baselineos 0.2.0-beta.1

package/dist/core/llm-tracer.js

@@ -0,0 +1,241 @@
+import Langfuse from 'langfuse';
+import { trace } from '@opentelemetry/api';
+
+// src/core/llm-tracer.ts
+var PiiBlockedError = class extends Error {
+  constructor(types) {
+    super(`PII detected in prompt \u2014 blocked: ${types.join(", ")}`);
+    this.types = types;
+    this.name = "PiiBlockedError";
+  }
+};
+var PATTERNS = {
+  // API keys before email to avoid partial overlap on sk- prefixes
+  "api-key": /\b(sk-ant-[A-Za-z0-9-_]{20,}|sk-[A-Za-z0-9]{32,}|(?:ghp|gho|ghu|ghs|ghr)_[A-Za-z0-9]{36}|AKIA[0-9A-Z]{16})\b/g,
+  "email": /\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Za-z]{2,}\b/g,
+  "phone": /\b(?:\+?1[\s.-]?)?\(?[0-9]{3}\)?[\s.-][0-9]{3}[\s.-][0-9]{4}\b/g,
+  "ssn": /\b(?!000|666|9\d{2})\d{3}-(?!00)\d{2}-(?!0000)\d{4}\b/g,
+  "credit-card": /\b(?:4[0-9]{12}(?:[0-9]{3})?|5[1-5][0-9]{14}|3[47][0-9]{13}|6(?:011|5[0-9]{2})[0-9]{12})\b/g,
+  "ip-address": /\b(?:(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\.){3}(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\b/g
+};
+var PiiDetector = class {
+  mode;
+  disabled;
+  bus;
+  constructor(config = {}) {
+    this.mode = config.mode ?? "redact";
+    this.disabled = new Set(config.disabledTypes ?? []);
+    this.bus = config.bus;
+  }
+  /**
+   * Scan text for PII without side effects.
+   * Safe to call in any context — no bus events, no span mutations.
+   */
+  scan(text) {
+    const matches = [];
+    for (const [type, pattern] of Object.entries(PATTERNS)) {
+      if (this.disabled.has(type)) continue;
+      pattern.lastIndex = 0;
+      let m;
+      while ((m = pattern.exec(text)) !== null) {
+        matches.push({ type, start: m.index, end: m.index + m[0].length });
+      }
+    }
+    matches.sort((a, b) => a.start - b.start || b.end - a.end);
+    const deduped = [];
+    let cursor = 0;
+    for (const match of matches) {
+      if (match.start >= cursor) {
+        deduped.push(match);
+        cursor = match.end;
+      }
+    }
+    const types = [...new Set(deduped.map((m) => m.type))];
+    const clean = deduped.length === 0;
+    let redacted = "";
+    let pos = 0;
+    for (const match of deduped) {
+      redacted += text.slice(pos, match.start);
+      redacted += `[REDACTED:${match.type.toUpperCase()}]`;
+      pos = match.end;
+    }
+    redacted += text.slice(pos);
+    return { clean, matches: deduped, types, redacted };
+  }
+  /**
+   * Process text before LLM routing.
+   *
+   * - Publishes 'governance:pii-detected' on the bus when PII is found.
+   * - Sets pii.detected and pii.types on the active OTel span.
+   * - redact mode: returns redacted text.
+   * - block mode: throws PiiBlockedError if PII found.
+   *
+   * @param text    Text to scan (typically task.description).
+   * @param taskId  Used for bus event correlation.
+   * @returns       Safe text (redacted or original when clean).
+   */
+  processText(text, taskId) {
+    const result = this.scan(text);
+    if (!result.clean) {
+      const span = trace.getActiveSpan();
+      if (span) {
+        span.setAttribute("pii.detected", true);
+        span.setAttribute("pii.types", result.types.join(","));
+        span.setAttribute("pii.count", result.matches.length);
+      }
+      this.bus?.broadcast("system", "governance:pii-detected", {
+        taskId,
+        types: result.types,
+        count: result.matches.length,
+        blocked: this.mode === "block",
+        mode: this.mode
+      }, "high");
+      if (this.mode === "block") {
+        throw new PiiBlockedError(result.types);
+      }
+    }
+    return result.redacted;
+  }
+};
+
+// src/core/llm-tracer.ts
+var LlmTracer = class {
+  langfuse;
+  piiScanner;
+  curator;
+  constructor(config = {}) {
+    this.langfuse = new Langfuse({
+      publicKey: config.publicKey ?? process.env["LANGFUSE_PUBLIC_KEY"] ?? "baseline-public-key",
+      secretKey: config.secretKey ?? process.env["LANGFUSE_SECRET_KEY"] ?? "baseline-secret-key",
+      baseUrl: config.baseUrl ?? process.env["LANGFUSE_BASE_URL"] ?? "http://localhost:3001",
+      flushAt: config.flushAt ?? 15,
+      flushInterval: config.flushInterval ?? 3e4
+    });
+    this.piiScanner = config.enablePiiMasking ?? true ? new PiiDetector({ mode: "redact" }) : null;
+    this.curator = config.curator ?? null;
+  }
+  /** Redact PII from a string if masking is enabled. No-op when scanner is null. */
+  maskPii(text) {
+    return this.piiScanner ? this.piiScanner.scan(text).redacted : text;
+  }
+  /**
+   * Log a single LLM generation to Langfuse.
+   *
+   * Multiple calls with the same `traceId` accumulate under one trace,
+   * so all generations for a task are grouped together automatically.
+   *
+   * @param traceId   task.id — the top-level trace identifier
+   * @param traceName task.title — human-readable trace label in the UI
+   * @param options   generation details
+   */
+  logGeneration(traceId, traceName, options) {
+    try {
+      const trace2 = this.langfuse.trace({
+        id: traceId,
+        name: traceName.slice(0, 200)
+      });
+      const maskedSystemPrompt = options.systemPrompt ? this.maskPii(options.systemPrompt) : void 0;
+      const maskedMessages = options.inputMessages.map((m) => ({
+        role: m.role,
+        content: typeof m.content === "string" ? this.maskPii(m.content) : m.content
+      }));
+      const maskedOutput = this.maskPii(options.output);
+      const input = maskedSystemPrompt ? [{ role: "system", content: maskedSystemPrompt }, ...maskedMessages] : maskedMessages;
+      trace2.generation({
+        name: options.name,
+        model: options.model,
+        input,
+        output: maskedOutput,
+        usage: {
+          input: options.inputTokens,
+          output: options.outputTokens,
+          total: options.inputTokens + options.outputTokens,
+          unit: "TOKENS"
+        },
+        startTime: options.startTime,
+        endTime: options.endTime,
+        metadata: options.metadata
+      });
+      if (this.curator) {
+        const firstUserMsg = options.inputMessages.find((m) => m.role === "user");
+        const inputText = typeof firstUserMsg?.content === "string" ? firstUserMsg.content : "";
+        const confidence = typeof options.metadata?.["confidence"] === "number" ? options.metadata["confidence"] : 0.8;
+        this.curator.ingest({
+          traceId,
+          title: traceName.slice(0, 200),
+          input: inputText,
+          output: options.output,
+          confidence,
+          benign: options.metadata?.["benign"] !== false,
+          tokens: options.inputTokens + options.outputTokens,
+          timestamp: options.endTime.toISOString(),
+          verified: typeof options.metadata?.["verified"] === "boolean" ? options.metadata["verified"] : void 0
+        });
+      }
+    } catch {
+    }
+  }
+  /** Flush pending spans. Call on graceful shutdown. */
+  async flush() {
+    try {
+      await this.langfuse.flushAsync();
+    } catch {
+    }
+  }
+  /** Shutdown Langfuse client (flushes pending spans). */
+  shutdown() {
+    this.flush().catch(() => {
+    });
+  }
+};
+/**
+ * PII Detector — SIGNAL-015
+ *
+ * Scans prompt text for personally identifiable information before it
+ * reaches the LLM router. Supports two enforcement modes:
+ *
+ *   redact — replaces detected values with [REDACTED:TYPE] (default)
+ *   block  — throws PiiBlockedError, halting task execution
+ *
+ * Detected types:
+ *   email        — RFC 5321 address pattern
+ *   phone        — US/international format (10-digit minimum)
+ *   ssn          — US Social Security Number (NNN-NN-NNNN)
+ *   credit-card  — Visa, Mastercard, Amex, Discover
+ *   api-key      — OpenAI sk-, Anthropic sk-ant-, GitHub ghp_, AWS AKIA
+ *   ip-address   — IPv4 addresses
+ *
+ * Integration:
+ *   Called by Orchestrator._performExecution() before engine.execute().
+ *   Publishes 'governance:pii-detected' on the AgentBus when PII is found.
+ *   Sets pii.detected and pii.types on the active OTel span.
+ *
+ * @license Apache-2.0
+ */
+/**
+ * LLM Tracer — SIGNAL-014
+ *
+ * Logs every LLM call made by AnthropicEngine to Langfuse for prompt-level
+ * observability. Captures model, token usage, latency, input/output, and
+ * task metadata. Traces are keyed by task.id, so a multi-step task (execute
+ * → self-verify → review) produces one Langfuse trace with multiple
+ * generation spans.
+ *
+ * Self-hosted Langfuse:
+ *   docker compose -f docker/docker-compose.monitoring.yml up -d
+ *   → http://localhost:3001  (admin@baselineos.dev / baseline)
+ *
+ * Configuration (env or constructor):
+ *   LANGFUSE_PUBLIC_KEY   — project public key (default: baseline-public-key)
+ *   LANGFUSE_SECRET_KEY   — project secret key (default: baseline-secret-key)
+ *   LANGFUSE_BASE_URL     — Langfuse server URL (default: http://localhost:3001)
+ *
+ * The tracer is fail-safe: any Langfuse error is swallowed silently so it
+ * can never interrupt task execution.
+ *
+ * @license Apache-2.0
+ */
+
+export { LlmTracer };
+//# sourceMappingURL=llm-tracer.js.map
+//# sourceMappingURL=llm-tracer.js.map

package/dist/core/llm-tracer.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/core/pii-detector.ts","../../src/core/llm-tracer.ts"],"names":["trace"],"mappings":";;;;AAoEO,IAAM,eAAA,GAAN,cAA8B,KAAA,CAAM;AAAA,EACzC,YAA4B,KAAA,EAAkB;AAC5C,IAAA,KAAA,CAAM,CAAA,uCAAA,EAAqC,KAAA,CAAM,IAAA,CAAK,IAAI,CAAC,CAAA,CAAE,CAAA;AADnC,IAAA,IAAA,CAAA,KAAA,GAAA,KAAA;AAE1B,IAAA,IAAA,CAAK,IAAA,GAAO,iBAAA;AAAA,EACd;AACF,CAAA;AAIA,IAAM,QAAA,GAAoC;AAAA;AAAA,EAExC,SAAA,EAAW,+GAAA;AAAA,EACX,OAAA,EAAS,qDAAA;AAAA,EACT,OAAA,EAAS,iEAAA;AAAA,EACT,KAAA,EAAO,wDAAA;AAAA,EACP,aAAA,EAAe,6FAAA;AAAA,EACf,YAAA,EAAc;AAChB,CAAA;AAIO,IAAM,cAAN,MAAkB;AAAA,EACd,IAAA;AAAA,EACQ,QAAA;AAAA,EACA,GAAA;AAAA,EAEjB,WAAA,CAAY,MAAA,GAA4B,EAAC,EAAG;AAC1C,IAAA,IAAA,CAAK,IAAA,GAAO,OAAO,IAAA,IAAQ,QAAA;AAC3B,IAAA,IAAA,CAAK,WAAW,IAAI,GAAA,CAAI,MAAA,CAAO,aAAA,IAAiB,EAAE,CAAA;AAClD,IAAA,IAAA,CAAK,MAAM,MAAA,CAAO,GAAA;AAAA,EACpB;AAAA;AAAA;AAAA;AAAA;AAAA,EAMA,KAAK,IAAA,EAA6B;AAChC,IAAA,MAAM,UAAsB,EAAC;AAE7B,IAAA,KAAA,MAAW,CAAC,IAAA,EAAM,OAAO,KAAK,MAAA,CAAO,OAAA,CAAQ,QAAQ,CAAA,EAA+B;AAClF,MAAA,IAAI,IAAA,CAAK,QAAA,CAAS,GAAA,CAAI,IAAI,CAAA,EAAG;AAG7B,MAAA,OAAA,CAAQ,SAAA,GAAY,CAAA;AACpB,MAAA,IAAI,CAAA;AACJ,MAAA,OAAA,CAAQ,CAAA,GAAI,OAAA,CAAQ,IAAA,CAAK,IAAI,OAAO,IAAA,EAAM;AACxC,QAAA,OAAA,CAAQ,IAAA,CAAK,EAAE,IAAA,EAAM,KAAA,EAAO,CAAA,CAAE,KAAA,EAAO,GAAA,EAAK,CAAA,CAAE,KAAA,GAAQ,CAAA,CAAE,CAAC,CAAA,CAAE,QAAQ,CAAA;AAAA,MACnE;AAAA,IACF;AAGA,IAAA,OAAA,CAAQ,IAAA,CAAK,CAAC,CAAA,EAAG,CAAA,KAAM,CAAA,CAAE,KAAA,GAAQ,CAAA,CAAE,KAAA,IAAS,CAAA,CAAE,GAAA,GAAM,CAAA,CAAE,GAAG,CAAA;AACzD,IAAA,MAAM,UAAsB,EAAC;AAC7B,IAAA,IAAI,MAAA,GAAS,CAAA;AACb,IAAA,KAAA,MAAW,SAAS,OAAA,EAAS;AAC3B,MAAA,IAAI,KAAA,CAAM,SAAS,MAAA,EAAQ;AACzB,QAAA,OAAA,CAAQ,KAAK,KAAK,CAAA;AAClB,QAAA,MAAA,GAAS,KAAA,CAAM,GAAA;AAAA,MACjB;AAAA,IACF;AAEA,IAAA,MAAM,KAAA,GAAQ,CAAC,GAAG,IAAI,GAAA,CAAI,OAAA,CAAQ,GAAA,CAAI,CAAC,CAAA,KAAM,CAAA,CAAE,IAAI,CAAC,CAAC,CAAA;AACrD,IAAA,MAAM,KAAA,GAAQ,QAAQ,MAAA,KAAW,CAAA;AAGjC,IAAA,IAAI,QAAA,GAAW,EAAA;AACf,IAAA,IAAI,GAAA,GAAM,CAAA;AACV,IAAA,KAAA,MAAW,SAAS,OAAA,EAAS;AAC3B,MAAA,QAAA,IAAY,IAAA,CAAK,KAAA,CAAM,GAAA,EAAK,KAAA,CAAM,KAAK,CAAA;AACvC,MAAA,QAAA,IAAY,CAAA,UAAA,EAAa,KAAA,CAAM,IAAA,CAAK,WAAA,EAAa,CAAA,CAAA,CAAA;AACjD,MAAA,GAAA,GAAM,KAAA,CAAM,GAAA;AAAA,IACd;AACA,IAAA,QAAA,IAAY,IAAA,CAAK,MAAM,GAAG,CAAA;AAE1B,IAAA,OAAO,EAAE,KAAA,EAAO,OAAA,EAAS,OAAA,EAAS,OAAO,QAAA,EAAS;AAAA,EACpD;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAcA,WAAA,CAAY,MAAc,MAAA,EAAyB;AACjD,IAAA,MAAM,MAAA,GAAS,IAAA,CAAK,IAAA,CAAK,IAAI,CAAA;AAE7B,IAAA,IAAI,CAAC,OAAO,KAAA,EAAO;AAEjB,MAAA,MAAM,IAAA,GAAO,MAAM,aAAA,EAAc;AACjC,MAAA,IAAI,IAAA,EAAM;AACR,QAAA,IAAA,CAAK,YAAA,CAAa,gBAAgB,IAAI,CAAA;AACtC,QAAA,IAAA,CAAK,aAAa,WAAA,EAAa,MAAA,CAAO,KAAA,CAAM,IAAA,CAAK,GAAG,CAAC,CAAA;AACrD,QAAA,IAAA,CAAK,YAAA,CAAa,WAAA,EAAa,MAAA,CAAO,OAAA,CAAQ,MAAM,CAAA;AAAA,MACtD;AAGA,MAAA,IAAA,CAAK,GAAA,EAAK,SAAA,CAAU,QAAA,EAAU,yBAAA,EAA2B;AAAA,QACvD,MAAA;AAAA,QACA,OAAO,MAAA,CAAO,KAAA;AAAA,QACd,KAAA,EAAO,OAAO,OAAA,CAAQ,MAAA;AAAA,QACtB,OAAA,EAAS,KAAK,IAAA,KAAS,OAAA;AAAA,QACvB,MAAM,IAAA,CAAK;AAAA,SACV,MAAM,CAAA;AAET,MAAA,IAAI,IAAA,CAAK,SAAS,OAAA,EAAS;AACzB,QAAA,MAAM,IAAI,eAAA,CAAgB,MAAA,CAAO,KAAK,CAAA;AAAA,MACxC;AAAA,IACF;AAEA,IAAA,OAAO,MAAA,CAAO,QAAA;AAAA,EAChB;AACF,CAAA;;;AC9GO,IAAM,YAAN,MAAgB;AAAA,EACJ,QAAA;AAAA,EACA,UAAA;AAAA,EACA,OAAA;AAAA,EAEjB,WAAA,CAAY,MAAA,GAA0B,EAAC,EAAG;AACxC,IAAA,IAAA,CAAK,QAAA,GAAW,IAAI,QAAA,CAAS;AAAA,MAC3B,WAAc,MAAA,CAAO,SAAA,IAAiB,OAAA,CAAQ,GAAA,CAAI,qBAAqB,CAAA,IAAM,qBAAA;AAAA,MAC7E,WAAc,MAAA,CAAO,SAAA,IAAiB,OAAA,CAAQ,GAAA,CAAI,qBAAqB,CAAA,IAAM,qBAAA;AAAA,MAC7E,SAAc,MAAA,CAAO,OAAA,IAAiB,OAAA,CAAQ,GAAA,CAAI,mBAAmB,CAAA,IAAQ,uBAAA;AAAA,MAC7E,OAAA,EAAc,OAAO,OAAA,IAAiB,EAAA;AAAA,MACtC,aAAA,EAAe,OAAO,aAAA,IAAiB;AAAA,KACxC,CAAA;AAED,IAAA,IAAA,CAAK,UAAA,GAAc,MAAA,CAAO,gBAAA,IAAoB,IAAA,GAC1C,IAAI,YAAY,EAAE,IAAA,EAAM,QAAA,EAAU,CAAA,GAClC,IAAA;AACJ,IAAA,IAAA,CAAK,OAAA,GAAU,OAAO,OAAA,IAAW,IAAA;AAAA,EACnC;AAAA;AAAA,EAGQ,QAAQ,IAAA,EAAsB;AACpC,IAAA,OAAO,KAAK,UAAA,GAAa,IAAA,CAAK,WAAW,IAAA,CAAK,IAAI,EAAE,QAAA,GAAW,IAAA;AAAA,EACjE;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA;AAAA,EAYA,aAAA,CACE,OAAA,EACA,SAAA,EACA,OAAA,EACM;AACN,IAAA,IAAI;AACF,MAAA,MAAMA,MAAAA,GAAQ,IAAA,CAAK,QAAA,CAAS,KAAA,CAAM;AAAA,QAChC,EAAA,EAAM,OAAA;AAAA,QACN,IAAA,EAAM,SAAA,CAAU,KAAA,CAAM,CAAA,EAAG,GAAG;AAAA,OAC7B,CAAA;AAGD,MAAA,MAAM,qBAAqB,OAAA,CAAQ,YAAA,GAC/B,KAAK,OAAA,CAAQ,OAAA,CAAQ,YAAY,CAAA,GACjC,KAAA,CAAA;AACJ,MAAA,MAAM,cAAA,GAAiB,OAAA,CAAQ,aAAA,CAAc,GAAA,CAAI,CAAC,CAAA,MAAO;AAAA,QACvD,MAAM,CAAA,CAAE,IAAA;AAAA,QACR,OAAA,EAAS,OAAO,CAAA,CAAE,OAAA,KAAY,QAAA,GAAW,KAAK,OAAA,CAAQ,CAAA,CAAE,OAAO,CAAA,GAAI,CAAA,CAAE;AAAA,OACvE,CAAE,CAAA;AACF,MAAA,MAAM,YAAA,GAAe,IAAA,CAAK,OAAA,CAAQ,OAAA,CAAQ,MAAM,CAAA;AAEhD,MAAA,MAAM,KAAA,GAAQ,kBAAA,GACV,CAAC,EAAE,IAAA,EAAM,QAAA,EAAU,OAAA,EAAS,kBAAA,EAAmB,EAAG,GAAG,cAAc,CAAA,GACnE,cAAA;AAEJ,MAAAA,OAAM,UAAA,CAAW;AAAA,QACf,MAAW,OAAA,CAAQ,IAAA;AAAA,QACnB,OAAW,OAAA,CAAQ,KAAA;AAAA,QACnB,KAAA;AAAA,QACA,MAAA,EAAW,YAAA;AAAA,QACX,KAAA,EAAO;AAAA,UACL,OAAQ,OAAA,CAAQ,WAAA;AAAA,UAChB,QAAQ,OAAA,CAAQ,YAAA;AAAA,UAChB,KAAA,EAAQ,OAAA,CAAQ,WAAA,GAAc,OAAA,CAAQ,YAAA;AAAA,UACtC,IAAA,EAAQ;AAAA,SACV;AAAA,QACA,WAAW,OAAA,CAAQ,SAAA;AAAA,QACnB,SAAW,OAAA,CAAQ,OAAA;AAAA,QACnB,UAAW,OAAA,CAAQ;AAAA,OACpB,CAAA;AAGD,MAAA,IAAI,KAAK,OAAA,EAAS;AAChB,QAAA,MAAM,YAAA,GAAe,QAAQ,aAAA,CAAc,IAAA,CAAK,CAAC,CAAA,KAAM,CAAA,CAAE,SAAS,MAAM,CAAA;AACxE,QAAA,MAAM,YAAY,OAAO,YAAA,EAAc,OAAA,KAAY,QAAA,GAAW,aAAa,OAAA,GAAU,EAAA;AACrF,QAAA,MAAM,UAAA,GAAa,OAAO,OAAA,CAAQ,QAAA,GAAW,YAAY,MAAM,QAAA,GAC1D,OAAA,CAAQ,QAAA,CAAS,YAAY,CAAA,GAC9B,GAAA;AACJ,QAAA,IAAA,CAAK,QAAQ,MAAA,CAAO;AAAA,UAClB,OAAA;AAAA,UACA,KAAA,EAAO,SAAA,CAAU,KAAA,CAAM,CAAA,EAAG,GAAG,CAAA;AAAA,UAC7B,KAAA,EAAO,SAAA;AAAA,UACP,QAAQ,OAAA,CAAQ,MAAA;AAAA,UAChB,UAAA;AAAA,UACA,MAAA,EAAQ,OAAA,CAAQ,QAAA,GAAW,QAAQ,CAAA,KAAM,KAAA;AAAA,UACzC,MAAA,EAAQ,OAAA,CAAQ,WAAA,GAAc,OAAA,CAAQ,YAAA;AAAA,UACtC,SAAA,EAAW,OAAA,CAAQ,OAAA,CAAQ,WAAA,EAAY;AAAA,UACvC,QAAA,EAAU,OAAO,OAAA,CAAQ,QAAA,GAAW,UAAU,MAAM,SAAA,GAC/C,OAAA,CAAQ,QAAA,CAAS,UAAU,CAAA,GAC5B,KAAA;AAAA,SACL,CAAA;AAAA,MACH;AAAA,IACF,CAAA,CAAA,MAAQ;AAAA,IAER;AAAA,EACF;AAAA;AAAA,EAGA,MAAM,KAAA,GAAuB;AAC3B,IAAA,IAAI;AACF,MAAA,MAAM,IAAA,CAAK,SAAS,UAAA,EAAW;AAAA,IACjC,CAAA,CAAA,MAAQ;AAAA,IAER;AAAA,EACF;AAAA;AAAA,EAGA,QAAA,GAAiB;AACf,IAAA,IAAA,CAAK,KAAA,EAAM,CAAE,KAAA,CAAM,MAAM;AAAA,IAAC,CAAC,CAAA;AAAA,EAC7B;AACF","file":"llm-tracer.js","sourcesContent":["/**\n * PII Detector — SIGNAL-015\n *\n * Scans prompt text for personally identifiable information before it\n * reaches the LLM router. Supports two enforcement modes:\n *\n *   redact — replaces detected values with [REDACTED:TYPE] (default)\n *   block  — throws PiiBlockedError, halting task execution\n *\n * Detected types:\n *   email        — RFC 5321 address pattern\n *   phone        — US/international format (10-digit minimum)\n *   ssn          — US Social Security Number (NNN-NN-NNNN)\n *   credit-card  — Visa, Mastercard, Amex, Discover\n *   api-key      — OpenAI sk-, Anthropic sk-ant-, GitHub ghp_, AWS AKIA\n *   ip-address   — IPv4 addresses\n *\n * Integration:\n *   Called by Orchestrator._performExecution() before engine.execute().\n *   Publishes 'governance:pii-detected' on the AgentBus when PII is found.\n *   Sets pii.detected and pii.types on the active OTel span.\n *\n * @license Apache-2.0\n */\n\nimport { trace } from '@opentelemetry/api';\nimport type { AgentBus } from './agent-bus.js';\n\n// ─── Types ────────────────────────────────────────────────────────────────────\n\nexport type PiiType =\n  | 'email'\n  | 'phone'\n  | 'ssn'\n  | 'credit-card'\n  | 'api-key'\n  | 'ip-address';\n\nexport type PiiMode = 'redact' | 'block';\n\nexport interface PiiMatch {\n  type: PiiType;\n  start: number;\n  end: number;\n}\n\nexport interface PiiScanResult {\n  /** true when no PII was detected */\n  clean: boolean;\n  /** all matches found, sorted by start position */\n  matches: PiiMatch[];\n  /** unique PII types present */\n  types: PiiType[];\n  /** input text with all PII replaced by [REDACTED:TYPE]; equals input when clean=true */\n  redacted: string;\n}\n\nexport interface PiiDetectorConfig {\n  /** Enforcement mode. Default: 'redact' */\n  mode?: PiiMode;\n  /** PII types to skip. Useful for staged rollouts. */\n  disabledTypes?: PiiType[];\n  /** AgentBus for publishing governance:pii-detected events. */\n  bus?: AgentBus;\n}\n\n// ─── Error ────────────────────────────────────────────────────────────────────\n\nexport class PiiBlockedError extends Error {\n  constructor(public readonly types: PiiType[]) {\n    super(`PII detected in prompt — blocked: ${types.join(', ')}`);\n    this.name = 'PiiBlockedError';\n  }\n}\n\n// ─── Patterns ────────────────────────────────────────────────────────────────\n\nconst PATTERNS: Record<PiiType, RegExp> = {\n  // API keys before email to avoid partial overlap on sk- prefixes\n  'api-key': /\\b(sk-ant-[A-Za-z0-9-_]{20,}|sk-[A-Za-z0-9]{32,}|(?:ghp|gho|ghu|ghs|ghr)_[A-Za-z0-9]{36}|AKIA[0-9A-Z]{16})\\b/g,\n  'email': /\\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\\.[A-Za-z]{2,}\\b/g,\n  'phone': /\\b(?:\\+?1[\\s.-]?)?\\(?[0-9]{3}\\)?[\\s.-][0-9]{3}[\\s.-][0-9]{4}\\b/g,\n  'ssn': /\\b(?!000|666|9\\d{2})\\d{3}-(?!00)\\d{2}-(?!0000)\\d{4}\\b/g,\n  'credit-card': /\\b(?:4[0-9]{12}(?:[0-9]{3})?|5[1-5][0-9]{14}|3[47][0-9]{13}|6(?:011|5[0-9]{2})[0-9]{12})\\b/g,\n  'ip-address': /\\b(?:(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\\.){3}(?:25[0-5]|2[0-4][0-9]|[01]?[0-9][0-9]?)\\b/g,\n};\n\n// ─── PiiDetector ─────────────────────────────────────────────────────────────\n\nexport class PiiDetector {\n  readonly mode: PiiMode;\n  private readonly disabled: Set<PiiType>;\n  private readonly bus?: AgentBus;\n\n  constructor(config: PiiDetectorConfig = {}) {\n    this.mode = config.mode ?? 'redact';\n    this.disabled = new Set(config.disabledTypes ?? []);\n    this.bus = config.bus;\n  }\n\n  /**\n   * Scan text for PII without side effects.\n   * Safe to call in any context — no bus events, no span mutations.\n   */\n  scan(text: string): PiiScanResult {\n    const matches: PiiMatch[] = [];\n\n    for (const [type, pattern] of Object.entries(PATTERNS) as Array<[PiiType, RegExp]>) {\n      if (this.disabled.has(type)) continue;\n\n      // Reset lastIndex — patterns are defined with /g flag\n      pattern.lastIndex = 0;\n      let m: RegExpExecArray | null;\n      while ((m = pattern.exec(text)) !== null) {\n        matches.push({ type, start: m.index, end: m.index + m[0].length });\n      }\n    }\n\n    // Sort by start position, resolve overlaps (keep longest match)\n    matches.sort((a, b) => a.start - b.start || b.end - a.end);\n    const deduped: PiiMatch[] = [];\n    let cursor = 0;\n    for (const match of matches) {\n      if (match.start >= cursor) {\n        deduped.push(match);\n        cursor = match.end;\n      }\n    }\n\n    const types = [...new Set(deduped.map((m) => m.type))];\n    const clean = deduped.length === 0;\n\n    // Build redacted string\n    let redacted = '';\n    let pos = 0;\n    for (const match of deduped) {\n      redacted += text.slice(pos, match.start);\n      redacted += `[REDACTED:${match.type.toUpperCase()}]`;\n      pos = match.end;\n    }\n    redacted += text.slice(pos);\n\n    return { clean, matches: deduped, types, redacted };\n  }\n\n  /**\n   * Process text before LLM routing.\n   *\n   * - Publishes 'governance:pii-detected' on the bus when PII is found.\n   * - Sets pii.detected and pii.types on the active OTel span.\n   * - redact mode: returns redacted text.\n   * - block mode: throws PiiBlockedError if PII found.\n   *\n   * @param text    Text to scan (typically task.description).\n   * @param taskId  Used for bus event correlation.\n   * @returns       Safe text (redacted or original when clean).\n   */\n  processText(text: string, taskId?: string): string {\n    const result = this.scan(text);\n\n    if (!result.clean) {\n      // Set OTel span attributes on the active span (task.perform)\n      const span = trace.getActiveSpan();\n      if (span) {\n        span.setAttribute('pii.detected', true);\n        span.setAttribute('pii.types', result.types.join(','));\n        span.setAttribute('pii.count', result.matches.length);\n      }\n\n      // Publish governance event\n      this.bus?.broadcast('system', 'governance:pii-detected', {\n        taskId,\n        types: result.types,\n        count: result.matches.length,\n        blocked: this.mode === 'block',\n        mode: this.mode,\n      }, 'high');\n\n      if (this.mode === 'block') {\n        throw new PiiBlockedError(result.types);\n      }\n    }\n\n    return result.redacted;\n  }\n}\n","/**\n * LLM Tracer — SIGNAL-014\n *\n * Logs every LLM call made by AnthropicEngine to Langfuse for prompt-level\n * observability. Captures model, token usage, latency, input/output, and\n * task metadata. Traces are keyed by task.id, so a multi-step task (execute\n * → self-verify → review) produces one Langfuse trace with multiple\n * generation spans.\n *\n * Self-hosted Langfuse:\n *   docker compose -f docker/docker-compose.monitoring.yml up -d\n *   → http://localhost:3001  (admin@baselineos.dev / baseline)\n *\n * Configuration (env or constructor):\n *   LANGFUSE_PUBLIC_KEY   — project public key (default: baseline-public-key)\n *   LANGFUSE_SECRET_KEY   — project secret key (default: baseline-secret-key)\n *   LANGFUSE_BASE_URL     — Langfuse server URL (default: http://localhost:3001)\n *\n * The tracer is fail-safe: any Langfuse error is swallowed silently so it\n * can never interrupt task execution.\n *\n * @license Apache-2.0\n */\n\nimport Langfuse from 'langfuse';\nimport { PiiDetector } from './pii-detector.js';\nimport type { TraceCurator } from './trace-curator.js';\n\n// ─── Types ────────────────────────────────────────────────────────────────────\n\nexport interface LlmTracerConfig {\n  /** Langfuse project public key. Default: LANGFUSE_PUBLIC_KEY env or 'baseline-public-key' */\n  publicKey?: string;\n  /** Langfuse project secret key. Default: LANGFUSE_SECRET_KEY env or 'baseline-secret-key' */\n  secretKey?: string;\n  /** Langfuse server base URL. Default: LANGFUSE_BASE_URL env or http://localhost:3001 */\n  baseUrl?: string;\n  /** Flush batch size (default: 15) */\n  flushAt?: number;\n  /** Flush interval in ms (default: 30_000) */\n  flushInterval?: number;\n  /**\n   * Scan and redact PII from system prompts, input messages, and output\n   * before sending to Langfuse. Default: true (SIGNAL-037).\n   */\n  enablePiiMasking?: boolean;\n  /**\n   * TraceCurator to ingest sampled traces into the ground-truth dataset.\n   * When provided, each logGeneration() call feeds a TraceRecord into the\n   * curator buffer for downstream curation and eval dataset expansion (SIGNAL-049).\n   */\n  curator?: TraceCurator;\n}\n\nexport interface LlmGenerationOptions {\n  /** Span name within the trace, e.g. 'execute', 'self-verify', 'review', 'agent-loop:3' */\n  name: string;\n  /** Model identifier, e.g. 'claude-sonnet-4-6' */\n  model: string;\n  /** System prompt sent to the model */\n  systemPrompt?: string;\n  /** User + assistant turns passed to the model */\n  inputMessages: Array<{ role: string; content: unknown }>;\n  /** Raw text output from the model */\n  output: string;\n  inputTokens: number;\n  outputTokens: number;\n  startTime: Date;\n  endTime: Date;\n  /** Arbitrary metadata attached to the generation (taskId, agentId, etc.) */\n  metadata?: Record<string, unknown>;\n}\n\n// ─── LlmTracer ────────────────────────────────────────────────────────────────\n\nexport class LlmTracer {\n  private readonly langfuse: Langfuse;\n  private readonly piiScanner: PiiDetector | null;\n  private readonly curator: TraceCurator | null;\n\n  constructor(config: LlmTracerConfig = {}) {\n    this.langfuse = new Langfuse({\n      publicKey:    config.publicKey     ?? process.env['LANGFUSE_PUBLIC_KEY']  ?? 'baseline-public-key',\n      secretKey:    config.secretKey     ?? process.env['LANGFUSE_SECRET_KEY']  ?? 'baseline-secret-key',\n      baseUrl:      config.baseUrl       ?? process.env['LANGFUSE_BASE_URL']    ?? 'http://localhost:3001',\n      flushAt:      config.flushAt       ?? 15,\n      flushInterval: config.flushInterval ?? 30_000,\n    });\n    // PII masking enabled by default (SIGNAL-037)\n    this.piiScanner = (config.enablePiiMasking ?? true)\n      ? new PiiDetector({ mode: 'redact' })\n      : null;\n    this.curator = config.curator ?? null;\n  }\n\n  /** Redact PII from a string if masking is enabled. No-op when scanner is null. */\n  private maskPii(text: string): string {\n    return this.piiScanner ? this.piiScanner.scan(text).redacted : text;\n  }\n\n  /**\n   * Log a single LLM generation to Langfuse.\n   *\n   * Multiple calls with the same `traceId` accumulate under one trace,\n   * so all generations for a task are grouped together automatically.\n   *\n   * @param traceId   task.id — the top-level trace identifier\n   * @param traceName task.title — human-readable trace label in the UI\n   * @param options   generation details\n   */\n  logGeneration(\n    traceId: string,\n    traceName: string,\n    options: LlmGenerationOptions,\n  ): void {\n    try {\n      const trace = this.langfuse.trace({\n        id:   traceId,\n        name: traceName.slice(0, 200),\n      });\n\n      // Redact PII before sending to Langfuse (SIGNAL-037)\n      const maskedSystemPrompt = options.systemPrompt\n        ? this.maskPii(options.systemPrompt)\n        : undefined;\n      const maskedMessages = options.inputMessages.map((m) => ({\n        role: m.role,\n        content: typeof m.content === 'string' ? this.maskPii(m.content) : m.content,\n      }));\n      const maskedOutput = this.maskPii(options.output);\n\n      const input = maskedSystemPrompt\n        ? [{ role: 'system', content: maskedSystemPrompt }, ...maskedMessages]\n        : maskedMessages;\n\n      trace.generation({\n        name:      options.name,\n        model:     options.model,\n        input,\n        output:    maskedOutput,\n        usage: {\n          input:  options.inputTokens,\n          output: options.outputTokens,\n          total:  options.inputTokens + options.outputTokens,\n          unit:   'TOKENS',\n        },\n        startTime: options.startTime,\n        endTime:   options.endTime,\n        metadata:  options.metadata,\n      });\n\n      // Feed into TraceCurator for ground-truth dataset expansion (SIGNAL-049)\n      if (this.curator) {\n        const firstUserMsg = options.inputMessages.find((m) => m.role === 'user');\n        const inputText = typeof firstUserMsg?.content === 'string' ? firstUserMsg.content : '';\n        const confidence = typeof options.metadata?.['confidence'] === 'number'\n          ? (options.metadata['confidence'] as number)\n          : 0.8;\n        this.curator.ingest({\n          traceId,\n          title: traceName.slice(0, 200),\n          input: inputText,\n          output: options.output,\n          confidence,\n          benign: options.metadata?.['benign'] !== false,\n          tokens: options.inputTokens + options.outputTokens,\n          timestamp: options.endTime.toISOString(),\n          verified: typeof options.metadata?.['verified'] === 'boolean'\n            ? (options.metadata['verified'] as boolean)\n            : undefined,\n        });\n      }\n    } catch {\n      // Silently ignore — tracer must never interrupt task execution\n    }\n  }\n\n  /** Flush pending spans. Call on graceful shutdown. */\n  async flush(): Promise<void> {\n    try {\n      await this.langfuse.flushAsync();\n    } catch {\n      // Silently ignore\n    }\n  }\n\n  /** Shutdown Langfuse client (flushes pending spans). */\n  shutdown(): void {\n    this.flush().catch(() => {});\n  }\n}\n"]}

package/dist/core/memory.d.ts

@@ -0,0 +1,86 @@
+/**
+ * BaselineOS Memory System
+ *
+ * Multi-scope persistent memory for AI agents.
+ *
+ * Scopes:
+ * - Working: Per-request, cleared after each call (fastest)
+ * - Session: Per-conversation, cleared on disconnect
+ * - Long-term: Persists across sessions (SQLite)
+ * - Shared: Visible across all agents and sessions
+ *
+ * @license Apache-2.0
+ */
+type MemoryScope = 'working' | 'session' | 'long-term' | 'shared';
+interface MemoryConfig {
+    persistPath: string;
+}
+interface StoreOptions {
+    scope?: MemoryScope;
+    sessionId?: string;
+    ttl?: number;
+    tags?: string[];
+    taskId?: string;
+    repo?: string;
+    persona?: string;
+}
+interface RetrieveOptions {
+    scope?: MemoryScope | 'all';
+    sessionId?: string;
+    taskId?: string;
+    repo?: string;
+    persona?: string;
+}
+interface MemoryEntry {
+    key: string;
+    value: unknown;
+    scope: MemoryScope;
+    sessionId?: string;
+    taskId?: string;
+    repo?: string;
+    persona?: string;
+    tags: string[];
+    createdAt: number;
+    updatedAt: number;
+    expiresAt?: number;
+    accessCount: number;
+}
+declare class MemorySystem {
+    private config;
+    private workingMemory;
+    private sessionMemory;
+    private longTermMemory;
+    private sharedMemory;
+    private db;
+    private initialized;
+    constructor(config?: Partial<MemoryConfig>);
+    initialize(): Promise<void>;
+    store(key: string, value: unknown, options?: StoreOptions): Promise<void>;
+    retrieve(key: string, options?: RetrieveOptions): Promise<unknown>;
+    delete(key: string, options?: RetrieveOptions): Promise<boolean>;
+    search(query: {
+        tags?: string[];
+        prefix?: string;
+        scope?: MemoryScope;
+        sessionId?: string;
+        taskId?: string;
+        repo?: string;
+        persona?: string;
+    }): Promise<MemoryEntry[]>;
+    clearWorking(): void;
+    clearSession(sessionId?: string): void;
+    flush(): Promise<void>;
+    close(): void;
+    private persistEntry;
+    private getFromScope;
+    private buildStorageKey;
+    private matchesContext;
+    getStats(): {
+        working: number;
+        session: number;
+        longTerm: number;
+        shared: number;
+    };
+}
+
+export { type MemoryConfig, type MemoryEntry, type MemoryScope, MemorySystem, type RetrieveOptions, type StoreOptions };