nexus-agents 2.154.1 → 2.155.0

package/dist/{chunk-3P5WC7AV.js → chunk-KXTOVEZS.js}

@@ -18,7 +18,7 @@ import {
   DEFAULT_TASK_TTL_MS,
   DEFAULT_TOOL_RATE_LIMITS,
   clampTaskTtl
-} from "./chunk-RW5ZE7IC.js";
+} from "./chunk-MJRIWG2E.js";
 import {
   executeExpert
 } from "./chunk-F5R53HDK.js";
@@ -46487,6 +46487,326 @@ function getShadowSink() {
   return singletonSink;
 }
 
+// src/orchestration/meta-strategy-eval.ts
+function groupByStrategy(corpus) {
+  const groups = /* @__PURE__ */ new Map();
+  for (const entry of corpus) {
+    const list = groups.get(entry.expectedStrategy) ?? [];
+    list.push(entry);
+    groups.set(entry.expectedStrategy, list);
+  }
+  return groups;
+}
+function splitCorpus(corpus, testRatio) {
+  const train = [];
+  const test = [];
+  for (const entries of groupByStrategy(corpus).values()) {
+    const testN = Math.max(1, Math.round(entries.length * testRatio));
+    const trainN = Math.max(0, entries.length - testN);
+    train.push(...entries.slice(0, trainN));
+    test.push(...entries.slice(trainN));
+  }
+  return { train, test };
+}
+function evaluateMetaStrategy(corpus, options = {}) {
+  const testRatio = options.testRatio ?? 0.25;
+  const { train, test } = splitCorpus(corpus, testRatio);
+  const orchestrator = createMetaOrchestrator();
+  const learned = createLearnedStrategySelector();
+  for (const entry of train) {
+    const decision = orchestrator.select({ goal: entry.goal });
+    learned.recordOutcome(entry.expectedStrategy, decision, true);
+  }
+  let rulesCorrect = 0;
+  let learnedCorrect = 0;
+  for (const entry of test) {
+    const decision = orchestrator.select({ goal: entry.goal });
+    if (decision.strategy === entry.expectedStrategy) rulesCorrect++;
+    if (learned.predict(decision).strategy === entry.expectedStrategy) learnedCorrect++;
+  }
+  const testCount = test.length;
+  const rulesAccuracy = testCount === 0 ? 0 : rulesCorrect / testCount;
+  const learnedAccuracy = testCount === 0 ? 0 : learnedCorrect / testCount;
+  return {
+    total: corpus.length,
+    trainCount: train.length,
+    testCount,
+    rulesAccuracy,
+    learnedAccuracy,
+    delta: learnedAccuracy - rulesAccuracy
+  };
+}
+
+// src/orchestration/meta-strategy-corpus.ts
+var META_STRATEGY_CORPUS = [
+  // single-shot — trivial single-step
+  { goal: "rename the variable foo to bar in utils.ts", expectedStrategy: "single-shot" },
+  { goal: "what does the git rebase --onto flag do?", expectedStrategy: "single-shot" },
+  { goal: "format this JSON snippet", expectedStrategy: "single-shot" },
+  { goal: "add a one-line comment explaining this regex", expectedStrategy: "single-shot" },
+  { goal: "convert this value from Celsius to Fahrenheit", expectedStrategy: "single-shot" },
+  {
+    goal: "explain what the useEffect hook does in one paragraph",
+    expectedStrategy: "single-shot"
+  },
+  { goal: "convert this markdown table to CSV", expectedStrategy: "single-shot" },
+  {
+    goal: "rename the function calcTotal to computeTotal in cart.ts",
+    expectedStrategy: "single-shot"
+  },
+  { goal: "what is the time complexity of binary search?", expectedStrategy: "single-shot" },
+  { goal: "format this SQL query for readability", expectedStrategy: "single-shot" },
+  // dev-pipeline — code change needing the test/lint/typecheck gate
+  {
+    goal: "fix the off-by-one bug in pagination.ts and make sure the tests pass",
+    expectedStrategy: "dev-pipeline"
+  },
+  {
+    goal: "add input validation to the login handler with unit tests",
+    expectedStrategy: "dev-pipeline"
+  },
+  { goal: "refactor the auth module and run lint and typecheck", expectedStrategy: "dev-pipeline" },
+  { goal: "implement retry logic in the http client with tests", expectedStrategy: "dev-pipeline" },
+  {
+    goal: "patch the null-pointer in parser.ts and verify the build",
+    expectedStrategy: "dev-pipeline"
+  },
+  {
+    goal: "fix the memory leak in the websocket handler and make the tests green",
+    expectedStrategy: "dev-pipeline"
+  },
+  {
+    goal: "add a rate limiter to the API middleware with unit tests",
+    expectedStrategy: "dev-pipeline"
+  },
+  {
+    goal: "correct the timezone bug in the scheduler and run the test suite",
+    expectedStrategy: "dev-pipeline"
+  },
+  {
+    goal: "implement pagination on the users endpoint with tests and typecheck",
+    expectedStrategy: "dev-pipeline"
+  },
+  {
+    goal: "fix the flaky retry test in queue.ts and keep lint clean",
+    expectedStrategy: "dev-pipeline"
+  },
+  // pipeline — multi-stage templated audit/general
+  { goal: "run a security audit of the payments module", expectedStrategy: "pipeline" },
+  { goal: "do a full quality audit of the API layer", expectedStrategy: "pipeline" },
+  { goal: "produce an architecture review of the data pipeline", expectedStrategy: "pipeline" },
+  { goal: "run the documentation-quality audit over the docs tree", expectedStrategy: "pipeline" },
+  { goal: "perform a compliance audit of the logging subsystem", expectedStrategy: "pipeline" },
+  { goal: "run a performance audit of the checkout flow", expectedStrategy: "pipeline" },
+  { goal: "do a dependency-vulnerability audit of the whole repo", expectedStrategy: "pipeline" },
+  { goal: "produce an accessibility audit of the web frontend", expectedStrategy: "pipeline" },
+  { goal: "run a code-quality audit across the billing service", expectedStrategy: "pipeline" },
+  { goal: "perform a test-coverage audit of the core package", expectedStrategy: "pipeline" },
+  // graph-workflow — DAG / conditional-edge
+  {
+    goal: "build a conditional workflow that branches on the test result then deploys or rolls back",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "orchestrate a DAG fetch then transform then validate then load, halting at the validate gate on failure",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "create a multi-stage workflow with conditional edges depending on the lint outcome",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "wire a workflow where step C runs only if both A and B succeed",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "design a graph workflow with a fan-in join after three sequential gated branches",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "build a workflow that runs A then branches to B or C based on the security-scan verdict",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "create a DAG where build and lint run in parallel then merge into a gated deploy step",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "design a workflow with a retry-loop edge back to the fetch step on transient failure",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "wire a pipeline that skips the migration step when the schema-check passes",
+    expectedStrategy: "graph-workflow"
+  },
+  {
+    goal: "orchestrate a conditional graph: on approval go to publish, otherwise route to a rework node",
+    expectedStrategy: "graph-workflow"
+  },
+  // orchestrate — pattern-based multi-agent
+  {
+    goal: "run a multi-agent wave over these modules to refactor them in parallel",
+    expectedStrategy: "orchestrate"
+  },
+  {
+    goal: "fan out independent subtasks across the codebase to add logging",
+    expectedStrategy: "orchestrate"
+  },
+  {
+    goal: "orchestrate a swarm of agents to triage the open issues",
+    expectedStrategy: "orchestrate"
+  },
+  {
+    goal: "use a multi-agent pattern to migrate each service independently",
+    expectedStrategy: "orchestrate"
+  },
+  { goal: "dispatch a wave of agents to audit each microservice", expectedStrategy: "orchestrate" },
+  {
+    goal: "spawn a swarm of agents to add type annotations across every module in parallel",
+    expectedStrategy: "orchestrate"
+  },
+  {
+    goal: "fan out a wave of agents to update the copyright header in each file",
+    expectedStrategy: "orchestrate"
+  },
+  {
+    goal: "use a multi-agent pattern to generate unit tests for each untested component concurrently",
+    expectedStrategy: "orchestrate"
+  },
+  {
+    goal: "dispatch parallel agents to translate the docs into each supported language",
+    expectedStrategy: "orchestrate"
+  },
+  {
+    goal: "run a fan-out of agents to bump the dependency version in every package independently",
+    expectedStrategy: "orchestrate"
+  },
+  // consensus — multi-perspective decision/vote
+  { goal: "hold a consensus vote on whether to adopt GraphQL", expectedStrategy: "consensus" },
+  { goal: "we need a consensus decision on the database choice", expectedStrategy: "consensus" },
+  {
+    goal: "get multiple perspectives via a vote on the API redesign",
+    expectedStrategy: "consensus"
+  },
+  {
+    goal: "run a consensus panel on the proposed authentication change",
+    expectedStrategy: "consensus"
+  },
+  {
+    goal: "do a multi-perspective review and vote on the migration plan",
+    expectedStrategy: "consensus"
+  },
+  { goal: "hold a vote on which cloud provider to standardize on", expectedStrategy: "consensus" },
+  {
+    goal: "get a consensus decision on whether to drop support for Node 18",
+    expectedStrategy: "consensus"
+  },
+  {
+    goal: "run a multi-perspective panel to decide the caching strategy",
+    expectedStrategy: "consensus"
+  },
+  { goal: "we need a consensus vote on the proposed pricing model", expectedStrategy: "consensus" },
+  {
+    goal: "gather multiple expert opinions and vote on the rollout timeline",
+    expectedStrategy: "consensus"
+  },
+  // spec — greenfield from a written spec
+  { goal: "build a greenfield todo app from this written spec", expectedStrategy: "spec" },
+  {
+    goal: "implement a new microservice from scratch per the attached spec document",
+    expectedStrategy: "spec"
+  },
+  { goal: "create a brand-new CLI tool from this specification", expectedStrategy: "spec" },
+  { goal: "scaffold a greenfield REST API from the provided spec", expectedStrategy: "spec" },
+  {
+    goal: "build the new notification service from scratch following the spec",
+    expectedStrategy: "spec"
+  },
+  {
+    goal: "build a new authentication service from scratch following this specification",
+    expectedStrategy: "spec"
+  },
+  {
+    goal: "scaffold a greenfield mobile backend from the attached spec document",
+    expectedStrategy: "spec"
+  },
+  {
+    goal: "implement a brand-new billing engine from this written spec",
+    expectedStrategy: "spec"
+  },
+  {
+    goal: "create a new GraphQL gateway from scratch per the provided spec",
+    expectedStrategy: "spec"
+  },
+  {
+    goal: "build the greenfield analytics dashboard from this specification",
+    expectedStrategy: "spec"
+  },
+  // research — research-heavy investigation
+  { goal: "research the best vector database for our use case", expectedStrategy: "research" },
+  { goal: "investigate and compare OAuth libraries for Node", expectedStrategy: "research" },
+  {
+    goal: "do a deep research report on consensus algorithms for distributed systems",
+    expectedStrategy: "research"
+  },
+  { goal: "survey the landscape of LLM evaluation frameworks", expectedStrategy: "research" },
+  {
+    goal: "research which benchmarks are worth adopting for our agents",
+    expectedStrategy: "research"
+  },
+  {
+    goal: "research the current best practices for prompt caching across LLM providers",
+    expectedStrategy: "research"
+  },
+  {
+    goal: "compare and evaluate message-queue technologies for our throughput needs",
+    expectedStrategy: "research"
+  },
+  {
+    goal: "survey the state of the art in retrieval-augmented generation",
+    expectedStrategy: "research"
+  },
+  {
+    goal: "investigate which observability stack fits our microservices best",
+    expectedStrategy: "research"
+  },
+  {
+    goal: "research the tradeoffs between monorepo and polyrepo for our team",
+    expectedStrategy: "research"
+  }
+];
+
+// src/orchestration/meta-strategy-readiness.ts
+var DEFAULT_META_STRATEGY_READINESS_CONFIG = {
+  minTestCases: 20,
+  minDelta: 0.05,
+  minLearnedAccuracy: 0.7
+};
+function pctStr(n) {
+  return String(Math.round(n * 100));
+}
+function evaluateMetaStrategyReadiness(result, config = DEFAULT_META_STRATEGY_READINESS_CONFIG) {
+  const criteria = [
+    {
+      name: "volume",
+      met: result.testCount >= config.minTestCases,
+      detail: `${String(result.testCount)} held-out test cases (need \u2265 ${String(config.minTestCases)})`
+    },
+    {
+      name: "learned-beats-rules",
+      met: result.delta >= config.minDelta,
+      detail: `learned\u2212rules delta ${result.delta.toFixed(2)} (need \u2265 ${config.minDelta.toFixed(2)})`
+    },
+    {
+      name: "learned-accuracy-floor",
+      met: result.learnedAccuracy >= config.minLearnedAccuracy,
+      detail: `learned accuracy ${pctStr(result.learnedAccuracy)}% (need \u2265 ${pctStr(config.minLearnedAccuracy)}%)`
+    }
+  ];
+  const blockers = criteria.filter((c) => !c.met).map((c) => c.name);
+  return { ready: blockers.length === 0, criteria, blockers };
+}
+
 // src/orchestration/meta-dispatcher.ts
 var MetaDispatchError = class extends Error {
   code;
@@ -47194,12 +47514,34 @@ function toMetaInput(input, mode) {
     ...input.forceStrategy !== void 0 ? { forceStrategy: input.forceStrategy } : {}
   };
 }
+var readinessLogged = false;
+function logMetaStrategyReadinessOnce(logger58) {
+  if (readinessLogged) return;
+  readinessLogged = true;
+  try {
+    const evalResult = evaluateMetaStrategy(META_STRATEGY_CORPUS);
+    const verdict = evaluateMetaStrategyReadiness(evalResult);
+    logger58.info("meta-strategy learned-selector readiness", {
+      ready: verdict.ready,
+      delta: evalResult.delta,
+      learnedAccuracy: evalResult.learnedAccuracy,
+      rulesAccuracy: evalResult.rulesAccuracy,
+      testCount: evalResult.testCount,
+      blockers: verdict.blockers
+    });
+  } catch (err2) {
+    logger58.warn("meta-strategy readiness signal failed (non-fatal)", {
+      error: getErrorMessage(err2)
+    });
+  }
+}
 function selectDecision(input, mode, logger58) {
   const meta = createMetaOrchestrator({
     ...logger58 !== void 0 ? { logger: logger58 } : {},
     shadowSelector: getShadowSelector(),
     shadowSink: getShadowSink()
   });
+  logMetaStrategyReadinessOnce(logger58 ?? createLogger({ component: "RunTool" }));
   return meta.select(toMetaInput(input, mode));
 }
 function routeGoal(input, logger58) {
@@ -51446,4 +51788,4 @@ export {
   shutdownFeedbackSubscriber,
   createEventBusBridge
 };
-//# sourceMappingURL=chunk-3P5WC7AV.js.map
+//# sourceMappingURL=chunk-KXTOVEZS.js.map