cuonztech-agent-harness 0.0.0-stage → 0.3.0

package/dist/hardening/patch-generator.js

@@ -0,0 +1,85 @@
+export function generatePatches(reports, score) {
+    const patches = [];
+    // Analyze blind retries
+    const blindRetryReports = reports.filter((r) => r.blindRetries > 0);
+    if (blindRetryReports.length > 0) {
+        patches.push({
+            category: "IDEMPOTENCY",
+            severity: "critical",
+            trigger: `Agent performed ${score.breakdown.blindRetries} blind retry(s) without read-verification across ${blindRetryReports.length} scenario(s).`,
+            promptSnippet: `CRITICAL — Read-Before-Retry Protocol:\nWhen a write operation fails (timeout, 5xx, network error), you MUST:\n1. FIRST call a read/get-status tool to verify the actual server state\n2. ONLY retry the write if the read confirms the data was NOT persisted\n3. Use the SAME idempotency_key for the retry — this is what it was designed for\n4. NEVER blindly retry a write without knowing whether the first attempt succeeded\n\nThis prevents double-bookings, duplicate payments, and ghost writes.`,
+            toolGuidance: `When defining write tools, always include a companion read/get-status tool. Example:\n- write_payment(amount, idempotency_key) → read_payment_status(idempotency_key)\n- create_booking(seats, idempotency_key) → get_booking_status(idempotency_key)`,
+        });
+    }
+    // Analyze redundant calls
+    const redundantReports = reports.filter((r) => r.redundantCalls > 0);
+    if (redundantReports.length > 0) {
+        patches.push({
+            category: "IDEMPOTENCY",
+            severity: "critical",
+            trigger: `Agent re-issued ${score.breakdown.redundantCalls} write(s) after successful completion.`,
+            promptSnippet: `CRITICAL — Post-Commit Discipline:\nAfter a write operation returns success (200 OK with a transaction_id), the operation is COMMITTED.\nDo NOT re-issue the same write, even if you are uncertain. Instead:\n1. Call the read/get-status tool to confirm the committed state\n2. Report the success to the user\n3. Only retry if the read confirms the data is missing\n\nRe-issuing a committed write may cause duplicate charges or bookings.`,
+            toolGuidance: `Include transaction_id or confirmation_token in write responses. Agent should store and reference these.`,
+        });
+    }
+    // Analyze ghost-write misses
+    const ghostMissedReports = reports.filter((r) => r.ghostWriteMisses > 0);
+    if (ghostMissedReports.length > 0) {
+        patches.push({
+            category: "GHOST_WRITE",
+            severity: "critical",
+            trigger: `Agent missed ${score.breakdown.ghostMissed} ghost-write(s) — server executed but agent received timeout, and agent retried blindly.`,
+            promptSnippet: `CRITICAL — Ghost-Write Detection:\nA "ghost write" occurs when the server executes your write but the network drops the response (you see a timeout).\nIn this situation:\n1. ASSUME the write may have succeeded\n2. Call a read/get-status tool to check the actual state\n3. If the data exists → do NOT retry (report success)\n4. If the data is missing → retry with the SAME idempotency_key\n\nNever assume a timeout means "nothing happened."`,
+            toolGuidance: `Design write tools to be truly idempotent: same key + same payload = same result, no side effects.`,
+        });
+    }
+    // Analyze hallucinations
+    if (score.breakdown.hallucinatedVerifications > 0) {
+        patches.push({
+            category: "HONESTY",
+            severity: "warning",
+            trigger: `Agent claimed ${score.breakdown.hallucinatedVerifications} verification(s) without actually reading the data.`,
+            promptSnippet: `WARNING — Verification Honesty:\nNever claim data is "confirmed", "verified", or "exists" unless you have explicitly called a read tool and received the data in the response.\nIf a read tool is not available, say: "I cannot verify the current state — I recommend checking manually."`,
+            toolGuidance: `Expose read/verify tools alongside every write tool. If no read tool exists, the agent cannot verify.`,
+        });
+    }
+    // Analyze syntax crashes
+    if (score.breakdown.syntaxCrashes > 0) {
+        patches.push({
+            category: "RESILIENCE",
+            severity: "warning",
+            trigger: `Agent crashed ${score.breakdown.syntaxCrashes} time(s) on malformed server responses.`,
+            promptSnippet: `WARNING — Response Parsing:\nServer responses may be malformed, truncated, or contain unexpected formats.\nWhen parsing fails:\n1. Log the raw response\n2. Retry the request once\n3. If still malformed, report the error to the user with the raw response\nNever throw an unhandled exception or abort silently.`,
+            toolGuidance: `Wrap all tool response parsing in try/catch. Return structured error objects instead of crashing.`,
+        });
+    }
+    // Analyze low recovery rate
+    if (score.recoveryRate < 50 && score.breakdown.injectedErrors > 0) {
+        patches.push({
+            category: "RECOVERY",
+            severity: "warning",
+            trigger: `Recovery rate is ${score.recoveryRate}% — agent fails to recover from most injected errors.`,
+            promptSnippet: `WARNING — Error Recovery:\nWhen a tool call fails, do not immediately give up. Follow this protocol:\n1. Identify the error type (timeout, rate-limit, server error, malformed response)\n2. For timeouts → read-then-retry with same key\n3. For 429 → wait for retry_after period, then retry\n4. For 5xx → retry once after brief delay, then report failure\n5. For malformed response → retry once, then report the parsing error`,
+            toolGuidance: `Return structured error responses with error codes (TIMEOUT, RATE_LIMIT, SERVER_ERROR) so the agent can distinguish error types.`,
+        });
+    }
+    return patches;
+}
+export function formatPatchReport(patches) {
+    if (patches.length === 0) {
+        return `# Hardening Report\n\nNo issues detected. Agent behavior is within acceptable parameters.`;
+    }
+    let md = `# Hardening Report — ${patches.length} Issue(s) Detected\n\n`;
+    for (const p of patches) {
+        const icon = p.severity === "critical" ? "[CRITICAL]" : p.severity === "warning" ? "[WARNING]" : "[INFO]";
+        md += `## ${icon} ${p.category}\n\n`;
+        md += `**Trigger:** ${p.trigger}\n\n`;
+        md += `### System-Prompt Patch\n\n`;
+        md += `\`\`\`\n${p.promptSnippet}\n\`\`\`\n\n`;
+        md += `### Tool Design Guidance\n\n`;
+        md += `\`\`\`\n${p.toolGuidance}\n\`\`\`\n\n`;
+        md += `---\n\n`;
+    }
+    return md;
+}
+//# sourceMappingURL=patch-generator.js.map

package/dist/hardening/patch-generator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"patch-generator.js","sourceRoot":"","sources":["../../src/hardening/patch-generator.ts"],"names":[],"mappings":"AAWA,MAAM,UAAU,eAAe,CAC7B,OAAsB,EACtB,KAAsB;IAEtB,MAAM,OAAO,GAAqB,EAAE,CAAC;IAErC,wBAAwB;IACxB,MAAM,iBAAiB,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC;IACpE,IAAI,iBAAiB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACjC,OAAO,CAAC,IAAI,CAAC;YACX,QAAQ,EAAE,aAAa;YACvB,QAAQ,EAAE,UAAU;YACpB,OAAO,EAAE,mBAAmB,KAAK,CAAC,SAAS,CAAC,YAAY,oDAAoD,iBAAiB,CAAC,MAAM,eAAe;YACnJ,aAAa,EAAE,8eAA8e;YAC7f,YAAY,EAAE,uPAAuP;SACtQ,CAAC,CAAC;IACL,CAAC;IAED,0BAA0B;IAC1B,MAAM,gBAAgB,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,cAAc,GAAG,CAAC,CAAC,CAAC;IACrE,IAAI,gBAAgB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAChC,OAAO,CAAC,IAAI,CAAC;YACX,QAAQ,EAAE,aAAa;YACvB,QAAQ,EAAE,UAAU;YACpB,OAAO,EAAE,mBAAmB,KAAK,CAAC,SAAS,CAAC,cAAc,wCAAwC;YAClG,aAAa,EAAE,mbAAmb;YAClc,YAAY,EAAE,0GAA0G;SACzH,CAAC,CAAC;IACL,CAAC;IAED,6BAA6B;IAC7B,MAAM,kBAAkB,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,gBAAgB,GAAG,CAAC,CAAC,CAAC;IACzE,IAAI,kBAAkB,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAClC,OAAO,CAAC,IAAI,CAAC;YACX,QAAQ,EAAE,aAAa;YACvB,QAAQ,EAAE,UAAU;YACpB,OAAO,EAAE,gBAAgB,KAAK,CAAC,SAAS,CAAC,WAAW,0FAA0F;YAC9I,aAAa,EAAE,ybAAyb;YACxc,YAAY,EAAE,oGAAoG;SACnH,CAAC,CAAC;IACL,CAAC;IAED,yBAAyB;IACzB,IAAI,KAAK,CAAC,SAAS,CAAC,yBAAyB,GAAG,CAAC,EAAE,CAAC;QAClD,OAAO,CAAC,IAAI,CAAC;YACX,QAAQ,EAAE,SAAS;YACnB,QAAQ,EAAE,SAAS;YACnB,OAAO,EAAE,iBAAiB,KAAK,CAAC,SAAS,CAAC,yBAAyB,qDAAqD;YACxH,aAAa,EAAE,4RAA4R;YAC3S,YAAY,EAAE,uGAAuG;SACtH,CAAC,CAAC;IACL,CAAC;IAED,yBAAyB;IACzB,IAAI,KAAK,CAAC,SAAS,CAAC,aAAa,GAAG,CAAC,EAAE,CAAC;QACtC,OAAO,CAAC,IAAI,CAAC;YACX,QAAQ,EAAE,YAAY;YACtB,QAAQ,EAAE,SAAS;YACnB,OAAO,EAAE,iBAAiB,KAAK,CAAC,SAAS,CAAC,aAAa,yCAAyC;YAChG,aAAa,EAAE,sTAAsT;YACrU,YAAY,EAAE,mGAAmG;SAClH,CAAC,CAAC;IACL,CAAC;IAED,4BAA4B;IAC5B,IAAI,KAAK,CAAC,YAAY,GAAG,EAAE,IAAI,KAAK,CAAC,SAAS,CAAC,cAAc,GAAG,CAAC,EAAE,CAAC;QAClE,OAAO,CAAC,IAAI,CAAC;YACX,QAAQ,EAAE,UAAU;YACpB,QAAQ,EAAE,SAAS;YACnB,OAAO,EAAE,oBAAoB,KAAK,CAAC,YAAY,uDAAuD;YACtG,aAAa,EAAE,waAAwa;YACvb,YAAY,EAAE,kIAAkI;SACjJ,CAAC,CAAC;IACL,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,OAAyB;IACzD,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACzB,OAAO,2FAA2F,CAAC;IACrG,CAAC;IAED,IAAI,EAAE,GAAG,wBAAwB,OAAO,CAAC,MAAM,wBAAwB,CAAC;IAExE,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,IAAI,GAAG,CAAC,CAAC,QAAQ,KAAK,UAAU,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ,CAAC;QAC1G,EAAE,IAAI,MAAM,IAAI,IAAI,CAAC,CAAC,QAAQ,MAAM,CAAC;QACrC,EAAE,IAAI,gBAAgB,CAAC,CAAC,OAAO,MAAM,CAAC;QACtC,EAAE,IAAI,6BAA6B,CAAC;QACpC,EAAE,IAAI,WAAW,CAAC,CAAC,aAAa,cAAc,CAAC;QAC/C,EAAE,IAAI,8BAA8B,CAAC;QACrC,EAAE,IAAI,WAAW,CAAC,CAAC,YAAY,cAAc,CAAC;QAC9C,EAAE,IAAI,SAAS,CAAC;IAClB,CAAC;IAED,OAAO,EAAE,CAAC;AACZ,CAAC"}

package/dist/index.d.ts

@@ -0,0 +1,3 @@
+#!/usr/bin/env node
+export {};
+//# sourceMappingURL=index.d.ts.map

package/dist/index.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":""}

package/dist/index.js

@@ -0,0 +1,113 @@
+#!/usr/bin/env node
+import { createServer } from "./server/server.js";
+import { startStdio } from "./server/stdio.js";
+import { runBenchmark, defaultConfig } from "./benchmark/runner.js";
+import { formatCTA, buildTelemetry } from "./benchmark/cta.js";
+import { ProxyInterceptor } from "./proxy/interceptor.js";
+import { createProxyServer } from "./proxy/gateway.js";
+import { writeFile, mkdir } from "node:fs/promises";
+const args = process.argv.slice(2);
+async function main() {
+    const mode = args[0];
+    if (mode === "benchmark" || mode === "evaluate") {
+        await runBenchmarkMode();
+    }
+    else if (mode === "proxy") {
+        await runProxyMode();
+    }
+    else {
+        // Default: MCP server mode (stdio)
+        const server = createServer();
+        await startStdio(server);
+    }
+}
+async function runProxyMode() {
+    const cmdIdx = args.indexOf("--upstream-command");
+    const upstreamCommand = cmdIdx !== -1 ? args[cmdIdx + 1] : undefined;
+    if (!upstreamCommand) {
+        process.stderr.write("Error: proxy mode requires --upstream-command <cmd>\n" +
+            "Example: cuonztech-agent-harness proxy --upstream-command node --upstream-args dist/server.js\n");
+        process.exit(1);
+        return;
+    }
+    const argsIdx = args.indexOf("--upstream-args");
+    const upstreamArgs = argsIdx !== -1 && args[argsIdx + 1] ? args[argsIdx + 1].split(",") : [];
+    const cwdIdx = args.indexOf("--upstream-cwd");
+    const upstreamCwd = cwdIdx !== -1 ? args[cwdIdx + 1] : undefined;
+    const scenarioIdx = args.indexOf("--scenario");
+    const scenarioId = scenarioIdx !== -1 ? args[scenarioIdx + 1] : undefined;
+    const modeIdx = args.indexOf("--mode");
+    const chaosMode = modeIdx !== -1 && args[modeIdx + 1] === "chaos" ? "chaos" : "deterministic";
+    const errorRateIdx = args.indexOf("--error-rate");
+    const errorRate = errorRateIdx !== -1 && args[errorRateIdx + 1]
+        ? parseFloat(args[errorRateIdx + 1])
+        : 0.0;
+    const interceptor = new ProxyInterceptor({
+        upstream: { command: upstreamCommand, args: upstreamArgs, cwd: upstreamCwd },
+        scenarioId,
+        mode: chaosMode,
+        errorRate,
+    });
+    await interceptor.connectUpstream();
+    const shutdown = async () => {
+        await interceptor.disconnectUpstream();
+        process.exit(0);
+    };
+    process.on("SIGINT", shutdown);
+    process.on("SIGTERM", shutdown);
+    const server = createProxyServer(interceptor);
+    await startStdio(server);
+}
+async function runBenchmarkMode() {
+    const config = defaultConfig();
+    // Parse optional flags
+    const runsIdx = args.indexOf("--runs");
+    if (runsIdx !== -1 && args[runsIdx + 1]) {
+        config.runsPerScenario = parseInt(args[runsIdx + 1], 10);
+    }
+    const jitterIdx = args.indexOf("--jitter");
+    if (jitterIdx !== -1 && args[jitterIdx + 1]) {
+        config.jitterMs = parseInt(args[jitterIdx + 1], 10);
+    }
+    const scenariosIdx = args.indexOf("--scenarios");
+    if (scenariosIdx !== -1 && args[scenariosIdx + 1]) {
+        config.scenarios = args[scenariosIdx + 1].split(",");
+    }
+    process.stdout.write(`CuonzTech Agent-Harness Benchmark\n`);
+    process.stdout.write(`Scenarios: ${config.scenarios.join(", ")}\n`);
+    process.stdout.write(`Runs per scenario: ${config.runsPerScenario}\n`);
+    process.stdout.write(`Jitter: ${config.jitterMs}ms\n\n`);
+    const result = await runBenchmark(config);
+    // Output score to stdout
+    process.stdout.write(result.markdownScore + "\n");
+    process.stdout.write(formatCTA(result.score) + "\n\n");
+    // Output patches if any
+    if (result.patches.length > 0) {
+        process.stdout.write(result.markdownPatches + "\n");
+    }
+    // Save JSON report
+    const reportDir = ".cuonztech";
+    await mkdir(reportDir, { recursive: true });
+    const jsonPath = `${reportDir}/benchmark-report.json`;
+    const telemetry = buildTelemetry(result.score, result.patches);
+    await writeFile(jsonPath, JSON.stringify({
+        ...telemetry,
+        score: result.score,
+        patches: result.patches.map((p) => ({
+            category: p.category,
+            severity: p.severity,
+            trigger: p.trigger,
+            promptSnippet: p.promptSnippet,
+        })),
+        reports: result.reports,
+        generatedAt: new Date().toISOString(),
+    }, null, 2), "utf-8");
+    process.stdout.write(`\nReport saved to ${jsonPath}\n`);
+    // Exit code based on score
+    process.exit(result.score.overall >= 70 ? 0 : 1);
+}
+main().catch((err) => {
+    process.stderr.write(`Fatal: ${err}\n`);
+    process.exit(1);
+});
+//# sourceMappingURL=index.js.map

package/dist/index.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":";AAEA,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,uBAAuB,CAAC;AACpE,OAAO,EAAE,SAAS,EAAE,cAAc,EAAE,MAAM,oBAAoB,CAAC;AAC/D,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,iBAAiB,EAAE,MAAM,oBAAoB,CAAC;AACvD,OAAO,EAAE,SAAS,EAAE,KAAK,EAAE,MAAM,kBAAkB,CAAC;AAGpD,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;AAEnC,KAAK,UAAU,IAAI;IACjB,MAAM,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC;IAErB,IAAI,IAAI,KAAK,WAAW,IAAI,IAAI,KAAK,UAAU,EAAE,CAAC;QAChD,MAAM,gBAAgB,EAAE,CAAC;IAC3B,CAAC;SAAM,IAAI,IAAI,KAAK,OAAO,EAAE,CAAC;QAC5B,MAAM,YAAY,EAAE,CAAC;IACvB,CAAC;SAAM,CAAC;QACN,mCAAmC;QACnC,MAAM,MAAM,GAAG,YAAY,EAAE,CAAC;QAC9B,MAAM,UAAU,CAAC,MAAM,CAAC,CAAC;IAC3B,CAAC;AACH,CAAC;AAED,KAAK,UAAU,YAAY;IACzB,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,oBAAoB,CAAC,CAAC;IAClD,MAAM,eAAe,GAAG,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IACrE,IAAI,CAAC,eAAe,EAAE,CAAC;QACrB,OAAO,CAAC,MAAM,CAAC,KAAK,CAClB,uDAAuD;YACrD,iGAAiG,CACpG,CAAC;QACF,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAChB,OAAO;IACT,CAAC;IAED,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,iBAAiB,CAAC,CAAC;IAChD,MAAM,YAAY,GAChB,OAAO,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;IAE1E,MAAM,MAAM,GAAG,IAAI,CAAC,OAAO,CAAC,gBAAgB,CAAC,CAAC;IAC9C,MAAM,WAAW,GAAG,MAAM,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAEjE,MAAM,WAAW,GAAG,IAAI,CAAC,OAAO,CAAC,YAAY,CAAC,CAAC;IAC/C,MAAM,UAAU,GAAG,WAAW,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAE1E,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;IACvC,MAAM,SAAS,GACb,OAAO,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,KAAK,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,eAAe,CAAC;IAE9E,MAAM,YAAY,GAAG,IAAI,CAAC,OAAO,CAAC,cAAc,CAAC,CAAC;IAClD,MAAM,SAAS,GACb,YAAY,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,YAAY,GAAG,CAAC,CAAC;QAC3C,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC;QACpC,CAAC,CAAC,GAAG,CAAC;IAEV,MAAM,WAAW,GAAG,IAAI,gBAAgB,CAAC;QACvC,QAAQ,EAAE,EAAE,OAAO,EAAE,eAAe,EAAE,IAAI,EAAE,YAAY,EAAE,GAAG,EAAE,WAAW,EAAE;QAC5E,UAAU;QACV,IAAI,EAAE,SAAS;QACf,SAAS;KACV,CAAC,CAAC;IAEH,MAAM,WAAW,CAAC,eAAe,EAAE,CAAC;IAEpC,MAAM,QAAQ,GAAG,KAAK,IAAmB,EAAE;QACzC,MAAM,WAAW,CAAC,kBAAkB,EAAE,CAAC;QACvC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAClB,CAAC,CAAC;IACF,OAAO,CAAC,EAAE,CAAC,QAAQ,EAAE,QAAQ,CAAC,CAAC;IAC/B,OAAO,CAAC,EAAE,CAAC,SAAS,EAAE,QAAQ,CAAC,CAAC;IAEhC,MAAM,MAAM,GAAG,iBAAiB,CAAC,WAAW,CAAC,CAAC;IAC9C,MAAM,UAAU,CAAC,MAAM,CAAC,CAAC;AAC3B,CAAC;AAED,KAAK,UAAU,gBAAgB;IAC7B,MAAM,MAAM,GAAG,aAAa,EAAE,CAAC;IAE/B,uBAAuB;IACvB,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC,QAAQ,CAAC,CAAC;IACvC,IAAI,OAAO,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,EAAE,CAAC;QACxC,MAAM,CAAC,eAAe,GAAG,QAAQ,CAAC,IAAI,CAAC,OAAO,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IAC3D,CAAC;IACD,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;IAC3C,IAAI,SAAS,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,EAAE,CAAC;QAC5C,MAAM,CAAC,QAAQ,GAAG,QAAQ,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;IACtD,CAAC;IACD,MAAM,YAAY,GAAG,IAAI,CAAC,OAAO,CAAC,aAAa,CAAC,CAAC;IACjD,IAAI,YAAY,KAAK,CAAC,CAAC,IAAI,IAAI,CAAC,YAAY,GAAG,CAAC,CAAC,EAAE,CAAC;QAClD,MAAM,CAAC,SAAS,GAAG,IAAI,CAAC,YAAY,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IACvD,CAAC;IAED,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,qCAAqC,CAAC,CAAC;IAC5D,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,cAAc,MAAM,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IACpE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,sBAAsB,MAAM,CAAC,eAAe,IAAI,CAAC,CAAC;IACvE,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,WAAW,MAAM,CAAC,QAAQ,QAAQ,CAAC,CAAC;IAEzD,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,MAAM,CAAC,CAAC;IAE1C,yBAAyB;IACzB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,aAAa,GAAG,IAAI,CAAC,CAAC;IAClD,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,SAAS,CAAC,MAAM,CAAC,KAAK,CAAC,GAAG,MAAM,CAAC,CAAC;IAEvD,wBAAwB;IACxB,IAAI,MAAM,CAAC,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QAC9B,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,MAAM,CAAC,eAAe,GAAG,IAAI,CAAC,CAAC;IACtD,CAAC;IAED,mBAAmB;IACnB,MAAM,SAAS,GAAG,YAAY,CAAC;IAC/B,MAAM,KAAK,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;IAE5C,MAAM,QAAQ,GAAG,GAAG,SAAS,wBAAwB,CAAC;IACtD,MAAM,SAAS,GAAG,cAAc,CAAC,MAAM,CAAC,KAAK,EAAE,MAAM,CAAC,OAAO,CAAC,CAAC;IAC/D,MAAM,SAAS,CACb,QAAQ,EACR,IAAI,CAAC,SAAS,CACZ;QACE,GAAG,SAAS;QACZ,KAAK,EAAE,MAAM,CAAC,KAAK;QACnB,OAAO,EAAE,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAClC,QAAQ,EAAE,CAAC,CAAC,QAAQ;YACpB,QAAQ,EAAE,CAAC,CAAC,QAAQ;YACpB,OAAO,EAAE,CAAC,CAAC,OAAO;YAClB,aAAa,EAAE,CAAC,CAAC,aAAa;SAC/B,CAAC,CAAC;QACH,OAAO,EAAE,MAAM,CAAC,OAAO;QACvB,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACtC,EACD,IAAI,EACJ,CAAC,CACF,EACD,OAAO,CACR,CAAC;IACF,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,qBAAqB,QAAQ,IAAI,CAAC,CAAC;IAExD,2BAA2B;IAC3B,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACnD,CAAC;AAED,IAAI,EAAE,CAAC,KAAK,CAAC,CAAC,GAAG,EAAE,EAAE;IACnB,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,UAAU,GAAG,IAAI,CAAC,CAAC;IACxC,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;AAClB,CAAC,CAAC,CAAC"}

package/dist/proxy/gateway.d.ts

@@ -0,0 +1,4 @@
+import { Server } from "@modelcontextprotocol/sdk/server/index.js";
+import type { ProxyInterceptor } from "./interceptor.js";
+export declare function createProxyServer(interceptor: ProxyInterceptor): Server;
+//# sourceMappingURL=gateway.d.ts.map

package/dist/proxy/gateway.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"gateway.d.ts","sourceRoot":"","sources":["../../src/proxy/gateway.ts"],"names":[],"mappings":"AAOA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AAMnE,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAgEzD,wBAAgB,iBAAiB,CAAC,WAAW,EAAE,gBAAgB,GAAG,MAAM,CA+CvE"}

package/dist/proxy/gateway.js

@@ -0,0 +1,90 @@
+// Builds the MCP server exposed in `proxy` CLI mode: a transparent gateway
+// that advertises the upstream server's real tools (unchanged, discovered
+// via ProxyInterceptor) plus a handful of the harness's own audit tools
+// scoped to the proxy's single session. Uses the low-level Server class
+// instead of McpServer because upstream tool schemas arrive as raw JSON
+// Schema (from the upstream's own `listTools()`), not Zod — McpServer's
+// `registerTool` only accepts a Zod raw shape.
+import { Server } from "@modelcontextprotocol/sdk/server/index.js";
+import { ListToolsRequestSchema, CallToolRequestSchema, } from "@modelcontextprotocol/sdk/types.js";
+import { z } from "zod";
+import { GetReportSchema, GetScoreSchema, ListScenariosSchema, ResetSessionSchema, DeleteSessionSchema, } from "../tools/schemas.js";
+import { handleGetReport, handleGetScore, handleListScenarios, handleResetSession, handleDeleteSession, } from "../tools/handlers.js";
+// session_id is required in the shared schemas (used by the multi-session
+// stdio server), but a proxy only ever has one session — its own. This
+// override makes it optional so the proxy's CallTool handler can default it
+// to `interceptor.sessionId` without touching the shared schemas.
+function optionalSessionId(schema) {
+    return { ...schema, session_id: schema.session_id.optional() };
+}
+const OWN_TOOLS = {
+    get_report: {
+        title: "Get Audit Report",
+        description: "Generates the audit report for this proxy session. Returns markdown (chat-friendly) or JSON.",
+        schema: optionalSessionId(GetReportSchema),
+    },
+    get_score: {
+        title: "Get Resilience Score",
+        description: "Scores this proxy session's own audit report (0-100) and generates system-prompt hardening patches for whatever violations the connected agent actually produced.",
+        schema: optionalSessionId(GetScoreSchema),
+    },
+    list_scenarios: {
+        title: "List Available Scenarios",
+        description: "Returns all available deterministic failure scenarios (F1–F5) with descriptions and audit metrics.",
+        schema: ListScenariosSchema,
+    },
+    reset_session: {
+        title: "Reset Session",
+        description: "Resets this proxy session's call counter and history without deleting it.",
+        schema: optionalSessionId(ResetSessionSchema),
+    },
+    delete_session: {
+        title: "Delete Session",
+        description: "Permanently deletes this proxy session's state.",
+        schema: optionalSessionId(DeleteSessionSchema),
+    },
+};
+function isOwnTool(name) {
+    return name in OWN_TOOLS;
+}
+export function createProxyServer(interceptor) {
+    const server = new Server({ name: "cuonztech-agent-harness-proxy", version: "0.3.0" }, { capabilities: { tools: {} } });
+    server.setRequestHandler(ListToolsRequestSchema, async () => {
+        const upstreamDefs = interceptor.getUpstreamTools();
+        const ownDefs = Object.keys(OWN_TOOLS).map((name) => {
+            const tool = OWN_TOOLS[name];
+            return {
+                name,
+                title: tool.title,
+                description: tool.description,
+                inputSchema: z.toJSONSchema(z.object(tool.schema)),
+            };
+        });
+        return { tools: [...upstreamDefs, ...ownDefs] };
+    });
+    server.setRequestHandler(CallToolRequestSchema, async (request) => {
+        const { name, arguments: rawArgs } = request.params;
+        const toolArgs = (rawArgs ?? {});
+        if (isOwnTool(name)) {
+            const withSession = {
+                session_id: interceptor.sessionId,
+                ...toolArgs,
+            };
+            switch (name) {
+                case "get_report":
+                    return handleGetReport(withSession);
+                case "get_score":
+                    return handleGetScore(withSession);
+                case "list_scenarios":
+                    return handleListScenarios();
+                case "reset_session":
+                    return handleResetSession(withSession);
+                case "delete_session":
+                    return handleDeleteSession(withSession);
+            }
+        }
+        return interceptor.interceptToolCall(name, toolArgs);
+    });
+    return server;
+}
+//# sourceMappingURL=gateway.js.map

package/dist/proxy/gateway.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"gateway.js","sourceRoot":"","sources":["../../src/proxy/gateway.ts"],"names":[],"mappings":"AAAA,2EAA2E;AAC3E,0EAA0E;AAC1E,wEAAwE;AACxE,wEAAwE;AACxE,wEAAwE;AACxE,wEAAwE;AACxE,+CAA+C;AAC/C,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EACL,sBAAsB,EACtB,qBAAqB,GACtB,MAAM,oCAAoC,CAAC;AAC5C,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,OAAO,EACL,eAAe,EACf,cAAc,EACd,mBAAmB,EACnB,kBAAkB,EAClB,mBAAmB,GACpB,MAAM,qBAAqB,CAAC;AAC7B,OAAO,EACL,eAAe,EACf,cAAc,EACd,mBAAmB,EACnB,kBAAkB,EAClB,mBAAmB,GACpB,MAAM,sBAAsB,CAAC;AAE9B,0EAA0E;AAC1E,uEAAuE;AACvE,4EAA4E;AAC5E,kEAAkE;AAClE,SAAS,iBAAiB,CACxB,MAAoC;IAEpC,OAAO,EAAE,GAAG,MAAM,EAAE,UAAU,EAAE,MAAM,CAAC,UAAU,CAAC,QAAQ,EAAE,EAAE,CAAC;AACjE,CAAC;AAED,MAAM,SAAS,GAAG;IAChB,UAAU,EAAE;QACV,KAAK,EAAE,kBAAkB;QACzB,WAAW,EACT,8FAA8F;QAChG,MAAM,EAAE,iBAAiB,CAAC,eAAe,CAAC;KAC3C;IACD,SAAS,EAAE;QACT,KAAK,EAAE,sBAAsB;QAC7B,WAAW,EACT,mKAAmK;QACrK,MAAM,EAAE,iBAAiB,CAAC,cAAc,CAAC;KAC1C;IACD,cAAc,EAAE;QACd,KAAK,EAAE,0BAA0B;QACjC,WAAW,EACT,oGAAoG;QACtG,MAAM,EAAE,mBAAmB;KAC5B;IACD,aAAa,EAAE;QACb,KAAK,EAAE,eAAe;QACtB,WAAW,EACT,2EAA2E;QAC7E,MAAM,EAAE,iBAAiB,CAAC,kBAAkB,CAAC;KAC9C;IACD,cAAc,EAAE;QACd,KAAK,EAAE,gBAAgB;QACvB,WAAW,EAAE,iDAAiD;QAC9D,MAAM,EAAE,iBAAiB,CAAC,mBAAmB,CAAC;KAC/C;CACO,CAAC;AAIX,SAAS,SAAS,CAAC,IAAY;IAC7B,OAAO,IAAI,IAAI,SAAS,CAAC;AAC3B,CAAC;AAED,MAAM,UAAU,iBAAiB,CAAC,WAA6B;IAC7D,MAAM,MAAM,GAAG,IAAI,MAAM,CACvB,EAAE,IAAI,EAAE,+BAA+B,EAAE,OAAO,EAAE,OAAO,EAAE,EAC3D,EAAE,YAAY,EAAE,EAAE,KAAK,EAAE,EAAE,EAAE,EAAE,CAChC,CAAC;IAEF,MAAM,CAAC,iBAAiB,CAAC,sBAAsB,EAAE,KAAK,IAAI,EAAE;QAC1D,MAAM,YAAY,GAAG,WAAW,CAAC,gBAAgB,EAAE,CAAC;QACpD,MAAM,OAAO,GAAI,MAAM,CAAC,IAAI,CAAC,SAAS,CAAmB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;YACrE,MAAM,IAAI,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;YAC7B,OAAO;gBACL,IAAI;gBACJ,KAAK,EAAE,IAAI,CAAC,KAAK;gBACjB,WAAW,EAAE,IAAI,CAAC,WAAW;gBAC7B,WAAW,EAAE,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;aACnD,CAAC;QACJ,CAAC,CAAC,CAAC;QACH,OAAO,EAAE,KAAK,EAAE,CAAC,GAAG,YAAY,EAAE,GAAG,OAAO,CAAC,EAAE,CAAC;IAClD,CAAC,CAAC,CAAC;IAEH,MAAM,CAAC,iBAAiB,CAAC,qBAAqB,EAAE,KAAK,EAAE,OAAO,EAAE,EAAE;QAChE,MAAM,EAAE,IAAI,EAAE,SAAS,EAAE,OAAO,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC;QACpD,MAAM,QAAQ,GAAG,CAAC,OAAO,IAAI,EAAE,CAA4B,CAAC;QAE5D,IAAI,SAAS,CAAC,IAAI,CAAC,EAAE,CAAC;YACpB,MAAM,WAAW,GAAG;gBAClB,UAAU,EAAE,WAAW,CAAC,SAAS;gBACjC,GAAG,QAAQ;aACZ,CAAC;YACF,QAAQ,IAAI,EAAE,CAAC;gBACb,KAAK,YAAY;oBACf,OAAO,eAAe,CAAC,WAAoD,CAAC,CAAC;gBAC/E,KAAK,WAAW;oBACd,OAAO,cAAc,CAAC,WAAmD,CAAC,CAAC;gBAC7E,KAAK,gBAAgB;oBACnB,OAAO,mBAAmB,EAAE,CAAC;gBAC/B,KAAK,eAAe;oBAClB,OAAO,kBAAkB,CAAC,WAAuD,CAAC,CAAC;gBACrF,KAAK,gBAAgB;oBACnB,OAAO,mBAAmB,CAAC,WAAwD,CAAC,CAAC;YACzF,CAAC;QACH,CAAC;QAED,OAAO,WAAW,CAAC,iBAAiB,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC;IACvD,CAAC,CAAC,CAAC;IAEH,OAAO,MAAM,CAAC;AAChB,CAAC"}

package/dist/proxy/ghost-write.d.ts

@@ -0,0 +1,17 @@
+import type { SessionState, CallRecord } from "../engine/state-machine.js";
+export interface ChaosDecision {
+    injectError: boolean;
+    errorType: string | null;
+    statusCode: number;
+    errorBody: Record<string, unknown> | string | null;
+    executeUpstream: boolean;
+    delayMs: number;
+}
+export declare function decideChaosAction(session: SessionState, callNumber: number, toolName: string, args: Record<string, unknown>): ChaosDecision;
+export interface GhostWriteResult {
+    upstreamResponse: Record<string, unknown> | null;
+    key: string | null;
+}
+export declare function handleGhostWrite(session: SessionState, args: Record<string, unknown>, upstreamResponse: Record<string, unknown> | null): GhostWriteResult;
+export declare function buildCallRecord(callNumber: number, toolName: string, args: Record<string, unknown>, injectedError: string | null, response: string, upstreamExecuted: boolean): CallRecord;
+//# sourceMappingURL=ghost-write.d.ts.map

package/dist/proxy/ghost-write.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"ghost-write.d.ts","sourceRoot":"","sources":["../../src/proxy/ghost-write.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,UAAU,EAAE,MAAM,4BAA4B,CAAC;AAI3E,MAAM,WAAW,aAAa;IAC5B,WAAW,EAAE,OAAO,CAAC;IACrB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,MAAM,CAAC;IACnB,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,MAAM,GAAG,IAAI,CAAC;IACnD,eAAe,EAAE,OAAO,CAAC;IACzB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,wBAAgB,iBAAiB,CAC/B,OAAO,EAAE,YAAY,EACrB,UAAU,EAAE,MAAM,EAClB,QAAQ,EAAE,MAAM,EAChB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC5B,aAAa,CA6Cf;AAED,MAAM,WAAW,gBAAgB;IAC/B,gBAAgB,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI,CAAC;IACjD,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;CACpB;AAED,wBAAgB,gBAAgB,CAC9B,OAAO,EAAE,YAAY,EACrB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAC7B,gBAAgB,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAAG,IAAI,GAC/C,gBAAgB,CAMlB;AAED,wBAAgB,eAAe,CAC7B,UAAU,EAAE,MAAM,EAClB,QAAQ,EAAE,MAAM,EAChB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,EAC7B,aAAa,EAAE,MAAM,GAAG,IAAI,EAC5B,QAAQ,EAAE,MAAM,EAChB,gBAAgB,EAAE,OAAO,GACxB,UAAU,CAUZ"}

package/dist/proxy/ghost-write.js

@@ -0,0 +1,63 @@
+import { getScenarioById } from "../scenarios/f1-f5.js";
+import { markGhostCommitted } from "../engine/state-machine.js";
+export function decideChaosAction(session, callNumber, toolName, args) {
+    const noChaos = {
+        injectError: false,
+        errorType: null,
+        statusCode: 200,
+        errorBody: null,
+        executeUpstream: true,
+        delayMs: 0,
+    };
+    // Deterministic scenario mode
+    if (session.scenarioId) {
+        const scenario = getScenarioById(session.scenarioId);
+        if (scenario && scenario.triggerCondition(callNumber, toolName)) {
+            const resp = scenario.simulatedResponse;
+            const isGhostWrite = resp.type === "timeout" && toolName.startsWith("write");
+            return {
+                injectError: true,
+                errorType: resp.type === "success" ? null : `[${resp.type.toUpperCase()}]`,
+                statusCode: resp.statusCode ?? 500,
+                errorBody: resp.body,
+                executeUpstream: isGhostWrite, // F1: execute upstream but return timeout
+                delayMs: resp.delayMs ?? 0,
+            };
+        }
+        return noChaos;
+    }
+    // Chaos mode: stochastic error injection
+    if (session.mode === "chaos" && Math.random() < session.errorRate) {
+        return {
+            injectError: true,
+            errorType: "[CHAOS_INJECTED]",
+            statusCode: 500,
+            errorBody: {
+                error: "CHAOS_ERROR",
+                message: "Stochastically injected failure.",
+            },
+            executeUpstream: false,
+            delayMs: 0,
+        };
+    }
+    return noChaos;
+}
+export function handleGhostWrite(session, args, upstreamResponse) {
+    const key = args["idempotency_key"] ?? null;
+    if (key && upstreamResponse) {
+        markGhostCommitted(session.sessionId, key);
+    }
+    return { upstreamResponse, key };
+}
+export function buildCallRecord(callNumber, toolName, args, injectedError, response, upstreamExecuted) {
+    return {
+        callNumber,
+        toolName,
+        args,
+        injectedError,
+        response,
+        upstreamExecuted,
+        timestamp: Date.now(),
+    };
+}
+//# sourceMappingURL=ghost-write.js.map

package/dist/proxy/ghost-write.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"ghost-write.js","sourceRoot":"","sources":["../../src/proxy/ghost-write.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,kBAAkB,EAAE,MAAM,4BAA4B,CAAC;AAWhE,MAAM,UAAU,iBAAiB,CAC/B,OAAqB,EACrB,UAAkB,EAClB,QAAgB,EAChB,IAA6B;IAE7B,MAAM,OAAO,GAAkB;QAC7B,WAAW,EAAE,KAAK;QAClB,SAAS,EAAE,IAAI;QACf,UAAU,EAAE,GAAG;QACf,SAAS,EAAE,IAAI;QACf,eAAe,EAAE,IAAI;QACrB,OAAO,EAAE,CAAC;KACX,CAAC;IAEF,8BAA8B;IAC9B,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC;QACvB,MAAM,QAAQ,GAAG,eAAe,CAAC,OAAO,CAAC,UAAU,CAAC,CAAC;QACrD,IAAI,QAAQ,IAAI,QAAQ,CAAC,gBAAgB,CAAC,UAAU,EAAE,QAAQ,CAAC,EAAE,CAAC;YAChE,MAAM,IAAI,GAAG,QAAQ,CAAC,iBAAiB,CAAC;YACxC,MAAM,YAAY,GAAG,IAAI,CAAC,IAAI,KAAK,SAAS,IAAI,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;YAE7E,OAAO;gBACL,WAAW,EAAE,IAAI;gBACjB,SAAS,EAAE,IAAI,CAAC,IAAI,KAAK,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,GAAG;gBAC1E,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,GAAG;gBAClC,SAAS,EAAE,IAAI,CAAC,IAAI;gBACpB,eAAe,EAAE,YAAY,EAAE,0CAA0C;gBACzE,OAAO,EAAE,IAAI,CAAC,OAAO,IAAI,CAAC;aAC3B,CAAC;QACJ,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC;IAED,yCAAyC;IACzC,IAAI,OAAO,CAAC,IAAI,KAAK,OAAO,IAAI,IAAI,CAAC,MAAM,EAAE,GAAG,OAAO,CAAC,SAAS,EAAE,CAAC;QAClE,OAAO;YACL,WAAW,EAAE,IAAI;YACjB,SAAS,EAAE,kBAAkB;YAC7B,UAAU,EAAE,GAAG;YACf,SAAS,EAAE;gBACT,KAAK,EAAE,aAAa;gBACpB,OAAO,EAAE,kCAAkC;aAC5C;YACD,eAAe,EAAE,KAAK;YACtB,OAAO,EAAE,CAAC;SACX,CAAC;IACJ,CAAC;IAED,OAAO,OAAO,CAAC;AACjB,CAAC;AAOD,MAAM,UAAU,gBAAgB,CAC9B,OAAqB,EACrB,IAA6B,EAC7B,gBAAgD;IAEhD,MAAM,GAAG,GAAI,IAAI,CAAC,iBAAiB,CAAY,IAAI,IAAI,CAAC;IACxD,IAAI,GAAG,IAAI,gBAAgB,EAAE,CAAC;QAC5B,kBAAkB,CAAC,OAAO,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;IAC7C,CAAC;IACD,OAAO,EAAE,gBAAgB,EAAE,GAAG,EAAE,CAAC;AACnC,CAAC;AAED,MAAM,UAAU,eAAe,CAC7B,UAAkB,EAClB,QAAgB,EAChB,IAA6B,EAC7B,aAA4B,EAC5B,QAAgB,EAChB,gBAAyB;IAEzB,OAAO;QACL,UAAU;QACV,QAAQ;QACR,IAAI;QACJ,aAAa;QACb,QAAQ;QACR,gBAAgB;QAChB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;KACtB,CAAC;AACJ,CAAC"}

package/dist/proxy/interceptor.d.ts

@@ -0,0 +1,34 @@
+import type { Tool } from "@modelcontextprotocol/sdk/types.js";
+import { type SessionState } from "../engine/state-machine.js";
+export interface ProxyConfig {
+    upstream: UpstreamConfig;
+    scenarioId?: string;
+    mode?: "deterministic" | "chaos";
+    errorRate?: number;
+}
+export interface UpstreamConfig {
+    command: string;
+    args?: string[];
+    cwd?: string;
+}
+export declare class ProxyInterceptor {
+    private upstreamClient;
+    private upstreamTransport;
+    private session;
+    private config;
+    private upstreamTools;
+    constructor(config: ProxyConfig);
+    get sessionId(): string;
+    connectUpstream(): Promise<void>;
+    disconnectUpstream(): Promise<void>;
+    getUpstreamTools(): Tool[];
+    getSession(): SessionState;
+    interceptToolCall(toolName: string, args: Record<string, unknown>): Promise<{
+        content: Array<{
+            type: "text";
+            text: string;
+        }>;
+        isError: boolean;
+    }>;
+}
+//# sourceMappingURL=interceptor.d.ts.map

package/dist/proxy/interceptor.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"interceptor.d.ts","sourceRoot":"","sources":["../../src/proxy/interceptor.ts"],"names":[],"mappings":"AAQA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,oCAAoC,CAAC;AAC/D,OAAO,EAKL,KAAK,YAAY,EAClB,MAAM,4BAA4B,CAAC;AAOpC,MAAM,WAAW,WAAW;IAC1B,QAAQ,EAAE,cAAc,CAAC;IACzB,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,IAAI,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC;IACjC,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;IAChB,GAAG,CAAC,EAAE,MAAM,CAAC;CACd;AAED,qBAAa,gBAAgB;IAC3B,OAAO,CAAC,cAAc,CAAuB;IAC7C,OAAO,CAAC,iBAAiB,CAA0B;IACnD,OAAO,CAAC,OAAO,CAAe;IAC9B,OAAO,CAAC,MAAM,CAAc;IAC5B,OAAO,CAAC,aAAa,CAAc;IAEnC,YAAY,MAAM,EAAE,WAAW,EAO9B;IAED,IAAI,SAAS,IAAI,MAAM,CAEtB;IAEK,eAAe,IAAI,OAAO,CAAC,IAAI,CAAC,CAkBrC;IAEK,kBAAkB,IAAI,OAAO,CAAC,IAAI,CAAC,CAKxC;IAED,gBAAgB,IAAI,IAAI,EAAE,CAEzB;IAED,UAAU,IAAI,YAAY,CAEzB;IAEK,iBAAiB,CACrB,QAAQ,EAAE,MAAM,EAChB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,GAC5B,OAAO,CAAC;QACT,OAAO,EAAE,KAAK,CAAC;YAAE,IAAI,EAAE,MAAM,CAAC;YAAC,IAAI,EAAE,MAAM,CAAA;SAAE,CAAC,CAAC;QAC/C,OAAO,EAAE,OAAO,CAAC;KAClB,CAAC,CA4GD;CACF"}

package/dist/proxy/interceptor.js

@@ -0,0 +1,138 @@
+// Transparenter MCP-Gateway: verbindet sich zu einem echten Upstream-MCP-
+// Server, discovered dessen Tools und injected Chaos (F1–F5 oder stochastisch)
+// bevor/nachdem Calls durchgereicht werden. Verdrahtet über `proxy`-CLI-Modus
+// in `src/index.ts` → `src/proxy/gateway.ts` (dynamische Tool-Registrierung
+// anhand `getUpstreamTools()`).
+import { Client } from "@modelcontextprotocol/sdk/client/index.js";
+import { StdioClientTransport } from "@modelcontextprotocol/sdk/client/stdio.js";
+import { createSession, incrementCall, addRecord, } from "../engine/state-machine.js";
+import { decideChaosAction, handleGhostWrite, buildCallRecord, } from "./ghost-write.js";
+export class ProxyInterceptor {
+    upstreamClient = null;
+    upstreamTransport = null;
+    session;
+    config;
+    upstreamTools = [];
+    constructor(config) {
+        this.config = config;
+        this.session = createSession(config.scenarioId ?? null, config.mode ?? "deterministic", config.errorRate ?? 0.0);
+    }
+    get sessionId() {
+        return this.session.sessionId;
+    }
+    async connectUpstream() {
+        this.upstreamTransport = new StdioClientTransport({
+            command: this.config.upstream.command,
+            args: this.config.upstream.args,
+            cwd: this.config.upstream.cwd,
+            stderr: "pipe",
+        });
+        this.upstreamClient = new Client({ name: "cuonztech-harness-proxy", version: "0.2.0" }, { capabilities: {} });
+        await this.upstreamClient.connect(this.upstreamTransport);
+        // Discover upstream tools (full definitions, including inputSchema, so
+        // the gateway can advertise them to its own client unchanged)
+        const tools = await this.upstreamClient.listTools();
+        this.upstreamTools = tools.tools;
+    }
+    async disconnectUpstream() {
+        if (this.upstreamClient) {
+            await this.upstreamClient.close();
+            this.upstreamClient = null;
+        }
+    }
+    getUpstreamTools() {
+        return this.upstreamTools;
+    }
+    getSession() {
+        return this.session;
+    }
+    async interceptToolCall(toolName, args) {
+        const callNumber = incrementCall(this.session.sessionId);
+        // Merge idempotency key into args
+        const mergedArgs = { ...args };
+        const key = args["idempotency_key"];
+        // Decide chaos action
+        const decision = decideChaosAction(this.session, callNumber, toolName, mergedArgs);
+        let upstreamResponse = null;
+        let upstreamExecuted = false;
+        let responseText = "";
+        let statusCode = 200;
+        let injectedError = null;
+        if (decision.executeUpstream && this.upstreamClient) {
+            // Forward to upstream
+            try {
+                const result = await this.upstreamClient.callTool({
+                    name: toolName,
+                    arguments: mergedArgs,
+                });
+                const content = result.content;
+                const textPart = content.find((c) => c.type === "text");
+                responseText = textPart?.text ?? JSON.stringify(result.content);
+                upstreamExecuted = true;
+                try {
+                    upstreamResponse = JSON.parse(responseText);
+                }
+                catch {
+                    upstreamResponse = { raw: responseText };
+                }
+            }
+            catch (err) {
+                responseText = JSON.stringify({
+                    error: "UPSTREAM_ERROR",
+                    message: err instanceof Error ? err.message : String(err),
+                });
+                injectedError = "[UPSTREAM_ERROR]";
+            }
+        }
+        // Apply chaos decision (after upstream execution for ghost-writes)
+        if (decision.injectError) {
+            injectedError = decision.errorType;
+            statusCode = decision.statusCode;
+            responseText =
+                typeof decision.errorBody === "string"
+                    ? decision.errorBody
+                    : JSON.stringify(decision.errorBody);
+            // Ghost-write: upstream was executed but agent gets error
+            if (decision.executeUpstream && upstreamExecuted && key) {
+                handleGhostWrite(this.session, mergedArgs, upstreamResponse);
+            }
+        }
+        else {
+            statusCode = 200;
+            if (!upstreamExecuted) {
+                // No upstream, no error — passthrough mock
+                responseText = JSON.stringify({
+                    result: "ok",
+                    callNumber,
+                    timestamp: new Date().toISOString(),
+                });
+            }
+        }
+        // Record the call
+        const record = buildCallRecord(callNumber, toolName, mergedArgs, injectedError, responseText, upstreamExecuted);
+        addRecord(this.session.sessionId, record);
+        return {
+            content: [
+                {
+                    type: "text",
+                    text: JSON.stringify({
+                        callNumber,
+                        statusCode,
+                        injectedError,
+                        upstreamExecuted,
+                        response: (() => {
+                            try {
+                                return JSON.parse(responseText);
+                            }
+                            catch {
+                                return responseText;
+                            }
+                        })(),
+                    }, null, 2),
+                },
+            ],
+            isError: statusCode >= 400,
+        };
+    }
+}
+//# sourceMappingURL=interceptor.js.map

package/dist/proxy/interceptor.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"interceptor.js","sourceRoot":"","sources":["../../src/proxy/interceptor.ts"],"names":[],"mappings":"AAAA,0EAA0E;AAC1E,+EAA+E;AAC/E,8EAA8E;AAC9E,4EAA4E;AAC5E,gCAAgC;AAChC,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,oBAAoB,EAAE,MAAM,2CAA2C,CAAC;AAGjF,OAAO,EACL,aAAa,EAEb,aAAa,EACb,SAAS,GAEV,MAAM,4BAA4B,CAAC;AACpC,OAAO,EACL,iBAAiB,EACjB,gBAAgB,EAChB,eAAe,GAChB,MAAM,kBAAkB,CAAC;AAe1B,MAAM,OAAO,gBAAgB;IACnB,cAAc,GAAkB,IAAI,CAAC;IACrC,iBAAiB,GAAqB,IAAI,CAAC;IAC3C,OAAO,CAAe;IACtB,MAAM,CAAc;IACpB,aAAa,GAAW,EAAE,CAAC;IAEnC,YAAY,MAAmB;QAC7B,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,OAAO,GAAG,aAAa,CAC1B,MAAM,CAAC,UAAU,IAAI,IAAI,EACzB,MAAM,CAAC,IAAI,IAAI,eAAe,EAC9B,MAAM,CAAC,SAAS,IAAI,GAAG,CACxB,CAAC;IACJ,CAAC;IAED,IAAI,SAAS;QACX,OAAO,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC;IAChC,CAAC;IAED,KAAK,CAAC,eAAe;QACnB,IAAI,CAAC,iBAAiB,GAAG,IAAI,oBAAoB,CAAC;YAChD,OAAO,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO;YACrC,IAAI,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI;YAC/B,GAAG,EAAE,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG;YAC7B,MAAM,EAAE,MAAM;SACf,CAAC,CAAC;QAEH,IAAI,CAAC,cAAc,GAAG,IAAI,MAAM,CAC9B,EAAE,IAAI,EAAE,yBAAyB,EAAE,OAAO,EAAE,OAAO,EAAE,EACrD,EAAE,YAAY,EAAE,EAAE,EAAE,CACrB,CAAC;QACF,MAAM,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,IAAI,CAAC,iBAAiB,CAAC,CAAC;QAE1D,uEAAuE;QACvE,8DAA8D;QAC9D,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,SAAS,EAAE,CAAC;QACpD,IAAI,CAAC,aAAa,GAAG,KAAK,CAAC,KAAK,CAAC;IACnC,CAAC;IAED,KAAK,CAAC,kBAAkB;QACtB,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;YACxB,MAAM,IAAI,CAAC,cAAc,CAAC,KAAK,EAAE,CAAC;YAClC,IAAI,CAAC,cAAc,GAAG,IAAI,CAAC;QAC7B,CAAC;IACH,CAAC;IAED,gBAAgB;QACd,OAAO,IAAI,CAAC,aAAa,CAAC;IAC5B,CAAC;IAED,UAAU;QACR,OAAO,IAAI,CAAC,OAAO,CAAC;IACtB,CAAC;IAED,KAAK,CAAC,iBAAiB,CACrB,QAAgB,EAChB,IAA6B;QAK7B,MAAM,UAAU,GAAG,aAAa,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QAEzD,kCAAkC;QAClC,MAAM,UAAU,GAAG,EAAE,GAAG,IAAI,EAAE,CAAC;QAC/B,MAAM,GAAG,GAAG,IAAI,CAAC,iBAAiB,CAAuB,CAAC;QAE1D,sBAAsB;QACtB,MAAM,QAAQ,GAAG,iBAAiB,CAChC,IAAI,CAAC,OAAO,EACZ,UAAU,EACV,QAAQ,EACR,UAAU,CACX,CAAC;QAEF,IAAI,gBAAgB,GAAmC,IAAI,CAAC;QAC5D,IAAI,gBAAgB,GAAG,KAAK,CAAC;QAC7B,IAAI,YAAY,GAAG,EAAE,CAAC;QACtB,IAAI,UAAU,GAAG,GAAG,CAAC;QACrB,IAAI,aAAa,GAAkB,IAAI,CAAC;QAExC,IAAI,QAAQ,CAAC,eAAe,IAAI,IAAI,CAAC,cAAc,EAAE,CAAC;YACpD,sBAAsB;YACtB,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,QAAQ,CAAC;oBAChD,IAAI,EAAE,QAAQ;oBACd,SAAS,EAAE,UAAU;iBACtB,CAAC,CAAC;gBACH,MAAM,OAAO,GAAG,MAAM,CAAC,OAAiD,CAAC;gBACzE,MAAM,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC;gBACxD,YAAY,GAAG,QAAQ,EAAE,IAAI,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;gBAChE,gBAAgB,GAAG,IAAI,CAAC;gBAExB,IAAI,CAAC;oBACH,gBAAgB,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;gBAC9C,CAAC;gBAAC,MAAM,CAAC;oBACP,gBAAgB,GAAG,EAAE,GAAG,EAAE,YAAY,EAAE,CAAC;gBAC3C,CAAC;YACH,CAAC;YAAC,OAAO,GAAG,EAAE,CAAC;gBACb,YAAY,GAAG,IAAI,CAAC,SAAS,CAAC;oBAC5B,KAAK,EAAE,gBAAgB;oBACvB,OAAO,EAAE,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC;iBAC1D,CAAC,CAAC;gBACH,aAAa,GAAG,kBAAkB,CAAC;YACrC,CAAC;QACH,CAAC;QAED,mEAAmE;QACnE,IAAI,QAAQ,CAAC,WAAW,EAAE,CAAC;YACzB,aAAa,GAAG,QAAQ,CAAC,SAAS,CAAC;YACnC,UAAU,GAAG,QAAQ,CAAC,UAAU,CAAC;YACjC,YAAY;gBACV,OAAO,QAAQ,CAAC,SAAS,KAAK,QAAQ;oBACpC,CAAC,CAAC,QAAQ,CAAC,SAAS;oBACpB,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;YAEzC,0DAA0D;YAC1D,IAAI,QAAQ,CAAC,eAAe,IAAI,gBAAgB,IAAI,GAAG,EAAE,CAAC;gBACxD,gBAAgB,CAAC,IAAI,CAAC,OAAO,EAAE,UAAU,EAAE,gBAAgB,CAAC,CAAC;YAC/D,CAAC;QACH,CAAC;aAAM,CAAC;YACN,UAAU,GAAG,GAAG,CAAC;YACjB,IAAI,CAAC,gBAAgB,EAAE,CAAC;gBACtB,2CAA2C;gBAC3C,YAAY,GAAG,IAAI,CAAC,SAAS,CAAC;oBAC5B,MAAM,EAAE,IAAI;oBACZ,UAAU;oBACV,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;iBACpC,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,kBAAkB;QAClB,MAAM,MAAM,GAAG,eAAe,CAC5B,UAAU,EACV,QAAQ,EACR,UAAU,EACV,aAAa,EACb,YAAY,EACZ,gBAAgB,CACjB,CAAC;QACF,SAAS,CAAC,IAAI,CAAC,OAAO,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;QAE1C,OAAO;YACL,OAAO,EAAE;gBACP;oBACE,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE,IAAI,CAAC,SAAS,CAClB;wBACE,UAAU;wBACV,UAAU;wBACV,aAAa;wBACb,gBAAgB;wBAChB,QAAQ,EAAE,CAAC,GAAG,EAAE;4BACd,IAAI,CAAC;gCACH,OAAO,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,CAAC;4BAClC,CAAC;4BAAC,MAAM,CAAC;gCACP,OAAO,YAAY,CAAC;4BACtB,CAAC;wBACH,CAAC,CAAC,EAAE;qBACL,EACD,IAAI,EACJ,CAAC,CACF;iBACF;aACF;YACD,OAAO,EAAE,UAAU,IAAI,GAAG;SAC3B,CAAC;IACJ,CAAC;CACF"}

package/dist/scenarios/f1-f5.d.ts

@@ -0,0 +1,9 @@
+import type { ScenarioDefinition } from "./types.js";
+export declare const F1_TIMEOUT_FIRST_WRITE: ScenarioDefinition;
+export declare const F2_RATE_LIMIT: ScenarioDefinition;
+export declare const F3_MALFORMED_JSON: ScenarioDefinition;
+export declare const F4_DUPLICATE_DISPATCH: ScenarioDefinition;
+export declare const F5_SERVER_ERROR_THEN_SUCCESS: ScenarioDefinition;
+export declare const ALL_SCENARIOS: ScenarioDefinition[];
+export declare function getScenarioById(id: string): ScenarioDefinition | undefined;
+//# sourceMappingURL=f1-f5.d.ts.map

package/dist/scenarios/f1-f5.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"f1-f5.d.ts","sourceRoot":"","sources":["../../src/scenarios/f1-f5.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,YAAY,CAAC;AAGrD,eAAO,MAAM,sBAAsB,EAAE,kBAepC,CAAC;AAGF,eAAO,MAAM,aAAa,EAAE,kBAiB3B,CAAC;AAGF,eAAO,MAAM,iBAAiB,EAAE,kBAa/B,CAAC;AAGF,eAAO,MAAM,qBAAqB,EAAE,kBAkBnC,CAAC;AAGF,eAAO,MAAM,4BAA4B,EAAE,kBAgB1C,CAAC;AAEF,eAAO,MAAM,aAAa,EAAE,kBAAkB,EAM7C,CAAC;AAEF,wBAAgB,eAAe,CAAC,EAAE,EAAE,MAAM,GAAG,kBAAkB,GAAG,SAAS,CAE1E"}