@tangle-network/agent-runtime 0.94.13 → 0.96.1

package/dist/index.js

@@ -1,27 +1,14 @@
-import {
-  RESEARCH_SUPERVISOR_SYSTEM_PROMPT,
-  createAgentKnowledgeReadinessCheck,
-  createSupervisedKnowledgeUpdater,
-  formatSupervisedKnowledgeTask,
-  knowledgeReadinessDeliverable,
-  runKnowledgeImprovementJob,
-  runSupervisedKnowledgeUpdate
-} from "./chunk-HGRW27YY.js";
-import "./chunk-5QOB7H74.js";
 import {
   FileAgentCandidateExecutionClaimStore,
   buildAgentCandidateBundle,
-  captureAgentCandidateWorkspace,
-  captureAgentCandidateWorkspaceFiles,
-  createAgentCandidateWorkspacePort,
   createProtectedAgentCandidateModelPort,
   disposePreparedAgentCandidateExecution,
   recoverExpiredAgentCandidateExecution
-} from "./chunk-TVJQAYQM.js";
+} from "./chunk-6XKPVJAZ.js";
 import {
   mcpToolsForRuntimeMcp,
   mcpToolsForRuntimeMcpSubset
-} from "./chunk-33OG2NN3.js";
+} from "./chunk-WYC2XJF2.js";
 import {
   DEFAULT_ROUTER_BASE_URL,
   cleanModelId,
@@ -40,7 +27,7 @@ import {
   runLoopRunnerCli,
   selfImproveLoopRunner,
   worktreeLoopRunner
-} from "./chunk-XP5KDM3R.js";
+} from "./chunk-SGQ4YIQW.js";
 import "./chunk-SGKPNBXE.js";
 import {
   CircuitBreakerState,
@@ -67,46 +54,69 @@ import {
   sleep,
   slugifySpeaker,
   turnId
-} from "./chunk-VYA2YEKA.js";
+} from "./chunk-2KGAN2HM.js";
+import "./chunk-FXF2OL34.js";
+import {
+  InMemoryRuntimeSessionStore,
+  createIterableBackend,
+  createOpenAICompatibleBackend,
+  createSandboxPromptBackend,
+  normalizeBackendStreamEvent,
+  nowIso,
+  startOrResumeRuntimeSession,
+  touchSession
+} from "./chunk-Q2JSAVQ3.js";
+import {
+  RESEARCH_SUPERVISOR_SYSTEM_PROMPT,
+  buildKnowledgeImprovementExperimentBundles,
+  createAgentKnowledgeReadinessCheck,
+  createKnowledgeImprovementActivationExecutor,
+  createSupervisedKnowledgeUpdater,
+  formatSupervisedKnowledgeTask,
+  knowledgeReadinessDeliverable,
+  runKnowledgeImprovementJob,
+  runSupervisedKnowledgeUpdate
+} from "./chunk-3XKSBI2U.js";
+import "./chunk-5QOB7H74.js";
+import {
+  captureAgentCandidateWorkspace,
+  captureAgentCandidateWorkspaceFiles,
+  createAgentCandidateWorkspacePort,
+  sealAgentCandidateBundle
+} from "./chunk-BLQIYRVR.js";
 import {
-  applyImprovementWinnerToProfile,
+  AGENTIC_PROFILE_RESOURCE_ROOT,
+  agenticGenerator,
+  commandVerifier,
   improve,
   improvementDriver,
   rawTraceDistiller
-} from "./chunk-PCURO3DL.js";
+} from "./chunk-PSOCBNM3.js";
 import {
+  AGENT_CANDIDATE_EXECUTION_SUPPORT,
+  CANDIDATE_KNOWLEDGE_RETRIEVAL_CONFIG_ENV,
+  CANDIDATE_KNOWLEDGE_ROOT_ENV,
   CANDIDATE_TRACE_ENV,
   CANDIDATE_TRACE_TAGS,
   InMemoryAgentCandidateExecutionClaimStore,
   applyExactAgentProfileDiff,
   candidateExecutionClaim,
+  candidateKnowledgeExecutionPaths,
   canonicalCandidateDigest,
   executePreparedAgentCandidate,
   parseExactAgentProfile,
   parseExactAgentProfileDiff,
   persistCandidateOutputArtifact,
   prepareAgentCandidateExecution,
-  sealAgentCandidateBundle,
   verifyAgentCandidateBundle
-} from "./chunk-CNH7DF7Z.js";
-import "./chunk-U33YZ7B2.js";
+} from "./chunk-YOLKCWRV.js";
 import {
   composeRuntimeHooks,
   defineRuntimeHooks,
   notifyRuntimeDecisionPoint,
   notifyRuntimeHookEvent
-} from "./chunk-C3UKLQ54.js";
+} from "./chunk-UQ6PNNXM.js";
 import "./chunk-DPEUKJRO.js";
-import {
-  InMemoryRuntimeSessionStore,
-  createIterableBackend,
-  createOpenAICompatibleBackend,
-  createSandboxPromptBackend,
-  newRuntimeSession,
-  normalizeBackendStreamEvent,
-  nowIso,
-  touchSession
-} from "./chunk-AEG3NGJ2.js";
 import {
   INTELLIGENCE_WIRE_VERSION,
   buildLoopOtelSpans,
@@ -120,23 +130,12 @@ import {
   sanitizeAgentRuntimeEvent,
   sanitizeKnowledgeReadinessReport,
   sanitizeRuntimeStreamEvent
-} from "./chunk-ISTDY47H.js";
-import "./chunk-WIPGQ4GT.js";
-import "./chunk-VSWBYWFK.js";
+} from "./chunk-HZDEXTSL.js";
+import "./chunk-IMSNJSXH.js";
+import "./chunk-M6MD6JBS.js";
 import "./chunk-3MDZX7YU.js";
-import "./chunk-QDSOD7RC.js";
+import "./chunk-FD2MBMOH.js";
 import "./chunk-M22PL4GH.js";
-import {
-  mcpBuildPrompt,
-  mcpServeVerifier,
-  toolBuildPrompt
-} from "./chunk-6O73TRHW.js";
-import {
-  AGENTIC_PROFILE_RESOURCE_ROOT,
-  agenticGenerator,
-  commandVerifier
-} from "./chunk-D3H7F6L2.js";
-import "./chunk-EP6RVHMX.js";
 import {
   AgentEvalError,
   BackendTransportError,
@@ -145,10 +144,8 @@ import {
   NotFoundError,
   PlannerError,
   RuntimeRunStateError,
-  SessionMismatchError,
   ValidationError
 } from "./chunk-YEJR7IXO.js";
-import "./chunk-DGUM43GV.js";
 
 // src/durable/chat-engine.ts
 var encoder = new TextEncoder();
@@ -236,6 +233,142 @@ function deriveExecutionId(input) {
   return `${input.projectId}:${input.sessionId}:${input.turnIndex}`;
 }
 
+// src/improvement/build-prompts.ts
+function findingLines(findings) {
+  return findings.map((f) => {
+    const where = f.subject ? ` [${f.subject}]` : "";
+    const action = f.recommended_action ? ` \u2192 ${f.recommended_action}` : "";
+    return `- (${f.severity})${where} ${f.claim}${action}`;
+  });
+}
+function toolBuildPrompt(args) {
+  return [
+    "You are building a new TOOL for this codebase to address the gaps below.",
+    "Write the tool as a small, self-contained module PLUS tests that exercise it.",
+    "The tool must compile and its tests must pass \u2014 they will be run automatically;",
+    "if verification fails you will get the error and another attempt. Do not commit;",
+    "leave the changes in the working tree.",
+    "",
+    "Gaps the tool should close:",
+    ...findingLines(args.findings)
+  ].join("\n");
+}
+function mcpBuildPrompt(args) {
+  return [
+    "You are building a new MCP SERVER (Model Context Protocol) that exposes",
+    "tool(s) addressing the gaps below, so any harness can mount it.",
+    "Requirements that WILL be checked by booting the server:",
+    "- it starts over stdio and answers the MCP `initialize` handshake,",
+    "- `tools/list` returns at least one tool with a valid input schema.",
+    "Newline-delimited JSON-RPC 2.0, protocol version 2024-11-05. Include a start",
+    "command (e.g. a package.json `start` script or a clear entrypoint). If the",
+    "boot-and-probe fails you will get the error and another attempt. Do not",
+    "commit; leave the changes in the working tree.",
+    "",
+    "Capabilities the server should provide:",
+    ...findingLines(args.findings)
+  ].join("\n");
+}
+
+// src/improvement/mcp-serve-verifier.ts
+import { spawn } from "child_process";
+import { createInterface } from "readline";
+var PROTOCOL_VERSION = "2024-11-05";
+function mcpServeVerifier(spec) {
+  const timeoutMs = spec.timeoutMs ?? 3e4;
+  const minTools = spec.minTools ?? 1;
+  return (worktreePath) => new Promise((resolve, reject) => {
+    const child = spawn(spec.command, spec.args ?? [], {
+      cwd: worktreePath,
+      stdio: ["pipe", "pipe", "pipe"],
+      env: { ...process.env, ...spec.env }
+    });
+    const stderr = [];
+    let settled = false;
+    let nextId = 1;
+    const initId = nextId++;
+    let listId = -1;
+    const settle = (fn) => {
+      if (settled) return;
+      settled = true;
+      clearTimeout(timer);
+      rl.close();
+      child.kill("SIGKILL");
+      fn();
+    };
+    const withStderr = (msg) => stderr.length > 0 ? `${msg}
+stderr:
+${stderr.join("").slice(-2e3)}` : msg;
+    const pass = () => settle(() => resolve({ ok: true }));
+    const failCandidate = (msg) => settle(() => resolve({ ok: false, feedback: withStderr(msg) }));
+    const setupFault = (err) => settle(() => reject(err));
+    const failStdin = (err) => failCandidate(`writing to MCP server stdin failed: ${err.message}`);
+    const send = (msg) => {
+      try {
+        child.stdin.write(`${JSON.stringify(msg)}
+`);
+        return true;
+      } catch (err) {
+        failStdin(err);
+        return false;
+      }
+    };
+    child.on("error", (err) => {
+      const code = err.code;
+      setupFault(
+        code === "ENOENT" ? new Error(
+          `mcpServeVerifier: '${spec.command}' not found in PATH (setup bug, not a failed candidate)`
+        ) : new Error(`mcpServeVerifier: '${spec.command}' failed to spawn: ${err.message}`)
+      );
+    });
+    child.on("exit", (code, signal) => {
+      failCandidate(`MCP server exited (code ${code}, signal ${signal}) before serving`);
+    });
+    child.stdin.on("error", failStdin);
+    child.stderr.on("data", (d) => stderr.push(String(d)));
+    const rl = createInterface({ input: child.stdout });
+    rl.on("line", (line) => {
+      let msg;
+      try {
+        msg = JSON.parse(line);
+      } catch {
+        return;
+      }
+      if (!msg || typeof msg !== "object") return;
+      if (msg.id === initId) {
+        if (msg.error) return failCandidate(`initialize errored: ${JSON.stringify(msg.error)}`);
+        if (!send({ jsonrpc: "2.0", method: "notifications/initialized" })) return;
+        listId = nextId++;
+        send({ jsonrpc: "2.0", id: listId, method: "tools/list" });
+        return;
+      }
+      if (msg.id === listId) {
+        if (msg.error) return failCandidate(`tools/list errored: ${JSON.stringify(msg.error)}`);
+        const tools = msg.result?.tools;
+        if (!Array.isArray(tools)) return failCandidate("tools/list result has no tools array");
+        if (tools.length < minTools) {
+          return failCandidate(`tools/list returned ${tools.length} tool(s), need >= ${minTools}`);
+        }
+        return pass();
+      }
+    });
+    const timer = setTimeout(
+      () => failCandidate(`MCP server did not complete the handshake within ${timeoutMs}ms`),
+      timeoutMs
+    );
+    send({
+      jsonrpc: "2.0",
+      id: initId,
+      method: "initialize",
+      params: {
+        protocolVersion: PROTOCOL_VERSION,
+        capabilities: {},
+        clientInfo: { name: "agent-runtime-mcp-verify", version: "0" }
+      }
+    });
+  });
+}
+
 // src/improvement/profile-diff-proposer.ts
 import { canonicalJson } from "@tangle-network/agent-eval";
 import {
@@ -256,13 +389,13 @@ function profileDiffProposer(options) {
       const candidates = [];
       const normalizedBaseline = applyExactAgentProfileDiff(
         profile,
-        { schemaVersion: 1, kind: "agent-profile-diff" },
+        { kind: "agent-profile-diff" },
         "profile diff baseline"
       );
       const seen = /* @__PURE__ */ new Set([canonicalJson(normalizedBaseline)]);
       for (const [index, proposal] of proposals.entries()) {
         if (candidates.length >= ctx.populationSize || ctx.signal.aborted) break;
-        const diff = parseExactAgentProfileDiff(proposal.diff, `profile diff proposal ${index}`);
+        const diff = parseExactAgentProfileDiff(proposal, `profile diff proposal ${index}`);
         const axes = changedAgentProfileAxes(diff);
         if (axes.length === 0) continue;
         const candidate = applyExactAgentProfileDiff(
@@ -275,16 +408,16 @@ function profileDiffProposer(options) {
         seen.add(surface);
         candidates.push({
           surface,
-          label: proposal.label ?? diff.title ?? diff.id ?? `Change ${axes.join(", ")}`,
-          rationale: candidateRationale(proposal, diff, axes)
+          label: diff.title ?? diff.id ?? `Change ${axes.join(", ")}`,
+          rationale: candidateRationale(diff, axes)
         });
       }
       return candidates;
     }
   };
 }
-function candidateRationale(proposal, diff, axes) {
-  const reason = proposal.rationale ?? diff.rationale ?? `Changes profile axes: ${axes.join(", ")}.`;
+function candidateRationale(diff, axes) {
+  const reason = diff.rationale ?? `Changes profile axes: ${axes.join(", ")}.`;
   const source = diff.source ? [diff.source.kind, ...diff.source.artifacts ?? []].join(" \xB7 ") : "source unspecified";
   return `${reason} Source: ${source}. Diff: ${canonicalCandidateDigest(diff)}.`;
 }
@@ -306,7 +439,7 @@ function reflectiveGenerator(opts) {
   return {
     kind: "reflective",
     async generate({ worktreePath, findings }) {
-      const batch = await opts.improvementAdapter.proposeFromFindings(findings);
+      const batch = await opts.improvementProposalSource.proposeFromFindings(findings);
       if (batch.edits.length === 0) return { applied: false, summary: "" };
       let applied = 0;
       for (const edit of batch.edits) {
@@ -557,19 +690,16 @@ async function* runAgentTaskStream(options) {
     return;
   }
   const store = options.sessionStore;
-  const existing = options.sessionId ? await store?.get(options.sessionId) : void 0;
-  const shouldResume = Boolean(options.resume && existing);
-  let session = shouldResume && existing ? await resumeBackendSession(options.backend, existing, input, {
-    task,
-    knowledge,
-    signal: options.signal
-  }) : await startBackendSession(
-    options.backend,
+  const opened = await startOrResumeRuntimeSession({
+    backend: options.backend,
     input,
-    { task, knowledge, signal: options.signal },
-    options.sessionId
-  );
-  await store?.put(session);
+    context: { task, knowledge, signal: options.signal },
+    store,
+    sessionId: options.sessionId,
+    resume: options.resume
+  });
+  let session = opened.session;
+  const shouldResume = opened.resumed;
   const sessionEvent = streamEvent({
     type: shouldResume ? "session_resumed" : "session_created",
     task,
@@ -587,7 +717,7 @@ async function* runAgentTaskStream(options) {
   yield backendStart;
   let finalText = "";
   try {
-    for await (const rawEvent of options.backend.stream(input, {
+    for await (const rawEvent of options.backend.stream(opened.input, {
       task,
       knowledge,
       session,
@@ -709,17 +839,6 @@ async function runKnowledgePreflightStream(task, questions, acquisitionPlans, pr
 function streamEvent(event) {
   return { ...event, timestamp: nowIso() };
 }
-async function startBackendSession(backend, input, context, requestedSessionId) {
-  if (backend.start) return backend.start(input, { ...context, requestedSessionId });
-  return newRuntimeSession(backend.kind, requestedSessionId);
-}
-async function resumeBackendSession(backend, session, input, context) {
-  if (session.backend !== backend.kind) {
-    throw new SessionMismatchError(session.backend, backend.kind);
-  }
-  if (backend.resume) return backend.resume(session, input, context);
-  return touchSession({ ...session, status: "active" });
-}
 function buildReadiness(task, provider) {
   if (provider?.buildReadiness) return provider.buildReadiness(task);
   return scoreKnowledgeReadiness({
@@ -1394,8 +1513,11 @@ function randomSuffix2(len = 8) {
 }
 export {
   AGENTIC_PROFILE_RESOURCE_ROOT,
+  AGENT_CANDIDATE_EXECUTION_SUPPORT,
   AgentEvalError,
   BackendTransportError,
+  CANDIDATE_KNOWLEDGE_RETRIEVAL_CONFIG_ENV,
+  CANDIDATE_KNOWLEDGE_ROOT_ENV,
   CANDIDATE_TRACE_ENV,
   CANDIDATE_TRACE_TAGS,
   CircuitBreakerState,
@@ -1421,15 +1543,16 @@ export {
   ValidationError,
   agenticGenerator,
   applyExactAgentProfileDiff,
-  applyImprovementWinnerToProfile,
   applyRunRecordDefaults,
   auditLoopRunner,
   buildAgentCandidateBundle,
   buildForwardHeaders,
+  buildKnowledgeImprovementExperimentBundles,
   buildLoopOtelSpans,
   buildLoopSpanNodes,
   buildRuntimeEventOtelSpans,
   candidateExecutionClaim,
+  candidateKnowledgeExecutionPaths,
   captureAgentCandidateWorkspace,
   captureAgentCandidateWorkspaceFiles,
   cleanModelId,
@@ -1440,6 +1563,7 @@ export {
   createAgentKnowledgeReadinessCheck,
   createConversationBackend,
   createIterableBackend,
+  createKnowledgeImprovementActivationExecutor,
   createOpenAICompatibleBackend,
   createOtelExporter,
   createProtectedAgentCandidateModelPort,