@tangle-network/agent-runtime 0.102.0 → 0.102.1

package/dist/{chunk-4AMAFNUD.js → chunk-7FJJUC73.js}

@@ -16,12 +16,17 @@ import {
   verifiedResourceTextByDigest,
   verifyAgentCandidateBundle,
   verifyCanonicalCandidateDocument
-} from "./chunk-TGDHHHH4.js";
+} from "./chunk-LJLG734E.js";
+import {
+  defaultStructuralRolloutPolicy,
+  optimizerMethod
+} from "./chunk-ROILMKDL.js";
 import {
   assertModelAllowed,
   harnessInvocation,
-  runLocalHarness
-} from "./chunk-IKGCKPDX.js";
+  runLocalHarness,
+  runSettledCommand
+} from "./chunk-SQ3ZJH3Y.js";
 import {
   runAnalystLoop
 } from "./chunk-FD2MBMOH.js";
@@ -80,6 +85,7 @@ function agenticGenerator(opts = {}) {
       costLedger,
       costPhase
     }) {
+      signal.throwIfAborted();
       let reproducibleCostLedger;
       if (opts.codexReproducible) {
         if (!costLedger) {
@@ -97,7 +103,7 @@ function agenticGenerator(opts = {}) {
       const shots = Math.max(1, maxShots);
       let attemptNote = "";
       for (let shot = 0; shot < shots; shot++) {
-        if (signal.aborted) break;
+        signal.throwIfAborted();
         const taskPrompt = attemptNote ? `${basePrompt}
 
 ${attemptNote}` : basePrompt;
@@ -191,6 +197,7 @@ ${attemptNote}` : basePrompt;
           error: shotError
         });
         await emitShotReceipt(opts.onShotCompleted, receipt, execution, shotError);
+        signal.throwIfAborted();
         if (!execution) {
           throw new Error("agenticGenerator: author shot completed without a harness result");
         }
@@ -198,6 +205,7 @@ ${attemptNote}` : basePrompt;
         try {
           worktreeChanged = dirty(worktreePath);
         } catch (cause) {
+          signal.throwIfAborted();
           return rethrowShotSetupError(
             opts.onShotDisposition,
             receipt,
@@ -207,10 +215,12 @@ ${attemptNote}` : basePrompt;
           );
         }
         if (!worktreeChanged) {
+          signal.throwIfAborted();
           await emitShotDisposition(opts.onShotDisposition, receipt, {
             kind: "clean",
             worktreePath
           });
+          signal.throwIfAborted();
           attemptNote = EMPTY_TREE_NOTE;
           continue;
         }
@@ -219,6 +229,7 @@ ${attemptNote}` : basePrompt;
           try {
             problem = rawTraceEvidenceProblem(worktreePath, findings);
           } catch (cause) {
+            signal.throwIfAborted();
             return rethrowShotSetupError(
               opts.onShotDisposition,
               receipt,
@@ -228,28 +239,35 @@ ${attemptNote}` : basePrompt;
             );
           }
           if (problem) {
+            signal.throwIfAborted();
             await emitShotDisposition(opts.onShotDisposition, receipt, {
               kind: "rejected",
               worktreePath,
               stage: "raw-trace-evidence",
               feedback: problem
             });
+            signal.throwIfAborted();
             attemptNote = problem;
             continue;
           }
         }
         if (!verify) {
+          signal.throwIfAborted();
           await emitShotDisposition(opts.onShotDisposition, receipt, {
             kind: "accepted",
             worktreePath,
             verified: false
           });
+          signal.throwIfAborted();
           return acceptedCandidate(findings);
         }
         let result;
         try {
-          result = await verify(worktreePath);
+          signal.throwIfAborted();
+          result = await verify(worktreePath, signal);
+          signal.throwIfAborted();
         } catch (cause) {
+          signal.throwIfAborted();
           return rethrowShotSetupError(
             opts.onShotDisposition,
             receipt,
@@ -259,19 +277,23 @@ ${attemptNote}` : basePrompt;
           );
         }
         if (result.ok) {
+          signal.throwIfAborted();
           await emitShotDisposition(opts.onShotDisposition, receipt, {
             kind: "accepted",
             worktreePath,
             verified: true
           });
+          signal.throwIfAborted();
           return acceptedCandidate(findings);
         }
+        signal.throwIfAborted();
         await emitShotDisposition(opts.onShotDisposition, receipt, {
           kind: "rejected",
           worktreePath,
           stage: "verification",
           feedback: result.feedback ?? null
         });
+        signal.throwIfAborted();
         attemptNote = failureNote(result.feedback);
       }
       return { applied: false, summary: "" };
@@ -420,6 +442,7 @@ function shotExecutionSnapshot(result) {
     killedBySignal: result.killedBySignal,
     durationMs: result.durationMs,
     timedOut: result.timedOut,
+    ...result.aborted !== void 0 ? { aborted: result.aborted } : {},
     ...usage ? { usage } : {},
     ...evidence ? { evidence } : {}
   });
@@ -445,6 +468,7 @@ function shotReceipt(input) {
     durationMs: result?.durationMs ?? input.completedAt.getTime() - input.startedAt.getTime(),
     exitCode: result?.exitCode ?? null,
     timedOut: result?.timedOut ?? false,
+    aborted: result?.aborted ?? false,
     killedBySignal: result?.killedBySignal ?? null,
     stdoutBytes: result ? Buffer.byteLength(result.stdout) : null,
     stdoutSha256: result ? sha256(result.stdout) : null,
@@ -473,6 +497,9 @@ function shotFailure(result, exactPrompt, codexReproducible) {
   if (result.timedOut) {
     return new Error("agenticGenerator: author shot timed out");
   }
+  if (result.aborted) {
+    return new Error("agenticGenerator: author shot was cancelled by the caller");
+  }
   if (result.killedBySignal) {
     return new Error(`agenticGenerator: author shot was killed by ${result.killedBySignal}`);
   }
@@ -509,11 +536,20 @@ function sha256(value) {
 }
 function defaultBuildPrompt(args) {
   const lines = [
-    "You are improving this codebase based on an evaluation analysis.",
-    "Make the smallest set of edits that addresses the findings below, then stop.",
-    "Do not change unrelated code. Do not commit \u2014 leave changes in the working tree.",
+    "You are improving this codebase based on an evaluation analysis: real runs failed, an",
+    "analyst distilled the findings below, and your change will be measured on held-out tasks",
+    "against the unchanged baseline \u2014 only a real lift promotes it.",
+    "",
+    optimizerMethod,
     "",
-    "Findings:"
+    "THE SURFACE \u2014 what a deliverable change looks like here:",
+    "- edit this codebase in place: the smallest coherent change set that fully tests your",
+    "  hypothesis about the dominant failure mode (see the method above \u2014 no unrelated edits,",
+    "  they confound the measurement),",
+    "- keep the diff reviewable: a reviewer should be able to trace every hunk back to a finding,",
+    "- do not commit \u2014 leave changes in the working tree.",
+    "",
+    "FINDINGS \u2014 ranked evidence from real failed runs:"
   ];
   for (const f of args.findings) {
     const where = f.subject ? ` [${f.subject}]` : "";
@@ -588,39 +624,45 @@ function traceEvidencePaths(findings) {
   return [...new Set(out)];
 }
 function commandVerifier(command, args = [], timeoutMs = 3e5) {
-  return (worktreePath) => {
-    const result = spawnSync(command, args, {
-      cwd: worktreePath,
-      encoding: "utf-8",
-      timeout: timeoutMs
-    });
-    if (result.signal) {
-      return {
-        ok: false,
-        feedback: `verifier '${command}' killed by ${result.signal} (likely timeout after ${timeoutMs}ms)`
-      };
-    }
-    if (result.error) {
-      const code = result.error.code;
+  return async (worktreePath, signal) => {
+    let result;
+    try {
+      result = await runSettledCommand({
+        command,
+        args,
+        cwd: worktreePath,
+        timeoutMs,
+        ...signal ? { signal } : {}
+      });
+    } catch (err) {
+      signal?.throwIfAborted();
+      const code = err.code;
       if (code === "ENOENT") {
         throw new Error(
           `commandVerifier: '${command}' not found in PATH (setup bug, not a failed candidate)`
         );
       }
-      throw new Error(`commandVerifier: '${command}' failed to spawn: ${result.error.message}`);
+      const reason = err instanceof Error ? err.message : String(err);
+      throw new Error(`commandVerifier: '${command}' failed to spawn: ${reason}`);
+    }
+    if (result.timedOut || result.killedBySignal) {
+      return {
+        ok: false,
+        feedback: `verifier '${command}' ${result.killedBySignal ? `killed by ${result.killedBySignal}` : "timed out"} after ${timeoutMs}ms`
+      };
     }
-    if (result.status === 0) return { ok: true };
-    const out = `${result.stdout ?? ""}${result.stderr ?? ""}`.trim();
-    return { ok: false, feedback: out.length > 0 ? out : `exit ${result.status}` };
+    if (result.exitCode === 0) return { ok: true };
+    const out = `${result.stdout}${result.stderr}`.trim();
+    return { ok: false, feedback: out.length > 0 ? out : `exit ${result.exitCode}` };
   };
 }
-function summarize(findings) {
+function summarizeFindings(findings) {
   if (findings.length === 0) return "agentic improvement";
   if (findings.length === 1) return `agentic: ${truncate(findings[0].claim, 64)}`;
   return `agentic: ${findings.length} findings addressed`;
 }
 function acceptedCandidate(findings) {
-  const summary = summarize(findings);
+  const summary = summarizeFindings(findings);
   return {
     applied: true,
     summary,
@@ -663,6 +705,59 @@ function worktreeChangedPaths(worktreePath) {
   return result.stdout.split("\n").map((line) => line.trim()).filter((line) => line.length > 0).map((line) => line.slice(3).trim());
 }
 
+// src/improvement/findings.ts
+import { makeFinding } from "@tangle-network/agent-eval";
+var SEVERITIES = /* @__PURE__ */ new Set(["critical", "high", "medium", "low", "info"]);
+var LIFTED_FINDING_ANALYST_ID = "lifted-seed";
+function isAnalystFinding(value) {
+  if (!value || typeof value !== "object") return false;
+  const o = value;
+  return o.schema_version === "1.0.0" && typeof o.finding_id === "string" && typeof o.analyst_id === "string" && typeof o.severity === "string" && SEVERITIES.has(o.severity) && typeof o.area === "string" && typeof o.claim === "string" && typeof o.confidence === "number" && Array.isArray(o.evidence_refs);
+}
+function liftedClaim(value) {
+  if (typeof value === "string") return value.trim() || null;
+  if (value && typeof value === "object") {
+    const o = value;
+    for (const key of ["recommended_action", "claim", "lesson", "notes", "text", "message"]) {
+      const v = o[key];
+      if (typeof v === "string" && v.trim()) return v.trim();
+    }
+    try {
+      const json = JSON.stringify(value);
+      if (json && json !== "{}" && json !== "[]") {
+        return json.length > 400 ? `${json.slice(0, 399)}\u2026` : json;
+      }
+    } catch {
+    }
+  }
+  return null;
+}
+function toAnalystFindings(findings, opts = {}) {
+  const analystId = opts.analystId ?? LIFTED_FINDING_ANALYST_ID;
+  const area = opts.area ?? "seed";
+  const out = [];
+  for (const f of findings) {
+    if (isAnalystFinding(f)) {
+      out.push(f);
+      continue;
+    }
+    const claim = liftedClaim(f);
+    if (!claim) continue;
+    out.push(
+      makeFinding({
+        analyst_id: analystId,
+        severity: "info",
+        area,
+        confidence: 0.5,
+        claim,
+        evidence_refs: [],
+        ...f && typeof f === "object" ? { metadata: { raw: f } } : {}
+      })
+    );
+  }
+  return out;
+}
+
 // src/improvement/improvement-driver.ts
 import { spawnSync as spawnSync2 } from "child_process";
 import {
@@ -799,15 +894,18 @@ function resolveFindings(ctx) {
   const report = ctx.report;
   if (report && typeof report === "object" && "findings" in report) {
     const f = report.findings;
-    if (Array.isArray(f) && f.length > 0) return f;
+    if (Array.isArray(f) && f.length > 0) {
+      const lifted = toAnalystFindings(f, { analystId: "report-findings", area: "report" });
+      if (lifted.length > 0) return lifted;
+    }
   }
-  return ctx.findings;
+  return toAnalystFindings(ctx.findings ?? [], { analystId: "loop-context", area: "seed" });
 }
 
 // src/improvement/raw-trace-distiller.ts
 import { existsSync as existsSync2, readdirSync } from "fs";
 import { basename, join as join2, resolve as resolve2 } from "path";
-import { makeFinding } from "@tangle-network/agent-eval";
+import { makeFinding as makeFinding2 } from "@tangle-network/agent-eval";
 var ANALYST_ID = "raw-trace-distiller";
 var PASS_THRESHOLD = 0.999;
 function rawTraceDistiller(options = {}) {
@@ -829,7 +927,7 @@ function rawTraceDistiller(options = {}) {
         return options.fallbackFindings;
       }
       return [
-        makeFinding({
+        makeFinding2({
           analyst_id: ANALYST_ID,
           severity: "info",
           area: "raw-trace-context",
@@ -843,7 +941,7 @@ function rawTraceDistiller(options = {}) {
     }
     const findings = [];
     findings.push(
-      makeFinding({
+      makeFinding2({
         analyst_id: ANALYST_ID,
         severity: "high",
         area: "raw-trace-context",
@@ -871,7 +969,7 @@ function rawTraceDistiller(options = {}) {
 ${files}${more}` : header;
       }).join("\n");
       findings.push(
-        makeFinding({
+        makeFinding2({
           analyst_id: ANALYST_ID,
           severity: cand.composite < 0.5 ? "critical" : "high",
           area: "raw-trace-context",
@@ -976,8 +1074,123 @@ function truncate2(s, n) {
   return s.length <= n ? s : `${s.slice(0, n - 1)}\u2026`;
 }
 
+// src/improvement/rollout-policy.ts
+var ROLLOUT_POLICY_EXTENSION = "structural-rollout";
+var ROLLOUT_POLICY_BOUNDS = {
+  k: { min: 1, max: 10, step: 2 },
+  repairRounds: { min: 0, max: 3, step: 1 },
+  testgen: { min: 0, max: 10, step: 3 }
+};
+var MAX_CANDIDATES_PER_GENERATION = 4;
+var clamp = (v, min, max) => Math.min(max, Math.max(min, v));
+var isBoundedInt = (v, min) => typeof v === "number" && Number.isInteger(v) && v >= min;
+function parseRolloutPolicy(surface) {
+  if (typeof surface !== "string" || surface.trim().length === 0) return void 0;
+  let raw;
+  try {
+    raw = JSON.parse(surface);
+  } catch {
+    return void 0;
+  }
+  return normalizeRolloutPolicy(raw);
+}
+function normalizeRolloutPolicy(raw) {
+  if (typeof raw !== "object" || raw === null || Array.isArray(raw)) return void 0;
+  const bag = raw;
+  const k = bag.k ?? defaultStructuralRolloutPolicy.k;
+  const repairRounds = bag.repairRounds ?? defaultStructuralRolloutPolicy.repairRounds;
+  const testgen = bag.testgen ?? defaultStructuralRolloutPolicy.testgen;
+  if (!isBoundedInt(k, 1) || !isBoundedInt(repairRounds, 0) || !isBoundedInt(testgen, 0)) {
+    return void 0;
+  }
+  return {
+    k,
+    repairRounds,
+    testgen,
+    ...typeof bag.diverse === "boolean" ? { diverse: bag.diverse } : {},
+    ...typeof bag.temperature === "number" ? { temperature: bag.temperature } : {}
+  };
+}
+function serializeRolloutPolicy(policy) {
+  return JSON.stringify({
+    k: policy.k,
+    repairRounds: policy.repairRounds,
+    testgen: policy.testgen,
+    ...policy.diverse !== void 0 ? { diverse: policy.diverse } : {},
+    ...policy.temperature !== void 0 ? { temperature: policy.temperature } : {}
+  });
+}
+function structuralRolloutPolicyFromProfile(profile) {
+  const bag = profile.extensions?.[ROLLOUT_POLICY_EXTENSION];
+  if (bag === void 0) return void 0;
+  return normalizeRolloutPolicy(bag);
+}
+function applyRolloutPolicyToProfile(profile, policy) {
+  const bag = {
+    k: policy.k,
+    repairRounds: policy.repairRounds,
+    testgen: policy.testgen,
+    ...policy.diverse !== void 0 ? { diverse: policy.diverse } : {},
+    ...policy.temperature !== void 0 ? { temperature: policy.temperature } : {}
+  };
+  return {
+    ...profile,
+    extensions: { ...profile.extensions, [ROLLOUT_POLICY_EXTENSION]: bag }
+  };
+}
+function enumerateNeighborPolicies(policy) {
+  const moves = [
+    { dial: "k", delta: 1 },
+    { dial: "k", delta: -1 },
+    { dial: "repairRounds", delta: 1 },
+    { dial: "repairRounds", delta: -1 },
+    { dial: "testgen", delta: 1 },
+    { dial: "testgen", delta: -1 }
+  ];
+  const seen = /* @__PURE__ */ new Set([serializeRolloutPolicy(policy)]);
+  const neighbors = [];
+  for (const move of moves) {
+    const bounds = ROLLOUT_POLICY_BOUNDS[move.dial];
+    const next = clamp(policy[move.dial] + move.delta * bounds.step, bounds.min, bounds.max);
+    const candidate = { ...policy, [move.dial]: next };
+    const key = serializeRolloutPolicy(candidate);
+    if (seen.has(key)) continue;
+    seen.add(key);
+    neighbors.push(candidate);
+  }
+  return neighbors;
+}
+function candidateLabel(base, next) {
+  for (const dial of ["k", "repairRounds", "testgen"]) {
+    if (next[dial] !== base[dial]) return `${dial} ${base[dial]}\u2192${next[dial]}`;
+  }
+  return "unchanged";
+}
+function rolloutPolicyProposer() {
+  return {
+    kind: "rollout-policy",
+    async propose(ctx) {
+      const policy = parseRolloutPolicy(ctx.currentSurface);
+      if (!policy) return [];
+      const neighbors = enumerateNeighborPolicies(policy);
+      if (neighbors.length === 0) return [];
+      const cap = Math.max(1, Math.min(ctx.populationSize, MAX_CANDIDATES_PER_GENERATION));
+      const start = ctx.generation * cap % neighbors.length;
+      const window = [];
+      for (let i = 0; i < Math.min(cap, neighbors.length); i += 1) {
+        window.push(neighbors[(start + i) % neighbors.length]);
+      }
+      return window.map((candidate) => ({
+        surface: serializeRolloutPolicy(candidate),
+        label: candidateLabel(policy, candidate),
+        rationale: "bounded single-dial neighbor of the current structuralRollout policy; the held-out gate decides (deterministic enumeration \u2014 the dial space is tiny and prompt-style reflective proposals are a measured zero here)"
+      }));
+    }
+  };
+}
+
 // src/improvement/improve.ts
-import { canonicalJson } from "@tangle-network/agent-eval";
+import { canonicalJson, makeFinding as makeFinding3 } from "@tangle-network/agent-eval";
 import {
   gepaProposer,
   gitWorktreeAdapter,
@@ -1003,6 +1216,8 @@ function defaultGeneratorFor(surface, llm) {
       return skillOptProposer({ llm: llmClientOptions(llm), model, target: "agent skill document" });
     case "memory":
       return memoryCurationProposer();
+    case "rollout-policy":
+      return rolloutPolicyProposer();
     default:
       return void 0;
   }
@@ -1025,6 +1240,10 @@ function baselineSurfaceFor(profile, surface, skills) {
       return canonicalJson(profile);
     case "memory":
       return profileInstructions(profile);
+    case "rollout-policy": {
+      const policy = structuralRolloutPolicyFromProfile(profile);
+      return policy ? serializeRolloutPolicy(policy) : "";
+    }
     case "code":
       throw new ConfigError(
         "improve(): code requires the isolated baseline created from opts.code.repoRoot"
@@ -1060,7 +1279,23 @@ function generationFailureDistiller(staticFindings) {
     }
     if (failures.length === 0) return staticFindings;
     failures.sort((a, b) => a.composite - b.composite);
-    return failures.slice(0, CAP);
+    return failures.slice(0, CAP).map(
+      (f) => makeFinding3({
+        analyst_id: "generation-failure-distiller",
+        severity: f.error !== void 0 || f.composite < 0.5 ? "high" : "medium",
+        area: "generation-failure",
+        confidence: 1,
+        subject: f.scenario,
+        claim: `Scenario ${f.scenario} scored composite ${f.composite}${f.notes ? `: ${f.notes}` : ""}${f.error ? ` (error: ${f.error})` : ""}`,
+        evidence_refs: [],
+        metadata: {
+          scenario: f.scenario,
+          composite: f.composite,
+          ...f.notes ? { notes: f.notes } : {},
+          ...f.error !== void 0 ? { error: f.error } : {}
+        }
+      })
+    );
   };
 }
 function memoryGenerationDistiller(staticFindings) {
@@ -1070,6 +1305,14 @@ function memoryGenerationDistiller(staticFindings) {
     return fresh === staticFindings ? staticFindings : [...staticFindings, ...fresh];
   };
 }
+function defaultDistillerFor(opts, findings) {
+  const surface = opts.surface ?? "prompt";
+  const durableRun = opts.runDir !== void 0 && !opts.runDir.startsWith("mem://");
+  const useRawTraces = opts.rawTraceContext ?? (surface === "memory" ? false : durableRun);
+  if (useRawTraces) return rawTraceDistiller({ fallbackFindings: findings });
+  if (surface === "memory") return memoryGenerationDistiller(findings);
+  return generationFailureDistiller(findings);
+}
 async function discardPreparedBaseline(worktree, baselineWorktree, cause) {
   return rethrowAfterCleanup(
     cause,
@@ -1201,6 +1444,17 @@ function materializeImprovementProfileCandidate(profile, surface, winner, skills
         }
       };
       break;
+    case "rollout-policy": {
+      if (winner === "") return void 0;
+      const policy = normalizeRolloutPolicy(parseWinnerJson(winner, surface));
+      if (!policy) {
+        throw new ConfigError(
+          `improve(): the shipped 'rollout-policy' winner is not a valid StructuralRolloutPolicy (integer k >= 1, repairRounds >= 0, testgen >= 0), so it cannot be applied: ${winner}`
+        );
+      }
+      candidate = applyRolloutPolicyToProfile(profile, policy);
+      break;
+    }
     case "code":
       return void 0;
   }
@@ -1311,7 +1565,7 @@ async function improve(profile, findings, opts) {
       findings,
       ...promotionGate !== void 0 ? { gate: promotionGate } : {},
       ...analyzeGeneration === null ? {} : {
-        analyzeGeneration: analyzeGeneration ?? (rawTraceContext ? rawTraceDistiller({ fallbackFindings: findings }) : surface === "memory" ? memoryGenerationDistiller(findings) : generationFailureDistiller(findings))
+        analyzeGeneration: analyzeGeneration ?? defaultDistillerFor(opts, findings)
       }
     });
   } catch (cause) {
@@ -1357,7 +1611,7 @@ async function improve(profile, findings, opts) {
   return {
     candidate,
     decision: raw.gateDecision,
-    lift: raw.lift,
+    ...raw.lift !== void 0 ? { lift: raw.lift } : {},
     raw,
     dispose
   };
@@ -2039,9 +2293,26 @@ function assertNoJudgeDerivedProposalFindings(findings) {
 export {
   AGENTIC_PROFILE_RESOURCE_ROOT,
   agenticGenerator,
+  defaultBuildPrompt,
+  rawTraceEvidenceProblem,
+  requiresRawTraceEvidence,
   commandVerifier,
+  summarizeFindings,
+  worktreeChangedPaths,
+  LIFTED_FINDING_ANALYST_ID,
+  isAnalystFinding,
+  toAnalystFindings,
   improvementDriver,
   rawTraceDistiller,
+  ROLLOUT_POLICY_EXTENSION,
+  ROLLOUT_POLICY_BOUNDS,
+  parseRolloutPolicy,
+  normalizeRolloutPolicy,
+  serializeRolloutPolicy,
+  structuralRolloutPolicyFromProfile,
+  applyRolloutPolicyToProfile,
+  enumerateNeighborPolicies,
+  rolloutPolicyProposer,
   improve,
   AGENT_IMPROVEMENT_PROFILE_SURFACES,
   buildAgentImprovementActivationTargets,
@@ -2064,4 +2335,4 @@ export {
   verifyAgentImprovementActivation,
   verifyCandidateExecutionEvidence
 };
-//# sourceMappingURL=chunk-4AMAFNUD.js.map
+//# sourceMappingURL=chunk-7FJJUC73.js.map