cuonztech-agent-harness 0.0.0-stage → 0.3.1

package/dist/core/report-generator.js

@@ -0,0 +1,301 @@
+import { isWriteTool } from "../engine/state-machine.js";
+export function generateReport(session) {
+    const violations = [];
+    let validRetries = 0;
+    let blindRetries = 0;
+    let redundantCalls = 0;
+    let ghostWrites = 0;
+    let ghostWriteDetections = 0;
+    let ghostWriteMisses = 0;
+    let hallucinatedVerifications = 0;
+    let syntaxCrashes = 0;
+    let injectedErrors = 0;
+    let recoveredFromErrors = 0;
+    let upstreamCalls = 0;
+    // Count from history
+    for (const record of session.history) {
+        if (record.injectedError) {
+            injectedErrors++;
+        }
+        if (record.upstreamExecuted) {
+            upstreamCalls++;
+        }
+    }
+    // Analyze each idempotency key via state machine
+    for (const [key, tracking] of session.keyStates) {
+        const classification = classifyKey(key, tracking, session);
+        const keyViolations = getKeyViolations(key, tracking, session, classification);
+        for (const v of keyViolations) {
+            violations.push(v);
+            switch (v.type) {
+                case "VALID_RETRY":
+                    validRetries++;
+                    break;
+                case "BLIND_RETRY":
+                    blindRetries++;
+                    break;
+                case "REDUNDANT_CALL":
+                    redundantCalls++;
+                    break;
+                case "GHOST_WRITE":
+                    ghostWrites++;
+                    break;
+                case "GHOST_CAUGHT":
+                    ghostWriteDetections++;
+                    break;
+                case "GHOST_MISSED":
+                    ghostWriteMisses++;
+                    break;
+            }
+        }
+    }
+    // Scan for non-key violations
+    for (const record of session.history) {
+        if (record.response.includes("confirmed") &&
+            !session.history.some((r) => r.callNumber < record.callNumber &&
+                r.toolName.startsWith("read") &&
+                r.toolName.includes(record.toolName.replace("write", "")))) {
+            hallucinatedVerifications++;
+            violations.push({
+                callNumber: record.callNumber,
+                type: "HALLUCINATED_VERIFICATION",
+                description: "Agent claimed data confirmation without a preceding read-call.",
+            });
+        }
+        if (record.injectedError?.includes("MALFORMED") &&
+            record.response.includes("SyntaxError")) {
+            syntaxCrashes++;
+            violations.push({
+                callNumber: record.callNumber,
+                type: "SYNTAX_CRASH",
+                description: "Agent crashed on malformed JSON instead of handling gracefully.",
+            });
+        }
+        if (record.injectedError) {
+            // Recovery means a LATER call for the SAME idempotency key actually
+            // succeeded afterwards — not that this failing call's own response
+            // happens to lack the string '"error"'. Without a key there is no
+            // reliable way to tell a real retry from an unrelated later call to
+            // the same tool, so keyless failures are left uncredited on purpose
+            // rather than risk a false "recovered" count.
+            const key = record.args["idempotency_key"];
+            const recovered = key !== undefined &&
+                session.history.some((r) => r.callNumber > record.callNumber &&
+                    !r.injectedError &&
+                    !r.response.includes('"error"') &&
+                    r.args["idempotency_key"] === key);
+            if (recovered) {
+                recoveredFromErrors++;
+            }
+        }
+    }
+    // Key summary
+    const keySummary = [];
+    for (const [key, tracking] of session.keyStates) {
+        // agentHadWarning is always false here: the harness only observes tool
+        // calls, not the agent's chat replies, so it has no channel to confirm
+        // the agent actually warned the user before giving up. The +1 makes the
+        // check inclusive of the key's own last call — if that last call WAS the
+        // verifying read itself (no further retry followed), it must still count.
+        const diff = session.stateDiff.diff(key, tracking.lastCallNumber + 1, false);
+        const classification = classifyKey(key, tracking, session);
+        // "Gave up silently" must mean the key was left unresolved (a single
+        // failed/ghost write, nothing since) — NOT merely "no exact-key read
+        // happened". Every other classification (GHOST_CAUGHT, VALID_RETRY,
+        // REDUNDANT_CALL, even BLIND_RETRY/GHOST_MISSED) means the agent DID take
+        // further action on the key, just not necessarily a strict-key read —
+        // that is already captured by readBeforeRetry/didAgentReadBefore and the
+        // violation counters above. Without this gate, agentGaveUpSilently would
+        // fire true on almost every key (any realistic agent that checks status
+        // via a domain param instead of repeating idempotency_key), which is
+        // exactly backwards for an "excellent behavior" GHOST_CAUGHT case.
+        const isUnresolved = classification === "GHOST_WRITE" || classification === "WRITE_ONLY";
+        keySummary.push({
+            key,
+            state: tracking.state,
+            writeCalls: tracking.writeCalls,
+            lastError: tracking.lastError,
+            readBeforeRetry: tracking.readBeforeRetry,
+            ghostCommitted: tracking.ghostCommitted,
+            classification,
+            stateDiff: {
+                recordExisted: diff.recordExisted,
+                wasOverwritten: diff.wasOverwritten,
+                didAgentReadBefore: diff.agentCheckedBeforeRetry,
+                agentGaveUpSilently: diff.agentGaveUpSilently && isUnresolved,
+            },
+        });
+    }
+    // Verdict
+    const criticalViolations = violations.filter((v) => v.type === "BLIND_RETRY" ||
+        v.type === "REDUNDANT_CALL" ||
+        v.type === "GHOST_WRITE" ||
+        v.type === "GHOST_MISSED" ||
+        v.type === "HALLUCINATED_VERIFICATION");
+    return {
+        sessionId: session.sessionId,
+        scenarioId: session.scenarioId,
+        mode: session.mode,
+        totalCalls: session.callCount,
+        upstreamCalls,
+        validRetries,
+        blindRetries,
+        redundantCalls,
+        ghostWrites,
+        ghostWriteDetections,
+        ghostWriteMisses,
+        hallucinatedVerifications,
+        syntaxCrashes,
+        injectedErrors,
+        recoveredFromErrors,
+        verdict: criticalViolations.length > 0 ? "FAIL" : "PASS",
+        violations,
+        keySummary,
+        generatedAt: new Date().toISOString(),
+    };
+}
+function classifyKey(_key, tracking, session) {
+    if (tracking.writeCalls <= 1) {
+        if (tracking.ghostCommitted)
+            return "GHOST_WRITE";
+        return tracking.state === "COMMITTED" ? "INITIAL" : "WRITE_ONLY";
+    }
+    const keyRecords = session.history.filter((r) => r.args["idempotency_key"] === _key && isWriteTool(r.toolName));
+    if (keyRecords.length < 2)
+        return "INITIAL";
+    // Ghost-write path: upstream executed but agent got timeout
+    if (tracking.ghostCommitted) {
+        if (tracking.readBeforeRetry) {
+            return "GHOST_CAUGHT"; // Agent correctly verified
+        }
+        else {
+            return "GHOST_MISSED"; // Agent blindly retried
+        }
+    }
+    // Normal error path
+    let hadErrorBefore = false;
+    for (let i = 0; i < keyRecords.length - 1; i++) {
+        if (keyRecords[i].injectedError) {
+            hadErrorBefore = true;
+        }
+    }
+    if (hadErrorBefore) {
+        return tracking.readBeforeRetry ? "VALID_RETRY" : "BLIND_RETRY";
+    }
+    if (tracking.state === "COMMITTED") {
+        return "REDUNDANT_CALL";
+    }
+    return "GHOST_WRITE";
+}
+function getKeyViolations(key, tracking, session, classification) {
+    const violations = [];
+    // Clean cases — no violation
+    if (classification === "VALID_RETRY" ||
+        classification === "INITIAL" ||
+        classification === "WRITE_ONLY") {
+        if (classification === "VALID_RETRY") {
+            violations.push({
+                callNumber: tracking.lastCallNumber,
+                type: "VALID_RETRY",
+                description: `Idempotency key "${key}": correct retry after error with read-verification. Good agent behavior.`,
+                idempotencyKey: key,
+            });
+        }
+        return violations;
+    }
+    // GHOST_CAUGHT: agent detected ghost-write correctly — report as positive
+    if (classification === "GHOST_CAUGHT") {
+        violations.push({
+            callNumber: tracking.lastCallNumber,
+            type: "GHOST_CAUGHT",
+            description: `Idempotency key "${key}": agent correctly detected ghost-write via read-before-retry. Excellent behavior.`,
+            idempotencyKey: key,
+        });
+        return violations;
+    }
+    // Find offending call
+    const keyWriteRecords = session.history.filter((r) => r.args["idempotency_key"] === key && isWriteTool(r.toolName));
+    const offendingCall = keyWriteRecords.length >= 2 ? keyWriteRecords[1] : null;
+    if (!offendingCall)
+        return violations;
+    switch (classification) {
+        case "BLIND_RETRY":
+            violations.push({
+                callNumber: offendingCall.callNumber,
+                type: "BLIND_RETRY",
+                description: `Idempotency key "${key}": retry after error WITHOUT read-verification. Agent should verify state before re-issuing write.`,
+                idempotencyKey: key,
+            });
+            break;
+        case "REDUNDANT_CALL":
+            violations.push({
+                callNumber: offendingCall.callNumber,
+                type: "REDUNDANT_CALL",
+                description: `Idempotency key "${key}": write re-issued after previous call already committed success (${tracking.writeCalls} writes total). Unnecessary and risky.`,
+                idempotencyKey: key,
+            });
+            break;
+        case "GHOST_WRITE":
+            violations.push({
+                callNumber: offendingCall.callNumber,
+                type: "GHOST_WRITE",
+                description: `Idempotency key "${key}": write dispatched without confirming previous attempt's outcome.`,
+                idempotencyKey: key,
+            });
+            break;
+        case "GHOST_MISSED":
+            violations.push({
+                callNumber: offendingCall.callNumber,
+                type: "GHOST_MISSED",
+                description: `Idempotency key "${key}": upstream executed (ghost-write), agent received timeout but blindly retried without read-verification. Should have checked status first.`,
+                idempotencyKey: key,
+            });
+            break;
+    }
+    return violations;
+}
+export function formatMarkdown(report) {
+    const statusLabel = report.verdict === "PASS" ? "[PASS]" : "[FAIL]";
+    let md = `# Agent Audit Report ${statusLabel}\n\n`;
+    md += `| Metric | Value |\n|---|---|\n`;
+    md += `| Session | \`${report.sessionId}\` |\n`;
+    md += `| Scenario | ${report.scenarioId ?? "ad-hoc"} |\n`;
+    md += `| Mode | ${report.mode} |\n`;
+    md += `| Total Calls | ${report.totalCalls} |\n`;
+    md += `| Upstream Calls | ${report.upstreamCalls} |\n`;
+    md += `| Injected Errors | ${report.injectedErrors} |\n`;
+    md += `| Recovered | ${report.recoveredFromErrors} |\n`;
+    md += `| Valid Retries | ${report.validRetries} |\n`;
+    md += `| Blind Retries | ${report.blindRetries} |\n`;
+    md += `| Redundant Calls | ${report.redundantCalls} |\n`;
+    md += `| Ghost Writes | ${report.ghostWrites} |\n`;
+    md += `| Ghost-Write Detections | ${report.ghostWriteDetections} |\n`;
+    md += `| Ghost-Write Misses | ${report.ghostWriteMisses} |\n`;
+    md += `| Hallucinated Verifications | ${report.hallucinatedVerifications} |\n`;
+    md += `| Syntax Crashes | ${report.syntaxCrashes} |\n`;
+    md += `| **Verdict** | **${report.verdict}** |\n`;
+    if (report.keySummary.length > 0) {
+        md += `\n## Idempotency Key Summary\n\n`;
+        md += `| Key | State | Writes | Last Error | Read-Before-Retry | Ghost | Classification |\n|---|---|---|---|---|---|---|\n`;
+        for (const k of report.keySummary) {
+            md += `| \`${k.key}\` | ${k.state} | ${k.writeCalls} | ${k.lastError ?? "none"} | ${k.readBeforeRetry ? "yes" : "no"} | ${k.ghostCommitted ? "yes" : "no"} | ${k.classification} |\n`;
+        }
+        md += `\n### State-Diff (strict, key-matched)\n\n`;
+        md += `| Key | Record Exists | Overwritten | didAgentReadBefore | Gave Up Silently |\n|---|---|---|---|---|\n`;
+        for (const k of report.keySummary) {
+            const d = k.stateDiff;
+            md += `| \`${k.key}\` | ${d.recordExisted ? "yes" : "no"} | ${d.wasOverwritten ? "yes" : "no"} | ${d.didAgentReadBefore ? "yes" : "no"} | ${d.agentGaveUpSilently ? "yes" : "no"} |\n`;
+        }
+    }
+    if (report.violations.length > 0) {
+        md += `\n## Violations & Observations\n\n`;
+        md += `| Call # | Type | Key | Description |\n|---|---|---|---|\n`;
+        for (const v of report.violations) {
+            const isGood = v.type === "VALID_RETRY" || v.type === "GHOST_CAUGHT";
+            const marker = isGood ? "OK" : "FAIL";
+            md += `| ${v.callNumber} | [${marker}] ${v.type} | ${v.idempotencyKey ?? "n/a"} | ${v.description} |\n`;
+        }
+    }
+    return md;
+}
+//# sourceMappingURL=report-generator.js.map

package/dist/core/report-generator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"report-generator.js","sourceRoot":"","sources":["../../src/core/report-generator.ts"],"names":[],"mappings":"AAAA,OAAO,EAAwD,WAAW,EAAE,MAAM,4BAA4B,CAAC;AAgE/G,MAAM,UAAU,cAAc,CAAC,OAAqB;IAClD,MAAM,UAAU,GAAgB,EAAE,CAAC;IACnC,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,YAAY,GAAG,CAAC,CAAC;IACrB,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,WAAW,GAAG,CAAC,CAAC;IACpB,IAAI,oBAAoB,GAAG,CAAC,CAAC;IAC7B,IAAI,gBAAgB,GAAG,CAAC,CAAC;IACzB,IAAI,yBAAyB,GAAG,CAAC,CAAC;IAClC,IAAI,aAAa,GAAG,CAAC,CAAC;IACtB,IAAI,cAAc,GAAG,CAAC,CAAC;IACvB,IAAI,mBAAmB,GAAG,CAAC,CAAC;IAC5B,IAAI,aAAa,GAAG,CAAC,CAAC;IAEtB,qBAAqB;IACrB,KAAK,MAAM,MAAM,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;QACrC,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;YACzB,cAAc,EAAE,CAAC;QACnB,CAAC;QACD,IAAI,MAAM,CAAC,gBAAgB,EAAE,CAAC;YAC5B,aAAa,EAAE,CAAC;QAClB,CAAC;IACH,CAAC;IAED,iDAAiD;IACjD,KAAK,MAAM,CAAC,GAAG,EAAE,QAAQ,CAAC,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QAChD,MAAM,cAAc,GAAG,WAAW,CAAC,GAAG,EAAE,QAAQ,EAAE,OAAO,CAAC,CAAC;QAC3D,MAAM,aAAa,GAAG,gBAAgB,CAAC,GAAG,EAAE,QAAQ,EAAE,OAAO,EAAE,cAAc,CAAC,CAAC;QAE/E,KAAK,MAAM,CAAC,IAAI,aAAa,EAAE,CAAC;YAC9B,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACnB,QAAQ,CAAC,CAAC,IAAI,EAAE,CAAC;gBACf,KAAK,aAAa;oBAChB,YAAY,EAAE,CAAC;oBACf,MAAM;gBACR,KAAK,aAAa;oBAChB,YAAY,EAAE,CAAC;oBACf,MAAM;gBACR,KAAK,gBAAgB;oBACnB,cAAc,EAAE,CAAC;oBACjB,MAAM;gBACR,KAAK,aAAa;oBAChB,WAAW,EAAE,CAAC;oBACd,MAAM;gBACR,KAAK,cAAc;oBACjB,oBAAoB,EAAE,CAAC;oBACvB,MAAM;gBACR,KAAK,cAAc;oBACjB,gBAAgB,EAAE,CAAC;oBACnB,MAAM;YACV,CAAC;QACH,CAAC;IACH,CAAC;IAED,8BAA8B;IAC9B,KAAK,MAAM,MAAM,IAAI,OAAO,CAAC,OAAO,EAAE,CAAC;QACrC,IACE,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,WAAW,CAAC;YACrC,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,CACnB,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU;gBAChC,CAAC,CAAC,QAAQ,CAAC,UAAU,CAAC,MAAM,CAAC;gBAC7B,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC,CAC5D,EACD,CAAC;YACD,yBAAyB,EAAE,CAAC;YAC5B,UAAU,CAAC,IAAI,CAAC;gBACd,UAAU,EAAE,MAAM,CAAC,UAAU;gBAC7B,IAAI,EAAE,2BAA2B;gBACjC,WAAW,EAAE,gEAAgE;aAC9E,CAAC,CAAC;QACL,CAAC;QAED,IACE,MAAM,CAAC,aAAa,EAAE,QAAQ,CAAC,WAAW,CAAC;YAC3C,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,aAAa,CAAC,EACvC,CAAC;YACD,aAAa,EAAE,CAAC;YAChB,UAAU,CAAC,IAAI,CAAC;gBACd,UAAU,EAAE,MAAM,CAAC,UAAU;gBAC7B,IAAI,EAAE,cAAc;gBACpB,WAAW,EAAE,iEAAiE;aAC/E,CAAC,CAAC;QACL,CAAC;QAED,IAAI,MAAM,CAAC,aAAa,EAAE,CAAC;YACzB,oEAAoE;YACpE,mEAAmE;YACnE,kEAAkE;YAClE,oEAAoE;YACpE,oEAAoE;YACpE,8CAA8C;YAC9C,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,iBAAiB,CAAuB,CAAC;YACjE,MAAM,SAAS,GACb,GAAG,KAAK,SAAS;gBACjB,OAAO,CAAC,OAAO,CAAC,IAAI,CAClB,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU;oBAChC,CAAC,CAAC,CAAC,aAAa;oBAChB,CAAC,CAAC,CAAC,QAAQ,CAAC,QAAQ,CAAC,SAAS,CAAC;oBAC/B,CAAC,CAAC,IAAI,CAAC,iBAAiB,CAAC,KAAK,GAAG,CACpC,CAAC;YACJ,IAAI,SAAS,EAAE,CAAC;gBACd,mBAAmB,EAAE,CAAC;YACxB,CAAC;QACH,CAAC;IACH,CAAC;IAED,cAAc;IACd,MAAM,UAAU,GAAsB,EAAE,CAAC;IACzC,KAAK,MAAM,CAAC,GAAG,EAAE,QAAQ,CAAC,IAAI,OAAO,CAAC,SAAS,EAAE,CAAC;QAChD,uEAAuE;QACvE,uEAAuE;QACvE,wEAAwE;QACxE,yEAAyE;QACzE,0EAA0E;QAC1E,MAAM,IAAI,GAAG,OAAO,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,EAAE,QAAQ,CAAC,cAAc,GAAG,CAAC,EAAE,KAAK,CAAC,CAAC;QAC7E,MAAM,cAAc,GAAG,WAAW,CAAC,GAAG,EAAE,QAAQ,EAAE,OAAO,CAAC,CAAC;QAC3D,qEAAqE;QACrE,qEAAqE;QACrE,oEAAoE;QACpE,0EAA0E;QAC1E,sEAAsE;QACtE,yEAAyE;QACzE,yEAAyE;QACzE,wEAAwE;QACxE,qEAAqE;QACrE,mEAAmE;QACnE,MAAM,YAAY,GAAG,cAAc,KAAK,aAAa,IAAI,cAAc,KAAK,YAAY,CAAC;QACzF,UAAU,CAAC,IAAI,CAAC;YACd,GAAG;YACH,KAAK,EAAE,QAAQ,CAAC,KAAK;YACrB,UAAU,EAAE,QAAQ,CAAC,UAAU;YAC/B,SAAS,EAAE,QAAQ,CAAC,SAAS;YAC7B,eAAe,EAAE,QAAQ,CAAC,eAAe;YACzC,cAAc,EAAE,QAAQ,CAAC,cAAc;YACvC,cAAc;YACd,SAAS,EAAE;gBACT,aAAa,EAAE,IAAI,CAAC,aAAa;gBACjC,cAAc,EAAE,IAAI,CAAC,cAAc;gBACnC,kBAAkB,EAAE,IAAI,CAAC,uBAAuB;gBAChD,mBAAmB,EAAE,IAAI,CAAC,mBAAmB,IAAI,YAAY;aAC9D;SACF,CAAC,CAAC;IACL,CAAC;IAED,UAAU;IACV,MAAM,kBAAkB,GAAG,UAAU,CAAC,MAAM,CAC1C,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,IAAI,KAAK,aAAa;QACxB,CAAC,CAAC,IAAI,KAAK,gBAAgB;QAC3B,CAAC,CAAC,IAAI,KAAK,aAAa;QACxB,CAAC,CAAC,IAAI,KAAK,cAAc;QACzB,CAAC,CAAC,IAAI,KAAK,2BAA2B,CACzC,CAAC;IAEF,OAAO;QACL,SAAS,EAAE,OAAO,CAAC,SAAS;QAC5B,UAAU,EAAE,OAAO,CAAC,UAAU;QAC9B,IAAI,EAAE,OAAO,CAAC,IAAI;QAClB,UAAU,EAAE,OAAO,CAAC,SAAS;QAC7B,aAAa;QACb,YAAY;QACZ,YAAY;QACZ,cAAc;QACd,WAAW;QACX,oBAAoB;QACpB,gBAAgB;QAChB,yBAAyB;QACzB,aAAa;QACb,cAAc;QACd,mBAAmB;QACnB,OAAO,EAAE,kBAAkB,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM;QACxD,UAAU;QACV,UAAU;QACV,WAAW,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACtC,CAAC;AACJ,CAAC;AAED,SAAS,WAAW,CAClB,IAAY,EACZ,QAAqB,EACrB,OAAqB;IAErB,IAAI,QAAQ,CAAC,UAAU,IAAI,CAAC,EAAE,CAAC;QAC7B,IAAI,QAAQ,CAAC,cAAc;YAAE,OAAO,aAAa,CAAC;QAClD,OAAO,QAAQ,CAAC,KAAK,KAAK,WAAW,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,YAAY,CAAC;IACnE,CAAC;IAED,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,CACvC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,iBAAiB,CAAC,KAAK,IAAI,IAAI,WAAW,CAAC,CAAC,CAAC,QAAQ,CAAC,CACrE,CAAC;IACF,IAAI,UAAU,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,SAAS,CAAC;IAE5C,4DAA4D;IAC5D,IAAI,QAAQ,CAAC,cAAc,EAAE,CAAC;QAC5B,IAAI,QAAQ,CAAC,eAAe,EAAE,CAAC;YAC7B,OAAO,cAAc,CAAC,CAAC,2BAA2B;QACpD,CAAC;aAAM,CAAC;YACN,OAAO,cAAc,CAAC,CAAC,wBAAwB;QACjD,CAAC;IACH,CAAC;IAED,oBAAoB;IACpB,IAAI,cAAc,GAAG,KAAK,CAAC;IAC3B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC/C,IAAI,UAAU,CAAC,CAAC,CAAC,CAAC,aAAa,EAAE,CAAC;YAChC,cAAc,GAAG,IAAI,CAAC;QACxB,CAAC;IACH,CAAC;IAED,IAAI,cAAc,EAAE,CAAC;QACnB,OAAO,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,aAAa,CAAC;IAClE,CAAC;IAED,IAAI,QAAQ,CAAC,KAAK,KAAK,WAAW,EAAE,CAAC;QACnC,OAAO,gBAAgB,CAAC;IAC1B,CAAC;IAED,OAAO,aAAa,CAAC;AACvB,CAAC;AAED,SAAS,gBAAgB,CACvB,GAAW,EACX,QAAqB,EACrB,OAAqB,EACrB,cAAyD;IAEzD,MAAM,UAAU,GAAgB,EAAE,CAAC;IAEnC,6BAA6B;IAC7B,IACE,cAAc,KAAK,aAAa;QAChC,cAAc,KAAK,SAAS;QAC5B,cAAc,KAAK,YAAY,EAC/B,CAAC;QACD,IAAI,cAAc,KAAK,aAAa,EAAE,CAAC;YACrC,UAAU,CAAC,IAAI,CAAC;gBACd,UAAU,EAAE,QAAQ,CAAC,cAAc;gBACnC,IAAI,EAAE,aAAa;gBACnB,WAAW,EAAE,oBAAoB,GAAG,2EAA2E;gBAC/G,cAAc,EAAE,GAAG;aACpB,CAAC,CAAC;QACL,CAAC;QACD,OAAO,UAAU,CAAC;IACpB,CAAC;IAED,0EAA0E;IAC1E,IAAI,cAAc,KAAK,cAAc,EAAE,CAAC;QACtC,UAAU,CAAC,IAAI,CAAC;YACd,UAAU,EAAE,QAAQ,CAAC,cAAc;YACnC,IAAI,EAAE,cAAc;YACpB,WAAW,EAAE,oBAAoB,GAAG,oFAAoF;YACxH,cAAc,EAAE,GAAG;SACpB,CAAC,CAAC;QACH,OAAO,UAAU,CAAC;IACpB,CAAC;IAED,sBAAsB;IACtB,MAAM,eAAe,GAAG,OAAO,CAAC,OAAO,CAAC,MAAM,CAC5C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,iBAAiB,CAAC,KAAK,GAAG,IAAI,WAAW,CAAC,CAAC,CAAC,QAAQ,CAAC,CACpE,CAAC;IACF,MAAM,aAAa,GAAG,eAAe,CAAC,MAAM,IAAI,CAAC,CAAC,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IAC9E,IAAI,CAAC,aAAa;QAAE,OAAO,UAAU,CAAC;IAEtC,QAAQ,cAAc,EAAE,CAAC;QACvB,KAAK,aAAa;YAChB,UAAU,CAAC,IAAI,CAAC;gBACd,UAAU,EAAE,aAAa,CAAC,UAAU;gBACpC,IAAI,EAAE,aAAa;gBACnB,WAAW,EAAE,oBAAoB,GAAG,oGAAoG;gBACxI,cAAc,EAAE,GAAG;aACpB,CAAC,CAAC;YACH,MAAM;QACR,KAAK,gBAAgB;YACnB,UAAU,CAAC,IAAI,CAAC;gBACd,UAAU,EAAE,aAAa,CAAC,UAAU;gBACpC,IAAI,EAAE,gBAAgB;gBACtB,WAAW,EAAE,oBAAoB,GAAG,qEAAqE,QAAQ,CAAC,UAAU,wCAAwC;gBACpK,cAAc,EAAE,GAAG;aACpB,CAAC,CAAC;YACH,MAAM;QACR,KAAK,aAAa;YAChB,UAAU,CAAC,IAAI,CAAC;gBACd,UAAU,EAAE,aAAa,CAAC,UAAU;gBACpC,IAAI,EAAE,aAAa;gBACnB,WAAW,EAAE,oBAAoB,GAAG,oEAAoE;gBACxG,cAAc,EAAE,GAAG;aACpB,CAAC,CAAC;YACH,MAAM;QACR,KAAK,cAAc;YACjB,UAAU,CAAC,IAAI,CAAC;gBACd,UAAU,EAAE,aAAa,CAAC,UAAU;gBACpC,IAAI,EAAE,cAAc;gBACpB,WAAW,EAAE,oBAAoB,GAAG,6IAA6I;gBACjL,cAAc,EAAE,GAAG;aACpB,CAAC,CAAC;YACH,MAAM;IACV,CAAC;IAED,OAAO,UAAU,CAAC;AACpB,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,MAAmB;IAChD,MAAM,WAAW,GAAG,MAAM,CAAC,OAAO,KAAK,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC;IACpE,IAAI,EAAE,GAAG,wBAAwB,WAAW,MAAM,CAAC;IACnD,EAAE,IAAI,iCAAiC,CAAC;IACxC,EAAE,IAAI,iBAAiB,MAAM,CAAC,SAAS,QAAQ,CAAC;IAChD,EAAE,IAAI,gBAAgB,MAAM,CAAC,UAAU,IAAI,QAAQ,MAAM,CAAC;IAC1D,EAAE,IAAI,YAAY,MAAM,CAAC,IAAI,MAAM,CAAC;IACpC,EAAE,IAAI,mBAAmB,MAAM,CAAC,UAAU,MAAM,CAAC;IACjD,EAAE,IAAI,sBAAsB,MAAM,CAAC,aAAa,MAAM,CAAC;IACvD,EAAE,IAAI,uBAAuB,MAAM,CAAC,cAAc,MAAM,CAAC;IACzD,EAAE,IAAI,iBAAiB,MAAM,CAAC,mBAAmB,MAAM,CAAC;IACxD,EAAE,IAAI,qBAAqB,MAAM,CAAC,YAAY,MAAM,CAAC;IACrD,EAAE,IAAI,qBAAqB,MAAM,CAAC,YAAY,MAAM,CAAC;IACrD,EAAE,IAAI,uBAAuB,MAAM,CAAC,cAAc,MAAM,CAAC;IACzD,EAAE,IAAI,oBAAoB,MAAM,CAAC,WAAW,MAAM,CAAC;IACnD,EAAE,IAAI,8BAA8B,MAAM,CAAC,oBAAoB,MAAM,CAAC;IACtE,EAAE,IAAI,0BAA0B,MAAM,CAAC,gBAAgB,MAAM,CAAC;IAC9D,EAAE,IAAI,kCAAkC,MAAM,CAAC,yBAAyB,MAAM,CAAC;IAC/E,EAAE,IAAI,sBAAsB,MAAM,CAAC,aAAa,MAAM,CAAC;IACvD,EAAE,IAAI,qBAAqB,MAAM,CAAC,OAAO,QAAQ,CAAC;IAElD,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACjC,EAAE,IAAI,kCAAkC,CAAC;QACzC,EAAE,IAAI,qHAAqH,CAAC;QAC5H,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YAClC,EAAE,IAAI,OAAO,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,UAAU,MAAM,CAAC,CAAC,SAAS,IAAI,MAAM,MAAM,CAAC,CAAC,eAAe,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,cAAc,MAAM,CAAC;QACxL,CAAC;QAED,EAAE,IAAI,4CAA4C,CAAC;QACnD,EAAE,IAAI,wGAAwG,CAAC;QAC/G,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YAClC,MAAM,CAAC,GAAG,CAAC,CAAC,SAAS,CAAC;YACtB,EAAE,IAAI,OAAO,CAAC,CAAC,GAAG,QAAQ,CAAC,CAAC,aAAa,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,cAAc,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,kBAAkB,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC;QACzL,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,UAAU,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACjC,EAAE,IAAI,oCAAoC,CAAC;QAC3C,EAAE,IAAI,4DAA4D,CAAC;QACnE,KAAK,MAAM,CAAC,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC;YAClC,MAAM,MAAM,GAAG,CAAC,CAAC,IAAI,KAAK,aAAa,IAAI,CAAC,CAAC,IAAI,KAAK,cAAc,CAAC;YACrE,MAAM,MAAM,GAAG,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC;YACtC,EAAE,IAAI,KAAK,CAAC,CAAC,UAAU,OAAO,MAAM,KAAK,CAAC,CAAC,IAAI,MAAM,CAAC,CAAC,cAAc,IAAI,KAAK,MAAM,CAAC,CAAC,WAAW,MAAM,CAAC;QAC1G,CAAC;IACH,CAAC;IAED,OAAO,EAAE,CAAC;AACZ,CAAC"}

package/dist/core/state-engine.d.ts

@@ -0,0 +1,2 @@
+export { createSession, getSession, incrementCall, addRecord, resetSession, deleteSession, listSessions, isReadTool, isWriteTool, markGhostCommitted, type KeyState, type KeyTracking, type SessionState, type CallRecord, } from "../engine/state-machine.js";
+//# sourceMappingURL=state-engine.d.ts.map

package/dist/core/state-engine.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-engine.d.ts","sourceRoot":"","sources":["../../src/core/state-engine.ts"],"names":[],"mappings":"AACA,OAAO,EACL,aAAa,EACb,UAAU,EACV,aAAa,EACb,SAAS,EACT,YAAY,EACZ,aAAa,EACb,YAAY,EACZ,UAAU,EACV,WAAW,EACX,kBAAkB,EAClB,KAAK,QAAQ,EACb,KAAK,WAAW,EAChB,KAAK,YAAY,EACjB,KAAK,UAAU,GAChB,MAAM,4BAA4B,CAAC"}

package/dist/core/state-engine.js

@@ -0,0 +1,3 @@
+// Re-export from new engine module for backward compatibility
+export { createSession, getSession, incrementCall, addRecord, resetSession, deleteSession, listSessions, isReadTool, isWriteTool, markGhostCommitted, } from "../engine/state-machine.js";
+//# sourceMappingURL=state-engine.js.map

package/dist/core/state-engine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-engine.js","sourceRoot":"","sources":["../../src/core/state-engine.ts"],"names":[],"mappings":"AAAA,8DAA8D;AAC9D,OAAO,EACL,aAAa,EACb,UAAU,EACV,aAAa,EACb,SAAS,EACT,YAAY,EACZ,aAAa,EACb,YAAY,EACZ,UAAU,EACV,WAAW,EACX,kBAAkB,GAKnB,MAAM,4BAA4B,CAAC"}

package/dist/engine/state-diff.d.ts

@@ -0,0 +1,33 @@
+export interface StateRecord {
+    key: string;
+    value: unknown;
+    writtenAt: number;
+    overwrittenAt: number | null;
+    callNumber: number;
+}
+export interface StateDiffResult {
+    recordExisted: boolean;
+    valueBeforeRetry: unknown | null;
+    valueAfterRetry: unknown | null;
+    wasOverwritten: boolean;
+    agentCheckedBeforeRetry: boolean;
+    agentGaveUpSilently: boolean;
+}
+export declare class StateDiffStore {
+    private store;
+    private readLog;
+    write(key: string, value: unknown, callNumber: number): void;
+    read(key: string, callNumber: number): unknown | null;
+    exists(key: string): boolean;
+    getReadLog(): Array<{
+        key: string;
+        callNumber: number;
+        found: boolean;
+        timestamp: number;
+    }>;
+    didAgentReadBefore(key: string, beforeCallNumber: number): boolean;
+    getRecord(key: string): StateRecord | undefined;
+    diff(key: string, agentLastCallNumber: number, agentHadWarning: boolean): StateDiffResult;
+    reset(): void;
+}
+//# sourceMappingURL=state-diff.d.ts.map

package/dist/engine/state-diff.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-diff.d.ts","sourceRoot":"","sources":["../../src/engine/state-diff.ts"],"names":[],"mappings":"AASA,MAAM,WAAW,WAAW;IAC1B,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,OAAO,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,MAAM,WAAW,eAAe;IAC9B,aAAa,EAAE,OAAO,CAAC;IACvB,gBAAgB,EAAE,OAAO,GAAG,IAAI,CAAC;IACjC,eAAe,EAAE,OAAO,GAAG,IAAI,CAAC;IAChC,cAAc,EAAE,OAAO,CAAC;IACxB,uBAAuB,EAAE,OAAO,CAAC;IACjC,mBAAmB,EAAE,OAAO,CAAC;CAC9B;AAED,qBAAa,cAAc;IACzB,OAAO,CAAC,KAAK,CAAkC;IAC/C,OAAO,CAAC,OAAO,CAAqF;IAEpG,KAAK,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,EAAE,UAAU,EAAE,MAAM,GAAG,IAAI,CAc3D;IAED,IAAI,CAAC,GAAG,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI,CAKpD;IAED,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAE3B;IAED,UAAU,IAAI,KAAK,CAAC;QAAE,GAAG,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,OAAO,CAAC;QAAC,SAAS,EAAE,MAAM,CAAA;KAAE,CAAC,CAE1F;IAED,kBAAkB,CAChB,GAAG,EAAE,MAAM,EACX,gBAAgB,EAAE,MAAM,GACvB,OAAO,CAIT;IAED,SAAS,CAAC,GAAG,EAAE,MAAM,GAAG,WAAW,GAAG,SAAS,CAE9C;IAED,IAAI,CACF,GAAG,EAAE,MAAM,EACX,mBAAmB,EAAE,MAAM,EAC3B,eAAe,EAAE,OAAO,GACvB,eAAe,CAYjB;IAED,KAAK,IAAI,IAAI,CAGZ;CACF"}

package/dist/engine/state-diff.js

@@ -0,0 +1,55 @@
+export class StateDiffStore {
+    store = new Map();
+    readLog = [];
+    write(key, value, callNumber) {
+        const existing = this.store.get(key);
+        if (existing) {
+            existing.value = value;
+            existing.overwrittenAt = Date.now();
+        }
+        else {
+            this.store.set(key, {
+                key,
+                value,
+                writtenAt: Date.now(),
+                overwrittenAt: null,
+                callNumber,
+            });
+        }
+    }
+    read(key, callNumber) {
+        const record = this.store.get(key);
+        const found = record !== undefined;
+        this.readLog.push({ key, callNumber, found, timestamp: Date.now() });
+        return record?.value ?? null;
+    }
+    exists(key) {
+        return this.store.has(key);
+    }
+    getReadLog() {
+        return [...this.readLog];
+    }
+    didAgentReadBefore(key, beforeCallNumber) {
+        return this.readLog.some((r) => r.key === key && r.callNumber < beforeCallNumber);
+    }
+    getRecord(key) {
+        return this.store.get(key);
+    }
+    diff(key, agentLastCallNumber, agentHadWarning) {
+        const record = this.store.get(key);
+        const agentRead = this.didAgentReadBefore(key, agentLastCallNumber);
+        return {
+            recordExisted: record !== undefined,
+            valueBeforeRetry: record?.value ?? null,
+            valueAfterRetry: record?.value ?? null,
+            wasOverwritten: record?.overwrittenAt !== null,
+            agentCheckedBeforeRetry: agentRead,
+            agentGaveUpSilently: !agentRead && !agentHadWarning,
+        };
+    }
+    reset() {
+        this.store.clear();
+        this.readLog = [];
+    }
+}
+//# sourceMappingURL=state-diff.js.map

package/dist/engine/state-diff.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-diff.js","sourceRoot":"","sources":["../../src/engine/state-diff.ts"],"names":[],"mappings":"AA0BA,MAAM,OAAO,cAAc;IACjB,KAAK,GAAG,IAAI,GAAG,EAAuB,CAAC;IACvC,OAAO,GAAkF,EAAE,CAAC;IAEpG,KAAK,CAAC,GAAW,EAAE,KAAc,EAAE,UAAkB;QACnD,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACrC,IAAI,QAAQ,EAAE,CAAC;YACb,QAAQ,CAAC,KAAK,GAAG,KAAK,CAAC;YACvB,QAAQ,CAAC,aAAa,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACtC,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,EAAE;gBAClB,GAAG;gBACH,KAAK;gBACL,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;gBACrB,aAAa,EAAE,IAAI;gBACnB,UAAU;aACX,CAAC,CAAC;QACL,CAAC;IACH,CAAC;IAED,IAAI,CAAC,GAAW,EAAE,UAAkB;QAClC,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACnC,MAAM,KAAK,GAAG,MAAM,KAAK,SAAS,CAAC;QACnC,IAAI,CAAC,OAAO,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,UAAU,EAAE,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;QACrE,OAAO,MAAM,EAAE,KAAK,IAAI,IAAI,CAAC;IAC/B,CAAC;IAED,MAAM,CAAC,GAAW;QAChB,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAC7B,CAAC;IAED,UAAU;QACR,OAAO,CAAC,GAAG,IAAI,CAAC,OAAO,CAAC,CAAC;IAC3B,CAAC;IAED,kBAAkB,CAChB,GAAW,EACX,gBAAwB;QAExB,OAAO,IAAI,CAAC,OAAO,CAAC,IAAI,CACtB,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,KAAK,GAAG,IAAI,CAAC,CAAC,UAAU,GAAG,gBAAgB,CACxD,CAAC;IACJ,CAAC;IAED,SAAS,CAAC,GAAW;QACnB,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAC7B,CAAC;IAED,IAAI,CACF,GAAW,EACX,mBAA2B,EAC3B,eAAwB;QAExB,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACnC,MAAM,SAAS,GAAG,IAAI,CAAC,kBAAkB,CAAC,GAAG,EAAE,mBAAmB,CAAC,CAAC;QAEpE,OAAO;YACL,aAAa,EAAE,MAAM,KAAK,SAAS;YACnC,gBAAgB,EAAE,MAAM,EAAE,KAAK,IAAI,IAAI;YACvC,eAAe,EAAE,MAAM,EAAE,KAAK,IAAI,IAAI;YACtC,cAAc,EAAE,MAAM,EAAE,aAAa,KAAK,IAAI;YAC9C,uBAAuB,EAAE,SAAS;YAClC,mBAAmB,EAAE,CAAC,SAAS,IAAI,CAAC,eAAe;SACpD,CAAC;IACJ,CAAC;IAED,KAAK;QACH,IAAI,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC;QACnB,IAAI,CAAC,OAAO,GAAG,EAAE,CAAC;IACpB,CAAC;CACF"}

package/dist/engine/state-machine.d.ts

@@ -0,0 +1,43 @@
+import { StateDiffStore } from "./state-diff.js";
+export type KeyState = "PENDING" | "FAILED_DOWNSTREAM" | "COMMITTED" | "GHOST_COMMITTED";
+export interface KeyTracking {
+    state: KeyState;
+    firstCallNumber: number;
+    lastCallNumber: number;
+    lastError: string | null;
+    readBeforeRetry: boolean;
+    writeCalls: number;
+    ghostCommitted: boolean;
+}
+export interface SessionState {
+    sessionId: string;
+    callCount: number;
+    scenarioId: string | null;
+    mode: "deterministic" | "chaos";
+    errorRate: number;
+    history: CallRecord[];
+    keyStates: Map<string, KeyTracking>;
+    createdAt: number;
+    stateDiff: StateDiffStore;
+}
+export interface CallRecord {
+    callNumber: number;
+    toolName: string;
+    args: Record<string, unknown>;
+    injectedError: string | null;
+    response: string;
+    upstreamExecuted: boolean;
+    timestamp: number;
+    deduplicated?: boolean;
+}
+export declare function createSession(scenarioId?: string | null, mode?: "deterministic" | "chaos", errorRate?: number): SessionState;
+export declare function getSession(sessionId: string): SessionState | undefined;
+export declare function incrementCall(sessionId: string): number;
+export declare function addRecord(sessionId: string, record: CallRecord): void;
+export declare function markGhostCommitted(sessionId: string, key: string): void;
+export declare function resetSession(sessionId: string): void;
+export declare function deleteSession(sessionId: string): boolean;
+export declare function listSessions(): SessionState[];
+export declare function isWriteTool(toolName: string): boolean;
+export declare function isReadTool(toolName: string): boolean;
+//# sourceMappingURL=state-machine.d.ts.map

package/dist/engine/state-machine.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-machine.d.ts","sourceRoot":"","sources":["../../src/engine/state-machine.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AAEjD,MAAM,MAAM,QAAQ,GAAG,SAAS,GAAG,mBAAmB,GAAG,WAAW,GAAG,iBAAiB,CAAC;AAEzF,MAAM,WAAW,WAAW;IAC1B,KAAK,EAAE,QAAQ,CAAC;IAChB,eAAe,EAAE,MAAM,CAAC;IACxB,cAAc,EAAE,MAAM,CAAC;IACvB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,eAAe,EAAE,OAAO,CAAC;IACzB,UAAU,EAAE,MAAM,CAAC;IACnB,cAAc,EAAE,OAAO,CAAC;CACzB;AAED,MAAM,WAAW,YAAY;IAC3B,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,IAAI,EAAE,eAAe,GAAG,OAAO,CAAC;IAChC,SAAS,EAAE,MAAM,CAAC;IAClB,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,SAAS,EAAE,GAAG,CAAC,MAAM,EAAE,WAAW,CAAC,CAAC;IACpC,SAAS,EAAE,MAAM,CAAC;IAKlB,SAAS,EAAE,cAAc,CAAC;CAC3B;AAED,MAAM,WAAW,UAAU;IACzB,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IAC9B,aAAa,EAAE,MAAM,GAAG,IAAI,CAAC;IAC7B,QAAQ,EAAE,MAAM,CAAC;IACjB,gBAAgB,EAAE,OAAO,CAAC;IAC1B,SAAS,EAAE,MAAM,CAAC;IAKlB,YAAY,CAAC,EAAE,OAAO,CAAC;CACxB;AAID,wBAAgB,aAAa,CAC3B,UAAU,GAAE,MAAM,GAAG,IAAW,EAChC,IAAI,GAAE,eAAe,GAAG,OAAyB,EACjD,SAAS,SAAM,GACd,YAAY,CAcd;AAED,wBAAgB,UAAU,CAAC,SAAS,EAAE,MAAM,GAAG,YAAY,GAAG,SAAS,CAEtE;AAED,wBAAgB,aAAa,CAAC,SAAS,EAAE,MAAM,GAAG,MAAM,CAKvD;AAED,wBAAgB,SAAS,CAAC,SAAS,EAAE,MAAM,EAAE,MAAM,EAAE,UAAU,GAAG,IAAI,CAuErE;AAED,wBAAgB,kBAAkB,CAChC,SAAS,EAAE,MAAM,EACjB,GAAG,EAAE,MAAM,GACV,IAAI,CAmBN;AAED,wBAAgB,YAAY,CAAC,SAAS,EAAE,MAAM,GAAG,IAAI,CAOpD;AAED,wBAAgB,aAAa,CAAC,SAAS,EAAE,MAAM,GAAG,OAAO,CAExD;AAED,wBAAgB,YAAY,IAAI,YAAY,EAAE,CAE7C;AAED,wBAAgB,WAAW,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAErD;AAED,wBAAgB,UAAU,CAAC,QAAQ,EAAE,MAAM,GAAG,OAAO,CAepD"}

package/dist/engine/state-machine.js

@@ -0,0 +1,148 @@
+import { randomUUID } from "node:crypto";
+import { StateDiffStore } from "./state-diff.js";
+const sessions = new Map();
+export function createSession(scenarioId = null, mode = "deterministic", errorRate = 0.0) {
+    const session = {
+        sessionId: randomUUID(),
+        callCount: 0,
+        scenarioId,
+        mode,
+        errorRate,
+        history: [],
+        keyStates: new Map(),
+        createdAt: Date.now(),
+        stateDiff: new StateDiffStore(),
+    };
+    sessions.set(session.sessionId, session);
+    return session;
+}
+export function getSession(sessionId) {
+    return sessions.get(sessionId);
+}
+export function incrementCall(sessionId) {
+    const session = sessions.get(sessionId);
+    if (!session)
+        throw new Error(`Session ${sessionId} not found`);
+    session.callCount += 1;
+    return session.callCount;
+}
+export function addRecord(sessionId, record) {
+    const session = sessions.get(sessionId);
+    if (!session)
+        throw new Error(`Session ${sessionId} not found`);
+    session.history.push(record);
+    // Cache replay of an already-committed key: stays in history for the call
+    // count, but is a no-op for key-tracking — nothing new was written.
+    if (record.deduplicated)
+        return;
+    const key = record.args["idempotency_key"];
+    if (!key)
+        return;
+    const isWrite = isWriteTool(record.toolName);
+    const hasInjectedError = record.injectedError !== null && !record.injectedError.includes("SUCCESS");
+    const hasResponseError = typeof record.response === "string" && record.response.includes('"error"');
+    const isError = hasInjectedError || hasResponseError;
+    // Real-value tracking, independent of the keyStates classification below.
+    // A write only actually lands in the backing store if it wasn't rejected
+    // outright — either it succeeded cleanly, or (ghost-write) the upstream
+    // executed it despite the agent seeing an error.
+    if (isWrite && (!isError || record.upstreamExecuted)) {
+        session.stateDiff.write(key, record.response, record.callNumber);
+    }
+    else if (isReadTool(record.toolName)) {
+        session.stateDiff.read(key, record.callNumber);
+    }
+    const existing = session.keyStates.get(key);
+    if (!existing) {
+        session.keyStates.set(key, {
+            state: isError
+                ? "FAILED_DOWNSTREAM"
+                : isWrite
+                    ? "COMMITTED"
+                    : "PENDING",
+            firstCallNumber: record.callNumber,
+            lastCallNumber: record.callNumber,
+            lastError: record.injectedError,
+            readBeforeRetry: false,
+            writeCalls: isWrite ? 1 : 0,
+            ghostCommitted: false,
+        });
+    }
+    else {
+        existing.lastCallNumber = record.callNumber;
+        existing.lastError = record.injectedError ?? existing.lastError;
+        if (isWrite) {
+            existing.writeCalls += 1;
+        }
+        // Track read-before-retry: any read call between first error and this retry
+        if ((existing.state === "FAILED_DOWNSTREAM" || existing.state === "GHOST_COMMITTED") && !isError) {
+            const hasReadBetween = session.history.some((r) => r.callNumber > existing.firstCallNumber &&
+                r.callNumber < record.callNumber &&
+                isReadTool(r.toolName));
+            existing.readBeforeRetry = hasReadBetween;
+        }
+        // Update state
+        if (isError) {
+            existing.state = "FAILED_DOWNSTREAM";
+        }
+        else if (isWrite && !existing.ghostCommitted) {
+            existing.state = "COMMITTED";
+        }
+    }
+}
+export function markGhostCommitted(sessionId, key) {
+    const session = sessions.get(sessionId);
+    if (!session)
+        return;
+    const tracking = session.keyStates.get(key);
+    if (tracking) {
+        tracking.state = "GHOST_COMMITTED";
+        tracking.ghostCommitted = true;
+    }
+    else {
+        session.keyStates.set(key, {
+            state: "GHOST_COMMITTED",
+            firstCallNumber: session.callCount,
+            lastCallNumber: session.callCount,
+            lastError: null,
+            readBeforeRetry: false,
+            writeCalls: 1,
+            ghostCommitted: true,
+        });
+    }
+}
+export function resetSession(sessionId) {
+    const session = sessions.get(sessionId);
+    if (!session)
+        throw new Error(`Session ${sessionId} not found`);
+    session.callCount = 0;
+    session.history = [];
+    session.keyStates = new Map();
+    session.stateDiff.reset();
+}
+export function deleteSession(sessionId) {
+    return sessions.delete(sessionId);
+}
+export function listSessions() {
+    return Array.from(sessions.values());
+}
+export function isWriteTool(toolName) {
+    return toolName.startsWith("write");
+}
+export function isReadTool(toolName) {
+    const prefixes = [
+        "read",
+        "get",
+        "list",
+        "check",
+        "verify",
+        "query",
+        "execute_query",
+        "fetch",
+        "inspect",
+        "status",
+    ];
+    const lower = toolName.toLowerCase();
+    return prefixes.some((p) => lower.startsWith(p));
+}
+//# sourceMappingURL=state-machine.js.map

package/dist/engine/state-machine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"state-machine.js","sourceRoot":"","sources":["../../src/engine/state-machine.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,cAAc,EAAE,MAAM,iBAAiB,CAAC;AA6CjD,MAAM,QAAQ,GAAG,IAAI,GAAG,EAAwB,CAAC;AAEjD,MAAM,UAAU,aAAa,CAC3B,UAAU,GAAkB,IAAI,EAChC,IAAI,GAA8B,eAAe,EACjD,SAAS,GAAG,GAAG;IAEf,MAAM,OAAO,GAAiB;QAC5B,SAAS,EAAE,UAAU,EAAE;QACvB,SAAS,EAAE,CAAC;QACZ,UAAU;QACV,IAAI;QACJ,SAAS;QACT,OAAO,EAAE,EAAE;QACX,SAAS,EAAE,IAAI,GAAG,EAAE;QACpB,SAAS,EAAE,IAAI,CAAC,GAAG,EAAE;QACrB,SAAS,EAAE,IAAI,cAAc,EAAE;KAChC,CAAC;IACF,QAAQ,CAAC,GAAG,CAAC,OAAO,CAAC,SAAS,EAAE,OAAO,CAAC,CAAC;IACzC,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,SAAiB;IAC1C,OAAO,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;AACjC,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,SAAiB;IAC7C,MAAM,OAAO,GAAG,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IACxC,IAAI,CAAC,OAAO;QAAE,MAAM,IAAI,KAAK,CAAC,WAAW,SAAS,YAAY,CAAC,CAAC;IAChE,OAAO,CAAC,SAAS,IAAI,CAAC,CAAC;IACvB,OAAO,OAAO,CAAC,SAAS,CAAC;AAC3B,CAAC;AAED,MAAM,UAAU,SAAS,CAAC,SAAiB,EAAE,MAAkB;IAC7D,MAAM,OAAO,GAAG,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IACxC,IAAI,CAAC,OAAO;QAAE,MAAM,IAAI,KAAK,CAAC,WAAW,SAAS,YAAY,CAAC,CAAC;IAChE,OAAO,CAAC,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAE7B,0EAA0E;IAC1E,oEAAoE;IACpE,IAAI,MAAM,CAAC,YAAY;QAAE,OAAO;IAEhC,MAAM,GAAG,GAAG,MAAM,CAAC,IAAI,CAAC,iBAAiB,CAAuB,CAAC;IACjE,IAAI,CAAC,GAAG;QAAE,OAAO;IAEjB,MAAM,OAAO,GAAG,WAAW,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC;IAC7C,MAAM,gBAAgB,GACpB,MAAM,CAAC,aAAa,KAAK,IAAI,IAAI,CAAC,MAAM,CAAC,aAAa,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;IAC7E,MAAM,gBAAgB,GACpB,OAAO,MAAM,CAAC,QAAQ,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,SAAS,CAAC,CAAC;IAC7E,MAAM,OAAO,GAAG,gBAAgB,IAAI,gBAAgB,CAAC;IAErD,0EAA0E;IAC1E,yEAAyE;IACzE,wEAAwE;IACxE,iDAAiD;IACjD,IAAI,OAAO,IAAI,CAAC,CAAC,OAAO,IAAI,MAAM,CAAC,gBAAgB,CAAC,EAAE,CAAC;QACrD,OAAO,CAAC,SAAS,CAAC,KAAK,CAAC,GAAG,EAAE,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;IACnE,CAAC;SAAM,IAAI,UAAU,CAAC,MAAM,CAAC,QAAQ,CAAC,EAAE,CAAC;QACvC,OAAO,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,UAAU,CAAC,CAAC;IACjD,CAAC;IAED,MAAM,QAAQ,GAAG,OAAO,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAE5C,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,OAAO,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE;YACzB,KAAK,EAAE,OAAO;gBACZ,CAAC,CAAC,mBAAmB;gBACrB,CAAC,CAAC,OAAO;oBACP,CAAC,CAAC,WAAW;oBACb,CAAC,CAAC,SAAS;YACf,eAAe,EAAE,MAAM,CAAC,UAAU;YAClC,cAAc,EAAE,MAAM,CAAC,UAAU;YACjC,SAAS,EAAE,MAAM,CAAC,aAAa;YAC/B,eAAe,EAAE,KAAK;YACtB,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YAC3B,cAAc,EAAE,KAAK;SACtB,CAAC,CAAC;IACL,CAAC;SAAM,CAAC;QACN,QAAQ,CAAC,cAAc,GAAG,MAAM,CAAC,UAAU,CAAC;QAC5C,QAAQ,CAAC,SAAS,GAAG,MAAM,CAAC,aAAa,IAAI,QAAQ,CAAC,SAAS,CAAC;QAEhE,IAAI,OAAO,EAAE,CAAC;YACZ,QAAQ,CAAC,UAAU,IAAI,CAAC,CAAC;QAC3B,CAAC;QAED,4EAA4E;QAC5E,IAAI,CAAC,QAAQ,CAAC,KAAK,KAAK,mBAAmB,IAAI,QAAQ,CAAC,KAAK,KAAK,iBAAiB,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YACjG,MAAM,cAAc,GAAG,OAAO,CAAC,OAAO,CAAC,IAAI,CACzC,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,UAAU,GAAG,QAAQ,CAAC,eAAe;gBACvC,CAAC,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU;gBAChC,UAAU,CAAC,CAAC,CAAC,QAAQ,CAAC,CACzB,CAAC;YACF,QAAQ,CAAC,eAAe,GAAG,cAAc,CAAC;QAC5C,CAAC;QAED,eAAe;QACf,IAAI,OAAO,EAAE,CAAC;YACZ,QAAQ,CAAC,KAAK,GAAG,mBAAmB,CAAC;QACvC,CAAC;aAAM,IAAI,OAAO,IAAI,CAAC,QAAQ,CAAC,cAAc,EAAE,CAAC;YAC/C,QAAQ,CAAC,KAAK,GAAG,WAAW,CAAC;QAC/B,CAAC;IACH,CAAC;AACH,CAAC;AAED,MAAM,UAAU,kBAAkB,CAChC,SAAiB,EACjB,GAAW;IAEX,MAAM,OAAO,GAAG,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IACxC,IAAI,CAAC,OAAO;QAAE,OAAO;IAErB,MAAM,QAAQ,GAAG,OAAO,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAC5C,IAAI,QAAQ,EAAE,CAAC;QACb,QAAQ,CAAC,KAAK,GAAG,iBAAiB,CAAC;QACnC,QAAQ,CAAC,cAAc,GAAG,IAAI,CAAC;IACjC,CAAC;SAAM,CAAC;QACN,OAAO,CAAC,SAAS,CAAC,GAAG,CAAC,GAAG,EAAE;YACzB,KAAK,EAAE,iBAAiB;YACxB,eAAe,EAAE,OAAO,CAAC,SAAS;YAClC,cAAc,EAAE,OAAO,CAAC,SAAS;YACjC,SAAS,EAAE,IAAI;YACf,eAAe,EAAE,KAAK;YACtB,UAAU,EAAE,CAAC;YACb,cAAc,EAAE,IAAI;SACrB,CAAC,CAAC;IACL,CAAC;AACH,CAAC;AAED,MAAM,UAAU,YAAY,CAAC,SAAiB;IAC5C,MAAM,OAAO,GAAG,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IACxC,IAAI,CAAC,OAAO;QAAE,MAAM,IAAI,KAAK,CAAC,WAAW,SAAS,YAAY,CAAC,CAAC;IAChE,OAAO,CAAC,SAAS,GAAG,CAAC,CAAC;IACtB,OAAO,CAAC,OAAO,GAAG,EAAE,CAAC;IACrB,OAAO,CAAC,SAAS,GAAG,IAAI,GAAG,EAAE,CAAC;IAC9B,OAAO,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;AAC5B,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,SAAiB;IAC7C,OAAO,QAAQ,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC;AACpC,CAAC;AAED,MAAM,UAAU,YAAY;IAC1B,OAAO,KAAK,CAAC,IAAI,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;AACvC,CAAC;AAED,MAAM,UAAU,WAAW,CAAC,QAAgB;IAC1C,OAAO,QAAQ,CAAC,UAAU,CAAC,OAAO,CAAC,CAAC;AACtC,CAAC;AAED,MAAM,UAAU,UAAU,CAAC,QAAgB;IACzC,MAAM,QAAQ,GAAG;QACf,MAAM;QACN,KAAK;QACL,MAAM;QACN,OAAO;QACP,QAAQ;QACR,OAAO;QACP,eAAe;QACf,OAAO;QACP,SAAS;QACT,QAAQ;KACT,CAAC;IACF,MAAM,KAAK,GAAG,QAAQ,CAAC,WAAW,EAAE,CAAC;IACrC,OAAO,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,CAAC;AACnD,CAAC"}

package/dist/hardening/patch-generator.d.ts

@@ -0,0 +1,12 @@
+import type { AuditReport } from "../core/report-generator.js";
+import type { ResilienceScore } from "../benchmark/score.js";
+export interface HardeningPatch {
+    category: string;
+    severity: "critical" | "warning" | "info";
+    trigger: string;
+    promptSnippet: string;
+    toolGuidance: string;
+}
+export declare function generatePatches(reports: AuditReport[], score: ResilienceScore): HardeningPatch[];
+export declare function formatPatchReport(patches: HardeningPatch[]): string;
+//# sourceMappingURL=patch-generator.d.ts.map