@tangle-network/agent-runtime 0.94.0 → 0.94.2

package/dist/{chunk-4E2GM63H.js → chunk-MZH4HORF.js}

@@ -36,7 +36,7 @@ import {
   terminalRecord,
   usdToNanos,
   withinCandidateCleanupDeadline
-} from "./chunk-7I2AOKCI.js";
+} from "./chunk-KHLE3C5G.js";
 
 // src/candidate-execution/builder.ts
 import { verifyCodeSurface } from "@tangle-network/agent-eval/campaign";
@@ -938,4 +938,4 @@ export {
   createProtectedAgentCandidateModelPort,
   recoverExpiredAgentCandidateExecution
 };
-//# sourceMappingURL=chunk-4E2GM63H.js.map
+//# sourceMappingURL=chunk-MZH4HORF.js.map

package/dist/{chunk-RTVVH2KL.js → chunk-OAVVQAOQ.js}

@@ -1,6 +1,6 @@
 import {
   runLocalHarness
-} from "./chunk-PX6SXX3M.js";
+} from "./chunk-JFAGXZKR.js";
 
 // src/improvement/agentic-generator.ts
 import { spawnSync } from "child_process";
@@ -207,4 +207,4 @@ export {
   agenticGenerator,
   commandVerifier
 };
-//# sourceMappingURL=chunk-RTVVH2KL.js.map
+//# sourceMappingURL=chunk-OAVVQAOQ.js.map

package/dist/{chunk-MRYPFDHJ.js → chunk-PV6Q3OKU.js}

@@ -5,13 +5,13 @@ import {
   definePersona,
   runPersonified,
   worktreeFanout
-} from "./chunk-OO4EK3JB.js";
+} from "./chunk-L4CACVIJ.js";
 import {
   runAnalystLoop
 } from "./chunk-ZQZX77MM.js";
 import {
   createExecutorRegistry
-} from "./chunk-FHZHADIM.js";
+} from "./chunk-TTWMSNFZ.js";
 import {
   ConfigError
 } from "./chunk-YEJR7IXO.js";
@@ -196,4 +196,4 @@ export {
   runLoopRunnerCli,
   parseLoopRunnerArgv
 };
-//# sourceMappingURL=chunk-MRYPFDHJ.js.map
+//# sourceMappingURL=chunk-PV6Q3OKU.js.map

package/dist/{chunk-FHZHADIM.js → chunk-TTWMSNFZ.js}

@@ -9,7 +9,7 @@ import {
   InMemoryResultBlobStore,
   InMemorySpawnJournal,
   contentAddress
-} from "./chunk-3D2RHC4K.js";
+} from "./chunk-WIPGQ4GT.js";
 import {
   providerAsExecutor,
   resolveAgentEnvironmentProvider
@@ -33,7 +33,7 @@ import {
 import {
   harnessInvocation,
   runLocalHarness
-} from "./chunk-PX6SXX3M.js";
+} from "./chunk-JFAGXZKR.js";
 import {
   AgentEvalError,
   ConfigError,
@@ -620,6 +620,7 @@ function clampSpend(spend, budget) {
       output: Math.floor(spend.tokens.output * ratio)
     } : spend.tokens,
     usd: budget.maxUsd === void 0 ? spend.usd : Math.min(spend.usd, budget.maxUsd),
+    ...spend.usdKnown === false ? { usdKnown: false } : {},
     ms: spend.ms
   };
 }
@@ -1980,13 +1981,16 @@ async function runWorktreeHarness(opts) {
     const { command, args } = harnessInvocation(opts.harness, opts.profile, opts.taskPrompt, {
       // This helper created the candidate worktree above; autonomous Claude
       // edits are permitted only inside that isolated checkout.
-      dangerouslySkipPermissions: opts.harness === "claude"
+      dangerouslySkipPermissions: opts.harness === "claude",
+      ...opts.codexReproducible ? { codexReproducible: true } : {}
     });
     const harnessResult = await runHarness({
       harness: opts.harness,
       cwd: worktree.path,
       taskPrompt: opts.taskPrompt,
       invocation: { command, args },
+      ...opts.codexReproducible ? { codexReproducible: true } : {},
+      ...opts.codexReadDeniedPaths ? { codexReadDeniedPaths: opts.codexReadDeniedPaths } : {},
       ...opts.harnessTimeoutMs !== void 0 ? { timeoutMs: opts.harnessTimeoutMs } : {},
       ...opts.signal ? { signal: opts.signal } : {}
     });
@@ -2014,7 +2018,19 @@ async function runWorktreeHarness(opts) {
         killedBySignal: harnessResult.killedBySignal,
         durationMs: harnessResult.durationMs,
         stdout: harnessResult.stdout,
-        stderr: harnessResult.stderr
+        stderr: harnessResult.stderr,
+        ...harnessResult.usage ? { usage: harnessResult.usage } : {},
+        ...harnessResult.evidence ? {
+          cliVersion: harnessResult.evidence.cliVersion,
+          executableSha256: harnessResult.evidence.executableSha256,
+          effectivePromptSha256: harnessResult.evidence.effectivePromptSha256,
+          nonPromptArgsSha256: harnessResult.evidence.nonPromptArgsSha256,
+          controlledConfigSha256: harnessResult.evidence.controlledConfigSha256,
+          readDeniedPaths: [...harnessResult.evidence.readDeniedPaths],
+          readDeniedPathsSha256: harnessResult.evidence.readDeniedPathsSha256,
+          readDeniedPathCount: harnessResult.evidence.readDeniedPathCount,
+          executionPolicy: harnessResult.evidence.policy
+        } : {}
       },
       ...checks ? { checks } : {}
     };
@@ -2082,9 +2098,22 @@ function createWorktreeCliExecutor(options) {
   if (typeof options.taskPrompt !== "string" || options.taskPrompt.length === 0) {
     throw new ValidationError("createWorktreeCliExecutor: taskPrompt required");
   }
+  if (options.codexReproducible && options.harness !== "codex") {
+    throw new ValidationError(
+      'createWorktreeCliExecutor: codexReproducible requires harness "codex"'
+    );
+  }
+  if (options.codexReproducible && options.budgetExempt === true) {
+    throw new ValidationError("createWorktreeCliExecutor: codexReproducible cannot be budgetExempt");
+  }
+  if (options.codexReadDeniedPaths !== void 0 && !options.codexReproducible) {
+    throw new ValidationError(
+      "createWorktreeCliExecutor: codexReadDeniedPaths requires codexReproducible"
+    );
+  }
   const runId = options.runId ?? randomUUID();
   const controller = new AbortController();
-  const budgetExempt = options.budgetExempt ?? true;
+  const budgetExempt = options.budgetExempt ?? !options.codexReproducible;
   let run;
   let artifact;
   return {
@@ -2103,6 +2132,8 @@ function createWorktreeCliExecutor(options) {
         ...options.testCmd !== void 0 ? { testCmd: options.testCmd } : {},
         ...options.typecheckCmd !== void 0 ? { typecheckCmd: options.typecheckCmd } : {},
         ...options.harnessTimeoutMs !== void 0 ? { harnessTimeoutMs: options.harnessTimeoutMs } : {},
+        ...options.codexReproducible ? { codexReproducible: true } : {},
+        ...options.codexReadDeniedPaths ? { codexReadDeniedPaths: options.codexReadDeniedPaths } : {},
         ...options.checkTimeoutMs !== void 0 ? { checkTimeoutMs: options.checkTimeoutMs } : {},
         ...options.checkOutputCap !== void 0 ? { checkOutputCap: options.checkOutputCap } : {},
         ...linked ? { signal: linked } : {},
@@ -2110,13 +2141,20 @@ function createWorktreeCliExecutor(options) {
         ...options.runHarness ? { runHarness: options.runHarness } : {},
         ...options.runCommand ? { runCommand: options.runCommand } : {}
       });
+      const usage = run.result.harness.usage;
+      if (!budgetExempt && !usage) {
+        const completed = run;
+        run = void 0;
+        await completed.cleanup();
+        throw new ValidationError(
+          "createWorktreeCliExecutor: metered harness run returned no token usage"
+        );
+      }
       const spent = {
         iterations: 1,
-        // The worktree-harness core surfaces no token/usd usage, so tokens/usd are a genuine zero
-        // (NOT a fabricated cost). When budgetExempt is true the pool ignores this spend entirely;
-        // when explicitly false the scope debits exactly this captured spend — the real iteration.
-        tokens: zeroTokenUsage(),
+        tokens: usage ? { input: usage.inputTokens, output: usage.outputTokens } : { input: 0, output: 0 },
         usd: 0,
+        ...usage ? { usdKnown: false } : {},
         ms: Date.now() - started
       };
       artifact = { outRef: contentAddress(run.result), out: run.result, spent };
@@ -3090,6 +3128,8 @@ var cliWorktreeExecutor = (spec, ctx) => {
     ...seam.runId ? { runId: seam.runId } : {},
     ...seam.baseRef ? { baseRef: seam.baseRef } : {},
     ...seam.harnessTimeoutMs !== void 0 ? { harnessTimeoutMs: seam.harnessTimeoutMs } : {},
+    ...seam.codexReproducible ? { codexReproducible: true } : {},
+    ...seam.codexReadDeniedPaths ? { codexReadDeniedPaths: seam.codexReadDeniedPaths } : {},
     ...seam.testCmd !== void 0 ? { testCmd: seam.testCmd } : {},
     ...seam.typecheckCmd !== void 0 ? { typecheckCmd: seam.typecheckCmd } : {},
     ...seam.checkTimeoutMs !== void 0 ? { checkTimeoutMs: seam.checkTimeoutMs } : {},
@@ -3332,6 +3372,7 @@ function sumSpend(settled) {
     total.tokens.input += ev.spent.tokens.input;
     total.tokens.output += ev.spent.tokens.output;
     total.usd += ev.spent.usd;
+    if (ev.spent.usdKnown === false) total.usdKnown = false;
     total.ms += ev.spent.ms;
   }
   return total;
@@ -3344,6 +3385,7 @@ function sumMetered(events) {
     total.tokens.input += ev.spend.tokens.input;
     total.tokens.output += ev.spend.tokens.output;
     total.usd += ev.spend.usd;
+    if (ev.spend.usdKnown === false) total.usdKnown = false;
     total.ms += ev.spend.ms;
   }
   return total;
@@ -3488,6 +3530,11 @@ function createBudgetPool(root, now = Date.now) {
     }
     open.delete(ticket.id);
     const { tokens: rTokens, usd: rUsd, iterations: rIterations } = ticket.reserved;
+    if (usdCapped && spent.usdKnown === false) {
+      throw new Error(
+        `budget pool: ticket ${ticket.id} reported unknown dollar cost under a dollar-capped budget`
+      );
+    }
     const spentTokens = totalTokens(spent.tokens);
     if (spentTokens > rTokens) {
       throw new Error(
@@ -3517,6 +3564,11 @@ function createBudgetPool(root, now = Date.now) {
     }
   }
   function observe(spend) {
+    if (usdCapped && spend.usdKnown === false) {
+      throw new Error(
+        "budget pool: cannot observe unknown dollar cost under a dollar-capped budget"
+      );
+    }
     const tokens = totalTokens(spend.tokens);
     freeTokens -= tokens;
     committedTokens += tokens;
@@ -3741,6 +3793,7 @@ function accumulate(a, b) {
   a.tokens.input += b.tokens.input;
   a.tokens.output += b.tokens.output;
   a.usd += b.usd;
+  if (b.usdKnown === false) a.usdKnown = false;
   a.ms += b.ms;
 }
 function addSpend(a, b) {
@@ -3748,6 +3801,7 @@ function addSpend(a, b) {
     iterations: a.iterations + b.iterations,
     tokens: { input: a.tokens.input + b.tokens.input, output: a.tokens.output + b.tokens.output },
     usd: a.usd + b.usd,
+    ...a.usdKnown === false || b.usdKnown === false ? { usdKnown: false } : {},
     ms: a.ms + b.ms
   };
 }
@@ -6571,4 +6625,4 @@ export {
   workerFromBackend,
   supervise
 };
-//# sourceMappingURL=chunk-FHZHADIM.js.map
+//# sourceMappingURL=chunk-TTWMSNFZ.js.map