fix(improvement): honor configured dispatch deadlines

docs/api/index.md

@@ -6419,7 +6419,7 @@ Improvement-loop coordinates. Present when called through improvementDriver.
 
 ###### costLedger?
 
-`CostLedger`
+`CostLedgerHandle`
 
 Shared run-wide paid-call account supplied by agent-eval 0.117+.
 
@@ -13212,7 +13212,7 @@ Full-agentic `CandidateGenerator` (the `shots=N, sandbox=on` setting): run a rea
 
 > **commandVerifier**(`command`, `args?`, `timeoutMs?`): [`Verifier`](#verifier)
 
-Defined in: [improvement/agentic-generator.ts:860](https://github.com/tangle-network/agent-runtime/blob/main/src/improvement/agentic-generator.ts#L860)
+Defined in: [improvement/agentic-generator.ts:863](https://github.com/tangle-network/agent-runtime/blob/main/src/improvement/agentic-generator.ts#L863)
 
 A `Verifier` that runs a command in the worktree: exit 0 ⇒ ok, any other
  exit ⇒ failed with stdout+stderr as feedback. The common case — verify by

docs/canonical-api.md

@@ -2,7 +2,7 @@
 
 <!-- This doc is the JUDGMENT layer: the mental model (§1), the AgentProfile law (§1.5), and the anti-reinvention decision table (§2) — WHICH primitive to reach for and what NOT to build. The export INVENTORY (WHAT exists) and per-symbol signatures + `file:line` are GENERATED into `docs/api/` (TypeDoc + `scripts/gen-primitive-catalog.mjs`, do NOT hand-edit) — that is the mechanical reference: `docs/api/primitive-catalog.md` is the never-stale list of every primitive to reuse. The freshness gate (`pnpm docs:freshness`) FAILS CI if a version pin, a cited `file:line`, a decision-table symbol, or the generated catalog drifts from source — see `docs/MAINTAINING.md`. Keep this file the small, hand-curated spine; never re-list the inventory here — point at the catalog. -->
 
-> **Version 0.94.11.** The export inventory + per-symbol signatures live in the generated `docs/api/` reference: **`docs/api/primitive-catalog.md`** is the never-stale, grouped list of every primitive to reuse (own surface + the agent-eval judge / authenticity / verification / statistics / campaign / token-usage surfaces), with each one's import path and one-line summary read live from source; the per-module pages hold the full signatures. The pinned substrate is agent-eval `>=0.117.1 <0.118.0`; the sandbox substrate that materializes profiles into harness shapes is `@tangle-network/sandbox` (peer `>=0.8.0 <1.0.0`). The neutral contract types (`AgentProfile`, `AgentProfileMcpServer`, `HarnessType`, `ReasoningEffort`, `Part`/`ToolPart`/`ToolState`, plus environment-provider types) are owned by **`@tangle-network/agent-interface`** (peer `>=0.25.0 <0.26.0`) — the single source of truth. Substrate primitives are re-exported through `@tangle-network/agent-eval/contract` (or `/campaign`), not local to this package — the catalog's §2 shows exactly which subpath each lives under.
+> **Version 0.94.12.** The export inventory + per-symbol signatures live in the generated `docs/api/` reference: **`docs/api/primitive-catalog.md`** is the never-stale, grouped list of every primitive to reuse (own surface + the agent-eval judge / authenticity / verification / statistics / campaign / token-usage surfaces), with each one's import path and one-line summary read live from source; the per-module pages hold the full signatures. The pinned substrate is agent-eval `>=0.120.1 <0.121.0`; the sandbox substrate that materializes profiles into harness shapes is `@tangle-network/sandbox` (peer `>=0.8.0 <1.0.0`). The neutral contract types (`AgentProfile`, `AgentProfileMcpServer`, `HarnessType`, `ReasoningEffort`, `Part`/`ToolPart`/`ToolState`, plus environment-provider types) are owned by **`@tangle-network/agent-interface`** (peer `>=0.25.0 <0.26.0`) — the single source of truth. Substrate primitives are re-exported through `@tangle-network/agent-eval/contract` (or `/campaign`), not local to this package — the catalog's §2 shows exactly which subpath each lives under.
 >
 > **`./loops` is the runtime barrel** — `package.json` maps it to `src/runtime/index.ts`. Everything below labelled `/loops` is the recursive-atom + loop-kernel surface.
 >

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-runtime",
-  "version": "0.94.11",
+  "version": "0.94.12",
   "description": "Shared task-lifecycle skeleton for agents: a recursive loop kernel for chat turns, one-shot tasks, and multi-attempt loops, with trace capture and eval-gated self-improvement. Domain behavior lives in adapters; scoring and ship-gates in @tangle-network/agent-eval.",
   "homepage": "https://github.com/tangle-network/agent-runtime#readme",
   "repository": {
@@ -116,7 +116,7 @@
   },
   "devDependencies": {
     "@biomejs/biome": "^2.4.15",
-    "@tangle-network/agent-eval": "0.117.1",
+    "@tangle-network/agent-eval": "0.120.1",
     "@tangle-network/agent-interface": "^0.25.0",
     "@tangle-network/sandbox": "^0.9.7",
     "@types/node": "^25.9.3",
@@ -147,7 +147,7 @@
   "license": "MIT",
   "packageManager": "pnpm@10.28.0",
   "peerDependencies": {
-    "@tangle-network/agent-eval": ">=0.117.1 <0.118.0",
+    "@tangle-network/agent-eval": ">=0.120.1 <0.121.0",
     "@tangle-network/agent-interface": ">=0.25.0 <0.26.0",
     "@tangle-network/sandbox": ">=0.8.0 <1.0.0",
     "playwright": "^1.40.0"

src/improvement/agentic-generator.ts

@@ -38,7 +38,7 @@ import { existsSync, readFileSync, rmSync } from 'node:fs'
 import { join, resolve, sep } from 'node:path'
 import type {
   AnalystFinding,
-  CostLedger,
+  CostLedgerHandle,
   CostReceipt,
   CostReceiptInput,
   MaximumCharge,
@@ -246,10 +246,14 @@ export function agenticGenerator(opts: AgenticGeneratorOptions = {}): CandidateG
       costLedger,
       costPhase,
     }) {
-      if (opts.codexReproducible && !costLedger) {
-        throw new Error(
-          'agenticGenerator: reproducible Codex requires the run-wide CostLedger supplied by agent-eval 0.117+',
-        )
+      let reproducibleCostLedger: CostLedgerHandle | undefined
+      if (opts.codexReproducible) {
+        if (!costLedger) {
+          throw new Error(
+            'agenticGenerator: reproducible Codex requires the run-wide CostLedger supplied by agent-eval 0.117+',
+          )
+        }
+        reproducibleCostLedger = costLedger
       }
       const basePrompt = appendProfileResourcePaths(
         buildPrompt({ report, findings }),
@@ -314,13 +318,12 @@ export function agenticGenerator(opts: AgenticGeneratorOptions = {}): CandidateG
             return harnessResult
           }
 
-          if (opts.codexReproducible) {
-            const ledger = costLedger as CostLedger
+          if (reproducibleCostLedger) {
             const model = opts.profile?.model?.default
             if (!model) {
               throw new Error('agenticGenerator: reproducible Codex requires profile.model.default')
             }
-            const paid = await ledger.runPaidCall({
+            const paid = await reproducibleCostLedger.runPaidCall({
               channel: 'driver',
               phase: costPhase ?? 'search.proposal',
               actor: `agentic-generator:${harness}`,

src/improvement/improve.test.ts

@@ -474,7 +474,6 @@ describe('improve() — default proposer resolution (substrate export drift guar
       onProvenance: () => {
         provenanceCalls += 1
       },
-      collectWorkerRecords: () => [],
       expectUsage: 'assert',
       captureSource: 'eval-run',
       autoOnPromote: 'none',

src/improvement/improvement-driver.ts

@@ -21,7 +21,7 @@
  */
 
 import { spawnSync } from 'node:child_process'
-import type { AnalystFinding, CostLedger } from '@tangle-network/agent-eval'
+import type { AnalystFinding, CostLedgerHandle } from '@tangle-network/agent-eval'
 import {
   type CodeSurface,
   type LabeledScenarioStore,
@@ -66,7 +66,7 @@ export interface CandidateGenerator {
     generation?: number
     candidateIndex?: number
     /** Shared run-wide paid-call account supplied by agent-eval 0.117+. */
-    costLedger?: CostLedger
+    costLedger?: CostLedgerHandle
     /** Receipt attribution phase supplied alongside `costLedger`. */
     costPhase?: string
   }): Promise<{ applied: boolean; summary: string }>

tests/improve.test.ts

@@ -88,6 +88,41 @@ describe('improve() — facade over selfImprove', () => {
     expect(out.profile.prompt?.systemPrompt).toBe('BASELINE')
   })
 
+  it('forwards dispatchTimeoutMs through improve() to abort a stalled cell', async () => {
+    let observedAbort = false
+    const stalledAgent = (
+      _surface: MutableSurface,
+      _scenario: DemoScenario,
+      ctx: DispatchContext,
+    ): Promise<string> =>
+      new Promise(() => {
+        ctx.signal.addEventListener(
+          'abort',
+          () => {
+            observedAbort = true
+          },
+          { once: true },
+        )
+      })
+
+    await expect(
+      improve(profileWith('BASELINE'), [], {
+        surface: 'prompt',
+        gate: 'none',
+        generator: scriptedWinner,
+        scenarios: scenarios.slice(0, 2),
+        judge,
+        agent: stalledAgent,
+        budget: { holdoutFraction: 0.5 },
+        dispatchTimeoutMs: 25,
+        expectUsage: 'off',
+        maxConcurrency: 1,
+      }),
+    ).rejects.toThrow(/dispatch exceeded 25ms/)
+
+    expect(observedAbort).toBe(true)
+  })
+
   it('applies the promoted prompt back into the profile on a ship verdict', async () => {
     const profile = profileWith('BASELINE')
     const out = await improve(profile, [], {