feat(supervise): re-attach bridge sessions for continuity 'resume'

CHANGELOG.md

@@ -1,5 +1,18 @@
 # Changelog
 
+## 0.135.0
+
+### workerFromBackend honors `continuity: 'resume'` on the bridge backend
+
+The backend-derived worker seam re-attaches cli-bridge sessions (#694, the chat-transport resume executor's bridge arm).
+A bridge session id IS the harness conversation key — cli-bridge maps it to the CLI's own resume (opencode `-s <id>`, claude `--resume`) — so a resume spawn is real session re-attachment, never a fresh session wearing a `resume` stamp.
+
+- `workerFromBackend` records the session id each supervised bridge spawn was bound to, keyed by the worker id the Scope assigned. A `continuity: 'resume'` spawn binds the prior worker's recorded session id (`spawnContext.resume.ofWorker`) instead of deriving a fresh one, so the new worker continues the exact harness conversation. A fresh → resume → resume chain stays on ONE session.
+- The record is process-local by construction, which matches the kernel's documented resume boundary: a prior process's workers are not resume targets.
+- Fail-loud contract unchanged everywhere else: a non-bridge backend still refuses a resume spawn (its executors have no re-attachable session), and a resume of a worker this seam never bound a session for refuses by name. Every refusal throws BEFORE a worker exists, so the kernel never ledgers `continuity: 'resume'` over a session that was not re-attached.
+- `runGraph` handed a `backend: 'bridge'` config now honors a delegates edge's `continuity: 'resume'` natively — the revise-edge pattern (write fresh, revise by resuming the writer's session) runs without a custom `makeWorkerAgent`.
+- FIX, found by the live proof: the derived external session id now scopes by the spawning manager node. Assignment ordinals restart at `ordinal:0` under every manager, so the unscoped digest mapped worker 1 of EVERY run — and of every sibling manager — to ONE bridge session id; on a bridge with a persistent session store, a `'fresh'` spawn then silently continued a foreign run's harness conversation (measured live: three separate runs shared one claude conversation, `turns: 6`). Durable recovery within a run is unchanged (a replay re-issues the same manager node id and assignments). Migration: a journaled run from an older version replayed under this version derives DIFFERENT session ids and will not re-attach its old harness sessions.
+
 ## 0.134.4
 
 - Consume Core 0.8.0, Eval 0.145.11, Interface 0.52.0, Knowledge 7.2.6, Profile Materialize 0.14.2, and Sandbox 0.26.1 as one compatible dependency set.

docs/api/primitive-catalog.md

@@ -7,7 +7,7 @@
 
 # Primitive catalog — the never-stale anti-reinvention inventory
 
-> **GENERATED** from `@tangle-network/agent-runtime@0.134.9` and `@tangle-network/agent-eval@0.145.11` by `scripts/gen-primitive-catalog.mjs`. Do NOT hand-edit — run `pnpm run docs:api`. This is the mechanical companion to the JUDGMENT in `canonical-api.md` (§2 decision table + §1.5 AgentProfile law): that doc says WHICH primitive to reach for and what NOT to build; this catalog proves WHAT exists. Per-symbol signatures + `file:line` live in the per-module pages under `docs/api/`.
+> **GENERATED** from `@tangle-network/agent-runtime@0.135.0` and `@tangle-network/agent-eval@0.145.11` by `scripts/gen-primitive-catalog.mjs`. Do NOT hand-edit — run `pnpm run docs:api`. This is the mechanical companion to the JUDGMENT in `canonical-api.md` (§2 decision table + §1.5 AgentProfile law): that doc says WHICH primitive to reach for and what NOT to build; this catalog proves WHAT exists. Per-symbol signatures + `file:line` live in the per-module pages under `docs/api/`.
 
 ## 1. agent-runtime — own public surface
 

docs/api/runtime.md

@@ -26035,6 +26035,16 @@ construction, so a caller may hand back something the run only learns later —
 `supervise()` gives a traced run's workers their trace context without ordering the span recorder
 ahead of the worker seam.
 
+Continuity: the `bridge` backend honors `continuity: 'resume'` by session re-attachment. A
+bridge session id IS the harness conversation key (cli-bridge maps it to the CLI's own resume —
+opencode `-s <id>`, claude `--resume`), so this seam records the session id each supervised
+spawn was bound to, keyed by the worker id the Scope assigned, and a resume spawn binds the
+prior worker's recorded session id instead of deriving a fresh one. The record is process-local
+by construction, which matches the kernel's resume boundary (a prior process's workers are not
+resume targets). Every other backend keeps failing loud: their executors have no re-attachable
+session, and accepting the spawn would ledger `continuity: 'resume'` over a brand-new session —
+a stamp asserting something that never happened.
+
 #### Parameters
 
 ##### backend

docs/canonical-api.md

@@ -4,7 +4,7 @@
 Generated signatures and the complete export list live in docs/api/.
 Run pnpm docs:freshness after editing this file. -->
 
-> **Version 0.134.9.**
+> **Version 0.135.0.**
 > [`docs/api/primitive-catalog.md`](./api/primitive-catalog.md) lists every export and import path.
 > `agent-eval` must satisfy `>=0.145.11 <0.146.0`.
 > `sandbox` must satisfy `>=0.26.1 <0.27.0`.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "@tangle-network/agent-runtime",
-  "version": "0.134.9",
+  "version": "0.135.0",
   "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": {

src/runtime/index.ts

@@ -548,8 +548,9 @@ export {
 } from './supervise/budget'
 // The chat-transport leaf (#721): a worker that IS a model conversation on a bare
 // OpenAI-compatible /v1/chat/completions endpoint — no sandbox. Ships with its session store and
-// the continuity-honoring `makeWorkerAgent` seam (the resume consumer `workerFromBackend`
-// refuses to be), so conversation graphs and chat-shot loops compose from data.
+// the continuity-honoring `makeWorkerAgent` seam. `workerFromBackend` resumes only bridge-backed
+// workers; this leaf resumes its own sessions, so conversation graphs and chat-shot loops
+// compose from data on any transport.
 export {
   type ChatCompletionsTransport,
   type ChatSessionStore,

src/runtime/supervise/supervise.ts

@@ -141,6 +141,16 @@ import { WORKER_TRACE_PROPAGATION } from './worker-trace'
  * construction, so a caller may hand back something the run only learns later — which is exactly how
  * `supervise()` gives a traced run's workers their trace context without ordering the span recorder
  * ahead of the worker seam.
+ *
+ * Continuity: the `bridge` backend honors `continuity: 'resume'` by session re-attachment. A
+ * bridge session id IS the harness conversation key (cli-bridge maps it to the CLI's own resume —
+ * opencode `-s <id>`, claude `--resume`), so this seam records the session id each supervised
+ * spawn was bound to, keyed by the worker id the Scope assigned, and a resume spawn binds the
+ * prior worker's recorded session id instead of deriving a fresh one. The record is process-local
+ * by construction, which matches the kernel's resume boundary (a prior process's workers are not
+ * resume targets). Every other backend keeps failing loud: their executors have no re-attachable
+ * session, and accepting the spawn would ledger `continuity: 'resume'` over a brand-new session —
+ * a stamp asserting something that never happened.
  */
 export function workerFromBackend(
   backend: ExecutorConfig,
@@ -150,37 +160,53 @@ export function workerFromBackend(
   const capturedBackend = captureReusableExecutorConfig(backend, 'workerFromBackend')
   const unscopedNamespace = randomUUID()
   let unscopedOrdinal = 0
+  const bridgeSessionByWorker = new Map<string, string>()
   return (rawProfile, spawnContext) => {
     const parsed = agentProfileSchema.safeParse(rawProfile)
     if (!parsed.success) {
       throw new ValidationError(`workerFromBackend: invalid AgentProfile: ${parsed.error.message}`)
     }
     const profile = parsed.data
     assertBackendProfileMaterialization(profile, capturedBackend, 'workerFromBackend')
-    // Fail closed on a resume this seam cannot honor: workerFromBackend creates a NEW executor
-    // per spawn and has no session re-attachment, so accepting a 'resume' spawn would ledger
-    // `continuity: 'resume'` over a brand-new session — a stamp asserting something that never
-    // happened. Custom makeWorkerAgent seams that re-attach sessions are the resume consumers.
-    if (spawnContext?.continuity === 'resume') {
-      throw new ValidationError(
-        'workerFromBackend: this backend seam does not re-attach sessions and cannot honor ' +
-          "continuity: 'resume' — provide a makeWorkerAgent that resumes (it receives " +
-          "spawnContext.resume.ofWorker), or use continuity: 'fresh'",
-      )
-    }
+    // Resolved BEFORE any worker exists, so a resume this seam cannot honor fails the spawn
+    // itself and the kernel never ledgers a `continuity: 'resume'` stamp for it.
+    const resumeSessionId = bridgeResumeSessionId(
+      capturedBackend,
+      spawnContext,
+      bridgeSessionByWorker,
+    )
     const name = profile.name ?? 'worker'
     // A Scope assignment is stable across reconstruction. Direct callers that omit that context
     // still get isolation, but only Scope-backed calls claim durable external-session recovery.
     const assignmentId =
       spawnContext?.assignmentId ?? `unscoped:${unscopedNamespace}:${unscopedOrdinal++}`
+    // The derived execution id scopes by the spawning manager: assignment ordinals restart at
+    // `ordinal:0` under every manager, so an unscoped digest maps worker 1 of EVERY run (and of
+    // every sibling manager) to one external session — on a bridge with a persistent session
+    // store, a 'fresh' spawn then silently continues a foreign run's harness conversation.
+    // Durable recovery is preserved: a replay of the same run re-issues the same manager node id
+    // and the same assignments, so it derives the same session ids.
+    const executionScope = spawnContext?.parentNodeId
     const boundBackend = bindReusableExecutorExecutionId(
       capturedBackend,
-      externalExecutionId('supervised-worker', { assignmentId }),
+      resumeSessionId ??
+        externalExecutionId('supervised-worker', {
+          assignmentId,
+          ...(executionScope === undefined ? {} : { scope: executionScope }),
+        }),
     )
+    const boundSessionId = boundBackend.backend === 'bridge' ? boundBackend.sessionId : undefined
     const baseFactory = createExecutor(boundBackend)
     // Carry the configured factory into Scope. It is built only AFTER reservation with the real
     // child signal/context, so a rejected or already-completed keyed spawn creates no executor.
     const executorFactory = (spec: AgentSpec, ctx: ExecutorContext) => {
+      // Record the bridge session this worker is bound to under its Scope-assigned worker id,
+      // so a later resume of the node can re-attach it. A resumed worker records the SAME
+      // session under its own id, which is what keeps a fresh → resume → resume chain on one
+      // harness conversation.
+      if (boundSessionId !== undefined && ctx.node?.nodeId !== undefined) {
+        bridgeSessionByWorker.set(ctx.node.nodeId, boundSessionId)
+      }
       // Caller-supplied seams sit UNDER the per-child seams the Scope seeds, so the scope's
       // recursion and trace context always win on a key collision.
       const extraSeams = seams?.()
@@ -207,6 +233,43 @@ function externalExecutionId(kind: string, identity: unknown): string {
   return `${kind}-${digest.slice('sha256:'.length)}`
 }
 
+/**
+ * The bridge session id a `'resume'` spawn re-attaches, or `undefined` for a fresh spawn.
+ * Every refusal throws BEFORE a worker exists, which is what keeps the kernel's continuity
+ * ledger true: a `'resume'` stamp can only appear over a session that was actually re-attached.
+ */
+function bridgeResumeSessionId(
+  backend: ExecutorConfig,
+  spawnContext: WorkerSpawnContext | undefined,
+  sessions: ReadonlyMap<string, string>,
+): string | undefined {
+  if (spawnContext?.continuity !== 'resume') return undefined
+  if (backend.backend !== 'bridge') {
+    throw new ValidationError(
+      `workerFromBackend: the '${backend.backend}' backend seam does not re-attach sessions and ` +
+        "cannot honor continuity: 'resume' — only the 'bridge' backend resumes here (cli-bridge " +
+        'keys the harness conversation by session id). Provide a makeWorkerAgent that resumes ' +
+        "(it receives spawnContext.resume.ofWorker), or use continuity: 'fresh'",
+    )
+  }
+  const ofWorker = spawnContext.resume?.ofWorker
+  if (ofWorker === undefined) {
+    throw new ValidationError(
+      "workerFromBackend: a 'resume' spawn carries no resume lineage — the kernel stamps " +
+        'spawnContext.resume for ledgered spawns; a direct caller must pass { ofWorker, sequence }',
+    )
+  }
+  const prior = sessions.get(ofWorker)
+  if (prior === undefined) {
+    throw new ValidationError(
+      `workerFromBackend: no recorded bridge session for worker '${ofWorker}' — this seam ` +
+        're-attaches only sessions it bound in this process (the kernel resume boundary). ' +
+        "Spawn the node fresh first, or use continuity: 'fresh'",
+    )
+  }
+  return prior
+}
+
 /**
  * The `trace-unpropagated` declaration for a worker backend, or `undefined` when the backend HAS a
  * propagation channel. The census (`WORKER_TRACE_PROPAGATION`) says WHETHER a backend propagates;

src/testing/fixtures/agent-improvement-proposal.json

@@ -1,6 +1,6 @@
 {
   "changedSurfaces": ["prompt"],
-  "digest": "sha256:bdcb2a49dec1a39ec71d01f845da6a2c6e55282805e271eff52a1b176a938601",
+  "digest": "sha256:2c580eafb8ac241bcb7a5d90ed4d400dc469e70776be38e92d0ca3cca8b1b3ba",
   "evaluation": {
     "decision": {
       "contributingChecks": [
@@ -4882,7 +4882,7 @@
     ],
     "metadata": {
       "fixture": "agent-improvement-proposal",
-      "runtimeVersion": "0.134.9"
+      "runtimeVersion": "0.135.0"
     },
     "objectives": [
       {
@@ -4993,8 +4993,8 @@
       "baselineContentHash": "sha256:5c21ee53e513fc604cb09754e21c392b24a424da0ef37dbf8f1ee4a8a0b08f09",
       "candidateContentHash": "sha256:60fcbb1c728194bd51d7d19cb732d1c3f1881dce7e0a6266b41c8b98cfd65693",
       "kind": "agent-eval-loop",
-      "recordDigest": "sha256:b6bb5f018d1f2d5210293918085073c1fa2f78fc21faf01364dddf8560af69ac",
-      "runId": "agent-runtime-0.134.9-proposal-fixture",
+      "recordDigest": "sha256:cb02f870cf34f127846fa273bcd957b888dcbe8b6021f049272d2bafe021a8f5",
+      "runId": "agent-runtime-0.135.0-proposal-fixture",
       "schema": "agent-candidate-experiment"
     }
   },
@@ -5021,5 +5021,5 @@
   ],
   "kind": "agent-improvement-proposal",
   "proposedAt": "2026-07-10T01:00:00.000Z",
-  "runId": "agent-runtime-0.134.9-proposal-fixture"
+  "runId": "agent-runtime-0.135.0-proposal-fixture"
 }

src/testing/fixtures/agent-profile-improvement-proposal.json

@@ -1,6 +1,6 @@
 {
   "changedSurfaces": ["prompt", "skills"],
-  "digest": "sha256:5bf45cc86b707dbae89345609e3ee8fe0c4e7444f15bf4fbcc9d3369f5ae9274",
+  "digest": "sha256:fcf087fac77b1e1cc64c7735aa120df7cd66366d595ea903c156d04449eb9412",
   "evaluation": {
     "decision": {
       "contributingChecks": [
@@ -1715,7 +1715,7 @@
     ],
     "metadata": {
       "fixture": "agent-profile-improvement-proposal",
-      "runtimeVersion": "0.134.9"
+      "runtimeVersion": "0.135.0"
     },
     "objectives": [
       {
@@ -1826,7 +1826,7 @@
       "baselineContentHash": "sha256:21c495a37c418c10bde64fbaa188beddeed31f1f051ea60a6a6582a9ee0db704",
       "candidateContentHash": "sha256:103f77bc8481601eef1ad5fe6ba84a40dffabc3a44f421f8c8559121edab84e9",
       "kind": "agent-eval-loop",
-      "recordDigest": "sha256:15c4af0c9eca66190ac1690ecf4fcc5ec400919e947f310354d230d3a90d3406",
+      "recordDigest": "sha256:1615f2b547311bbba123dea28a11004d09b60191bcbab4637f0364fb39975868",
       "runId": "profile-improvement-1",
       "schema": "agent-profile-improvement-experiment"
     }