hackmyagent 0.11.13 → 0.11.14

package/dist/cli.js

@@ -1743,8 +1743,9 @@ Examples:
     .option('-b, --benchmark <name>', 'Run benchmark compliance check (e.g., oasb-1)')
     .option('-l, --level <level>', 'Benchmark level: L1 (Essential), L2 (Standard), L3 (Hardened)', 'L1')
     .option('-c, --category <name>', 'Filter to specific benchmark category')
-    .option('--deep', 'Enable LLM-powered semantic analysis (requires ANTHROPIC_API_KEY)')
-    .option('--scan-depth <depth>', 'CAAT scan depth: quick (config+creds only), standard (default), deep (+ LLM analysis)', 'standard')
+    .option('--deep', 'Maximum analysis: static + NanoMind + behavioral simulation + adaptive attacks (~30s per artifact)')
+    .option('--static-only', 'Disable NanoMind and simulation (static checks only, fast, deterministic)')
+    .option('--scan-depth <depth>', 'CAAT scan depth: quick (config+creds only), standard (default), deep (+ simulation)', 'standard')
     .option('--ci-publish', 'Submit scan results to registry CI endpoint (requires CI_SCAN_HMAC_SECRET env)')
     .option('--publish', 'Push scan results to the OpenA2A Registry')
     .option('--registry-report', 'Post results to OpenA2A Registry')
@@ -1817,17 +1818,41 @@ Examples:
             console.error(`Error: Invalid scan depth '${options.scanDepth}'. Use: ${validDepths.join(', ')}`);
             process.exit(1);
         }
-        // Deep mode: --deep flag OR --scan-depth deep
+        // Analysis mode: smart defaults, minimal flags
+        // Default: static + NanoMind (if daemon available)
+        // --deep: everything (static + NanoMind + simulation + adaptive attacks)
+        // --static-only: just static checks (CI/deterministic)
+        // --ci: implies --static-only
+        const isStaticOnly = options.staticOnly ?? false;
         const isDeep = options.deep ?? (scanDepth === 'deep');
-        const onProgress = isDeep && format === 'text'
+        // Auto-detect NanoMind daemon
+        let nanomindAvailable = false;
+        if (!isStaticOnly && !options.ci) {
+            try {
+                const { isDaemonAvailable } = await Promise.resolve().then(() => __importStar(require('./semantic/nanomind-analyzer.js')));
+                nanomindAvailable = await isDaemonAvailable();
+            }
+            catch { /* daemon not installed */ }
+        }
+        const onProgress = format === 'text'
             ? (msg) => process.stdout.write(msg)
             : undefined;
-        if (isDeep && format === 'text') {
-            if (!process.env.ANTHROPIC_API_KEY) {
-                console.log(`Layer 3: Semantic analysis — skipped (no ANTHROPIC_API_KEY)`);
-                console.log(`  Tip: Add HackMyAgent as an MCP server for free LLM analysis:`);
-                console.log(`  npx ${CLI_PREFIX} init-mcp\n`);
+        // Show analysis mode to user
+        if (format === 'text') {
+            if (isStaticOnly || options.ci) {
+                // Static only -- no extra output
+            }
+            else if (nanomindAvailable && isDeep) {
+                console.log(`Analysis: static + NanoMind + behavioral simulation + adaptive attacks\n`);
+            }
+            else if (nanomindAvailable) {
+                console.log(`Analysis: static + NanoMind (enhanced accuracy)\n`);
             }
+            else if (isDeep) {
+                console.log(`Analysis: static + behavioral simulation\n`);
+                console.log(`  Tip: Install NanoMind for even better results: nanomind-daemon start\n`);
+            }
+            // Default static-only: no message needed, it's the baseline
         }
         if (scanDepth === 'quick' && format === 'text') {
             console.log(`Scan depth: quick (config checks + credential detection only)\n`);
@@ -1845,6 +1870,53 @@ Examples:
             onProgress,
         });
         const scanDurationMs = Date.now() - scanStartMs;
+        // Behavioral simulation: auto-runs on --deep, or when NanoMind detects ambiguity
+        if (isDeep && format === 'text') {
+            try {
+                const { SimulationEngine, parseSkillProfile } = await Promise.resolve().then(() => __importStar(require('./simulation/index.js')));
+                const { readFileSync, readdirSync, statSync } = await Promise.resolve().then(() => __importStar(require('node:fs')));
+                const { join } = await Promise.resolve().then(() => __importStar(require('node:path')));
+                // Find skill files in target directory
+                const skillFiles = [];
+                const findSkills = (dir) => {
+                    try {
+                        for (const entry of readdirSync(dir)) {
+                            const fullPath = join(dir, entry);
+                            const stat = statSync(fullPath);
+                            if (stat.isDirectory() && !entry.startsWith('.') && entry !== 'node_modules') {
+                                findSkills(fullPath);
+                            }
+                            else if (entry.endsWith('.md') || entry.endsWith('.yaml') || entry.endsWith('.yml')) {
+                                skillFiles.push(fullPath);
+                            }
+                        }
+                    }
+                    catch { /* skip inaccessible dirs */ }
+                };
+                findSkills(targetDir);
+                if (skillFiles.length === 0) {
+                    process.stdout.write(`\n[Simulation] No skill/SOUL/MCP artifacts found. Simulation skipped.\n\n`);
+                }
+                else {
+                    process.stdout.write(`\n[Simulation] Running behavioral simulation on ${skillFiles.length} artifact(s)...\n`);
+                    const sim = new SimulationEngine({ useLLM: nanomindAvailable });
+                    for (const file of skillFiles.slice(0, 10)) { // Cap at 10 files
+                        const content = readFileSync(file, 'utf-8');
+                        const profile = parseSkillProfile(content, file.split('/').pop() ?? 'unknown');
+                        const simResult = await sim.runLayer3(profile);
+                        const icon = simResult.verdict === 'CLEAN' ? 'PASS' : simResult.verdict === 'SUSPICIOUS' ? 'WARN' : 'FAIL';
+                        process.stdout.write(`  [${icon}] ${file.split('/').pop()} — ${simResult.verdict} (${(simResult.confidence * 100).toFixed(0)}% confidence, ${simResult.failedProbes.length}/${simResult.probeCount} probes failed)\n`);
+                        // Auto-export training data
+                        const { exportSimulationTraining } = await Promise.resolve().then(() => __importStar(require('./attack-engine/training-pipeline.js')));
+                        exportSimulationTraining(content, simResult);
+                    }
+                    process.stdout.write(`[Simulation] Complete.\n\n`);
+                } // end skillFiles.length > 0
+            }
+            catch (err) {
+                process.stdout.write(`[Simulation] Skipped: ${err instanceof Error ? err.message : 'unknown error'}\n\n`);
+            }
+        }
         // OASB-2 composite mode: infrastructure (50%) + governance (50%)
         if (isOasb2) {
             const infraResult = generateBenchmarkReport(result.allFindings || result.findings, level, options.category);
@@ -4412,7 +4484,8 @@ Examples:
     .option('--tier <tier>', 'Override agent tier detection (BASIC, TOOL-USING, AGENTIC, MULTI-AGENT)')
     .option('--profile <profile>', 'Override agent profile (conversational, code-assistant, tool-agent, autonomous, orchestrator, custom)')
     .option('--fail-below <score>', 'Exit 1 if score below threshold (0-100)')
-    .option('--deep', 'Enable LLM semantic analysis for ambiguous controls (requires claude CLI or ANTHROPIC_API_KEY)')
+    .option('--deep', 'Maximum analysis: NanoMind + SOUL governance simulation (~15s)')
+    .option('--static-only', 'Disable NanoMind (static governance checks only)')
     .option('--publish', 'Push scan results to the OpenA2A Registry')
     .option('--registry-url <url>', 'Registry URL (default: REGISTRY_URL env)', validateRegistryUrl(process.env.REGISTRY_URL || 'https://api.oa2a.org'))
     .option('--contribute', 'Share anonymized scan findings with OpenA2A Registry (overrides config)')
@@ -5047,6 +5120,109 @@ program
     writeJsonStdout({ totalChecks: Object.keys(metadata).length, checks: metadata });
 });
 // Show help and exit 0 when no arguments provided
+// explain command: NanoMind-powered finding explanation
+program
+    .command('explain')
+    .argument('<findingId>', 'Finding ID to explain (e.g., SKILL-SEMANTIC-007 or CRED-001)')
+    .description('Explain a security finding in plain English using NanoMind')
+    .action(async (findingId) => {
+    console.log(`Explaining finding: ${findingId}\n`);
+    // Try NanoMind daemon first for dynamic explanation
+    const { isDaemonAvailable, explainFinding } = await Promise.resolve().then(() => __importStar(require('./semantic/nanomind-analyzer.js')));
+    const available = await isDaemonAvailable();
+    if (available) {
+        const explanation = await explainFinding(JSON.stringify({ findingId }));
+        if (explanation) {
+            console.log(explanation);
+            return;
+        }
+    }
+    // Fallback: static explanation from check metadata
+    const checkId = findingId.toUpperCase();
+    const staticExplanations = {
+        'CRED-001': 'Hardcoded credential detected. API keys, tokens, or passwords are embedded directly in source code. Replace with environment variable references ($VAR_NAME) and rotate the exposed credential immediately.',
+        'CRED-002': 'OpenAI API key pattern detected (sk-...). Move to environment variable OPENAI_API_KEY.',
+        'CRED-003': 'Anthropic API key pattern detected (sk-ant-...). Move to environment variable ANTHROPIC_API_KEY.',
+        'CRED-004': 'AWS credential pattern detected. Use AWS SDK credential chain or environment variables.',
+        'MCP-001': 'MCP server running without TLS. Agent-to-server communication is unencrypted.',
+        'SKILL-005': 'External endpoint in skill capability declaration. Verify the endpoint is trusted.',
+    };
+    const explanation = staticExplanations[checkId];
+    if (explanation) {
+        console.log(`${checkId}: ${explanation}`);
+    }
+    else {
+        console.log(`No explanation available for ${findingId}.`);
+        if (!available) {
+            console.log(`\nFor dynamic explanations, install NanoMind: npm install -g @nanomind/cli && nanomind-daemon start`);
+        }
+    }
+});
+// red-team command: NanoMind-powered adaptive attack engine
+program
+    .command('red-team')
+    .argument('<target>', 'Path to artifact to red-team (skill, SOUL.md, MCP config, system prompt)')
+    .description('Run adaptive attack session against an artifact. NanoMind generates target-specific attacks, observes responses, adapts, and maps defenses.')
+    .option('--iterations <n>', 'Max attack iterations per category', '5')
+    .option('--json', 'Output results as JSON')
+    .action(async (target, options) => {
+    const { readFileSync } = await Promise.resolve().then(() => __importStar(require('node:fs')));
+    const { runAttackSession, exportTrainingData } = await Promise.resolve().then(() => __importStar(require('./attack-engine/feedback-loop.js')));
+    const { exportAttackTraining } = await Promise.resolve().then(() => __importStar(require('./attack-engine/training-pipeline.js')));
+    let content;
+    try {
+        content = readFileSync(target, 'utf-8');
+    }
+    catch {
+        console.error(`Cannot read file: ${target}`);
+        process.exit(1);
+    }
+    const artifactType = target.toLowerCase().includes('soul') ? 'soul'
+        : target.toLowerCase().includes('mcp') ? 'mcp_tool'
+            : 'skill';
+    const name = target.split('/').pop() ?? 'unknown';
+    if (!options.json) {
+        console.log(`\nAdaptive Attack Engine`);
+        console.log(`Target: ${name} (${artifactType})`);
+        console.log(`Max iterations: ${options.iterations ?? 5} per category\n`);
+    }
+    const result = await runAttackSession(content, artifactType, name, {
+        maxIterations: parseInt(options.iterations ?? '5', 10),
+    });
+    if (options.json) {
+        console.log(JSON.stringify(result, null, 2));
+    }
+    else {
+        console.log(`Results:`);
+        console.log(`  Payloads generated: ${result.totalPayloads}`);
+        console.log(`  Successful attacks: ${result.successCount}`);
+        console.log(`  Partial successes:  ${result.partialCount}`);
+        console.log(`  Resilience score:   ${(result.defenseMap.resilienceScore * 100).toFixed(0)}%`);
+        console.log(`  Duration:           ${result.durationMs}ms\n`);
+        if (result.vulnerabilities.length > 0) {
+            console.log(`Vulnerabilities Found:`);
+            for (const vuln of result.vulnerabilities) {
+                console.log(`  [${vuln.severity.toUpperCase()}] ${vuln.title}`);
+                console.log(`    ${vuln.description}`);
+                console.log(`    Fix: ${vuln.remediation}\n`);
+            }
+        }
+        else {
+            console.log(`No vulnerabilities found. All defenses held.\n`);
+        }
+        if (result.defenseMap.strongCategories.length > 0) {
+            console.log(`Strong defenses: ${result.defenseMap.strongCategories.join(', ')}`);
+        }
+        if (result.defenseMap.weakCategories.length > 0) {
+            console.log(`Weak defenses:   ${result.defenseMap.weakCategories.join(', ')}`);
+        }
+    }
+    // Auto-export training data
+    const trainingCount = exportAttackTraining(result);
+    if (!options.json && trainingCount > 0) {
+        console.log(`\n${trainingCount} training samples exported to NanoMind corpus.`);
+    }
+});
 if (process.argv.length <= 2) {
     program.outputHelp();
     process.exit(0);