null;
+    let line = 1;
+    let column = 1;
+    for (let i = 0; i < lines.length; i++) {
+        const lineContent = lines[i];
+        specPattern.lastIndex = 0;
+        match = specPattern.exec(lineContent);
+        if (match) {
+            line = i + 1;
+            column = match.index + 1;
+            break;
+        }
+    }
+    return { line, column };
+}
+/**
+ * Extract dependencies from content
+ */
+function extractDependencies(content) {
+    const dependsOn = {
+        prompts: [],
+        datasets: [],
+        tools: [],
+        code: [],
+    };
+    // Extract from dependsOn option if present
+    const dependsOnMatch = content.match(/dependsOn\s*:\s*({[^}]+})/s);
+    if (dependsOnMatch) {
+        try {
+            const deps = eval(`(${dependsOnMatch[1]})`);
+            return {
+                prompts: deps.prompts || [],
+                datasets: deps.datasets || [],
+                tools: deps.tools || [],
+                code: deps.code || [],
+            };
+        }
+        catch (error) {
+            // Fall back to simple extraction
+        }
+    }
+    // Simple extraction as fallback
+    const patterns = {
+        prompts: /["']([^"']*\.md)["']/g,
+        datasets: /["']([^"']*\.json)["']/g,
+        tools: /["']([^"']*\.ts)["']/g,
+        code: /import.*from\s*["']([^"']+)["']/g,
+    };
+    for (const [type, pattern] of Object.entries(patterns)) {
+        let match;
+        while ((match = pattern.exec(content)) !== null) {
+            dependsOn[type].push(match[1]);
+        }
+    }
+    return dependsOn;
+}
+/**
+ * Generate suite path from tags or file structure
+ */
+function generateSuitePath(tags, filePath) {
+    // Use tags as primary suite path
+    if (tags.length > 0) {
+        return [tags[0]];
+    }
+    // Fall back to file structure
+    const parts = filePath.split("/");
+    if (parts.length > 1) {
+        return [parts[0]];
+    }
+    return ["general"];
+}
+/**
+ * Generate namespace from project root
+ */
+function generateNamespace(projectRoot) {
+    const hash = crypto.createHash("sha256");
+    hash.update(projectRoot);
+    return hash.digest("hex").slice(0, 8);
+}
+/**
+ * Normalize path to POSIX format
+ */
+function normalizePath(filePath, projectRoot) {
+    const relativePath = path.relative(projectRoot, filePath);
+    return relativePath.replace(/\\/g, "/");
+}
+/**
+ * Hash file content
+ */
+async function hashFile(filePath) {
+    const content = await fs.readFile(filePath, "utf-8");
+    const hash = crypto.createHash("sha256");
+    hash.update(content);
+    return `sha256:${hash.digest("hex")}`;
+}
+/**
+ * Write manifest to disk
+ */
+async function writeManifest(manifest, projectRoot) {
+    const evalaiDir = path.join(projectRoot, ".evalai");
+    // Ensure .evalai directory exists
+    await fs.mkdir(evalaiDir, { recursive: true });
+    // Write manifest.json
+    const manifestPath = path.join(evalaiDir, "manifest.json");
+    await fs.writeFile(manifestPath, JSON.stringify(manifest, null, 2), "utf-8");
+    // Write lock file
+    const lock = {
+        generatedAt: manifest.generatedAt,
+        fileHashes: Object.fromEntries(manifest.specFiles.map((f) => [f.filePath, f.fileHash])),
+    };
+    const lockPath = path.join(evalaiDir, "manifest.lock.json");
+    await fs.writeFile(lockPath, JSON.stringify(lock, null, 2), "utf-8");
+}
+/**
+ * Read existing manifest
+ */
+async function readManifest(projectRoot) {
+    const manifestPath = path.join(projectRoot, ".evalai", "manifest.json");
+    try {
+        const content = await fs.readFile(manifestPath, "utf-8");
+        return JSON.parse(content);
+    }
+    catch (error) {
+        return null;
+    }
+}
+/**
+ * Read existing lock file
+ */
+async function readLock(projectRoot) {
+    const lockPath = path.join(projectRoot, ".evalai", "manifest.lock.json");
+    try {
+        const content = await fs.readFile(lockPath, "utf-8");
+        return JSON.parse(content);
+    }
+    catch (error) {
+        return null;
+    }
+}

package/dist/cli/migrate.d.ts

@@ -0,0 +1,41 @@
+/**
+ * COMPAT-203: Config → DSL migration generator (file-based)
+ *
+ * CLI command: evalai migrate config --in evalai.config.json --out eval/legacy.spec.ts
+ * Generates defineEval() calls with comments and TODOs for manual completion
+ */
+import { Command } from "commander";
+/**
+ * Migration options
+ */
+interface MigrateOptions {
+    /** Input config file path */
+    input: string;
+    /** Output DSL file path */
+    output: string;
+    /** Include detailed comments */
+    verbose?: boolean;
+    /** Generate helper functions */
+    helpers?: boolean;
+    /** Preserve original test IDs */
+    preserveIds?: boolean;
+    /** Include provenance metadata */
+    provenance?: boolean;
+}
+/**
+ * Main migration function
+ */
+export declare function migrateConfig(options: MigrateOptions): Promise<void>;
+/**
+ * CLI command definition
+ */
+export declare function createMigrateCommand(): Command;
+/**
+ * Validate config file structure
+ */
+export declare function validateConfigFile(filePath: string): Promise<boolean>;
+/**
+ * Show migration preview without writing files
+ */
+export declare function previewMigration(filePath: string): Promise<void>;
+export {};