dimcode-darwin-x64 0.1.0-alpha.16 → 0.1.0-alpha.17

package/bin/dimcode

@@ -1,4 +1,4 @@
 [diffend] Oversized file quarantined before diffing.
 name: package/bin/dimcode
-size: 98714040 bytes
-sha256: 70eb60ca30debdd2e226d5c2914da6a56f6e67cc6fe6d6d6ca00eff8fad4b03d
+size: 104311000 bytes
+sha256: bbeb81fb9b8bc26ca3e2447036592629025ea0c193c02f2f1cf68a2f821a4bc6

package/default-plugins.json

@@ -1,7 +1,6 @@
 {
   "plugins": {
     "goal": "plugins/goal",
-    "session-title": "plugins/session-title",
     "workflow": "plugins/workflow",
     "scheduler": "plugins/scheduler",
     "plugin-authoring": "plugins/plugin-authoring"

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "dimcode-darwin-x64",
-  "version": "0.1.0-alpha.16",
+  "version": "0.1.0-alpha.17",
   "description": "Dim native CLI and App Server",
   "license": "SEE LICENSE IN NOTICE",
   "repository": {

package/plugins/scheduler/README.md

@@ -114,7 +114,7 @@ Run from the repository root, using Node 24:
 ```sh
 npm ci --prefix runtime/plugins/scheduler --ignore-scripts
 node runtime/node_modules/typescript/bin/tsc -p runtime/plugins/scheduler/tsconfig.json
-node --test runtime/plugins/scheduler/scheduler.test.mjs
+cargo test --locked -p dim-app-server --test transport_flow node_scheduler -- --ignored
 ```
 
 Before native flows, reinstall the local packages using the CI
@@ -124,12 +124,8 @@ package copy; source edits alone do not refresh it.
 
 | Public flow | Evidence |
 | --- | --- |
-| Save takeover, CAS failure, live permissions, shared cap/reuse vectors, tool restrictions and retirement | SDK service flows in `scheduler.test.mjs` |
-| Cron DST gap/fold, enabled anchor, missed windows and one-off disable | SDK flows with a controlled wall clock |
-| Held/consumed/applied input deadlines, uncertain withdrawal, late launch/reuse replies and revision rechecks | SDK execution flows |
-| Restart lookup/history, single-attempt reports, admitted-report cleanup and deletion retry | SDK persistence/lifetime flows |
 | Actual source defaults/protected fields, tool save, later live widening bounded by the saved cap, non-interactive execution, takeover, reporting and inner/outer-v1 records | `node_scheduler_saves_live_authority_runs_noninteractive_and_reports_to_the_latest_saver` |
-| An untitled interactive leader adopts its saving operator input before the raw Steer report; subsequent automatic title generation sends the original operator text to the real auxiliary HTTP binding and publishes its title | `node_scheduler_reports_preserve_the_first_adopted_operator_title` |
+| An untitled interactive leader adopts its saving operator input before the raw Steer report; subsequent automatic title generation sends the original operator text to the real auxiliary HTTP binding and publishes its title | [`scheduler_reports_do_not_replace_the_first_operator_input_for_titles`](../../../crates/dim-session-title/tests/scheduler_flow.rs) |
 | Required invalid-record startup, unchanged bytes, failed optional owner and usable native Sessions | `node_scheduler_invalid_inner_records_preserve_bytes_and_do_not_block_native_sessions` |
 | Agents index replacement keeps the owner and run loaded; owner reload preserves pending approval, input, deadline and anchor; completion/report occur once | `node_scheduler_owner_reload_tracks_the_existing_native_run_without_replaying_input_or_reports` |
 | Native timer windows, DST gap/fold, enabled-anchor changes, one-off disable and missed-window collapse | `node_scheduler_native_clock_preserves_anchors_collapses_gaps_and_handles_dst_once` |
@@ -141,7 +137,7 @@ package copy; source edits alone do not refresh it.
 | Real filesystem deletion/replace failures preserve the last ownership record and retry with no jobs; external Sessions survive | `node_scheduler_retention_retries_real_delete_and_record_failures_with_no_jobs` |
 | Fresh and pre-interrupt history prefixes preserve applied inputs without model consumption, deletion or another stop; later reuse consumption cannot change the terminal record | `node_scheduler_applied_inputs_interrupted_before_consumption_expire_without_another_stop` |
 | A stale history prefix followed by genuine notHeld re-reads native deletion, completion or consumption before deciding | `node_scheduler_not_held_rereads_native_deletion_completion_and_consumption` |
-| All shared cap and reuse vectors pass real authorizing saves, native launch policies and reused-policy checks | `node_scheduler_shared_policy_vectors_reach_native_save_launch_and_reuse` |
+| Shared cap/reuse vectors and unrepresentable saves use real native policies; default-policy groups share one host while job identities stay separate | `node_scheduler_shared_policy_vectors_reach_native_save_launch_and_reuse` |
 | Selected/other job edits, one-CAS conflict, frozen accepted prompt/authority/report target and accepted same-revision local writes | `node_scheduler_job_edits_cas_and_local_source_writes_preserve_accepted_snapshots` |
 | Both job origins preserve completed/failed/cancelled/timed-out/skipped/missed history; only selected tool outcomes send one Steer report with model-visible identities, and reload sends none | `node_scheduler_reports_only_selected_tool_outcomes_with_model_visible_identity` |
 | Closed/deleted report targets and a discarded reply after actual native publication preserve the outcome, record failure once and never redirect or resend after reload | `node_scheduler_failed_reports_keep_the_outcome_and_never_retry_or_redirect` |
@@ -153,8 +149,7 @@ A prefix captured before a user interrupt can permit a withdrawal command at the
 InputQueue returns `notHeld` for the applied input; the command does not imply an `input/deleted`
 fact. The native check preserves this distinction and still requires no model call or extra stop.
 
-The SDK suite exercises the public process/Session services; the native matrix
-runs with the repository's public Node binding setup in CI, including the required
+The native matrix exercises the public process/Session services and runs with the repository's public Node binding setup in CI, including the required
 invalid-record case. The test fixture wraps the real package's public ports
 and uses Node's built-in test clock. Native SQLite publication and filesystem
 permissions supply the report and retention faults; production interfaces are

package/plugins/scheduler/data.ts

@@ -9,10 +9,9 @@ export class ScheduleError extends Error {
 
 export const id = z.string().min(1)
 export const preset = z.enum(['read-only', 'workspace-write', 'full-access'])
-const fileAccess = z.enum(['read-only', 'workspace-read-write', 'unrestricted'])
-export const policy = z.strictObject({ preset, fileAccess: fileAccess.optional(),
-  toolApproval: z.enum(['ask', 'bypass']).optional(),
-  allowedTools: z.array(z.string()), disallowedTools: z.array(z.string()) }) satisfies z.ZodType<PermissionPolicy>
+export const policy = z.object({
+  preset, allowedTools: z.array(z.string()), disallowedTools: z.array(z.string()),
+}) satisfies z.ZodType<PermissionPolicy>
 export const timestamp = z.iso.datetime({ precision: 3 }).max(24)
 export const address = z.strictObject({ session_id: id, branch_id: id })
 const cronField = '[0-9*/,-]+'
@@ -46,37 +45,17 @@ export type JobPatch = z.infer<typeof managementJob>
 /** The same conservative policy cap used by S2; lists are sets. */
 export function cap(requested: PermissionPolicy, bound: PermissionPolicy) {
   const ranks = preset.options
-  const legacyReadOnly = [requested, bound].some(value => approval(value) === 'legacy-read-only')
-  if (legacyReadOnly && [requested, bound].some(value => value.toolApproval === 'ask')) {
-    throw new ScheduleError('schedule_invalid', 'Explicit Ask cannot be capped by legacy read-only tool rules')
-  }
   const disallowedTools = [...new Set([...requested.disallowedTools, ...bound.disallowedTools])]
+  const allows = (value: PermissionPolicy, name: string) =>
+    value.preset === 'full-access' || value.allowedTools.includes(name)
   const allowedTools = [...new Set([...requested.allowedTools, ...bound.allowedTools])].filter(name =>
-    (approval(requested) === 'bypass' || requested.allowedTools.includes(name))
-    && (approval(bound) === 'bypass' || bound.allowedTools.includes(name)) && !disallowedTools.includes(name))
-  return { preset: ranks[Math.min(ranks.indexOf(requested.preset), ranks.indexOf(bound.preset))],
-    ...(requested.fileAccess !== undefined || bound.fileAccess !== undefined ? {
-      fileAccess: fileAccess.options[Math.min(fileRank(requested), fileRank(bound))],
-    } : {}),
-    ...(!legacyReadOnly && (requested.toolApproval !== undefined || bound.toolApproval !== undefined) ? {
-      toolApproval: approval(requested) === 'bypass' && approval(bound) === 'bypass' ? 'bypass' as const : 'ask' as const,
-    } : {}), allowedTools, disallowedTools } satisfies PermissionPolicy
-}
-
-function fileRank(value: PermissionPolicy): number {
-  return value.fileAccess === undefined ? preset.options.indexOf(value.preset) : fileAccess.options.indexOf(value.fileAccess)
-}
-
-function approval(value: PermissionPolicy) {
-  return value.toolApproval ?? (value.preset === 'full-access' ? 'bypass' : value.preset === 'read-only' ? 'legacy-read-only' : 'ask')
+    allows(requested, name) && allows(bound, name) && !disallowedTools.includes(name))
+  return { preset: ranks[Math.min(ranks.indexOf(requested.preset), ranks.indexOf(bound.preset))], allowedTools, disallowedTools } satisfies PermissionPolicy
 }
 
 export function within(actual: PermissionPolicy, bound: PermissionPolicy): boolean {
-  let capped: PermissionPolicy
-  try { capped = cap(actual, bound) }
-  catch (error) { if (error instanceof ScheduleError) return false; throw error }
+  const capped = cap(actual, bound)
   return actual.preset === capped.preset
-    && fileRank(actual) === fileRank(capped) && approval(actual) === approval(capped)
     && [actual.allowedTools, actual.disallowedTools].every((names, index) => {
       const result = index === 0 ? capped.allowedTools : capped.disallowedTools
       return new Set(names).size === result.length && names.every(name => result.includes(name))

package/plugins/scheduler/settings.ts

@@ -139,14 +139,6 @@ export const settings: SourceDeclaration = {
                     "full-access"
                   ]
                 },
-                "fileAccess": {
-                  "type": "string",
-                  "enum": ["read-only", "workspace-read-write", "unrestricted"]
-                },
-                "toolApproval": {
-                  "type": "string",
-                  "enum": ["ask", "bypass"]
-                },
                 "allowedTools": {
                   "type": "array",
                   "items": {
@@ -165,7 +157,6 @@ export const settings: SourceDeclaration = {
                 "allowedTools",
                 "disallowedTools"
               ],
-              "additionalProperties": false
             },
             "created_by": {
               "anyOf": [

package/plugins/workflow/README.md

@@ -103,13 +103,20 @@ An `ambiguous` reply (duplicates written before branches accepted each id once)
 delivered and records a violation. Missing destinations are undeliverable.
 
 Run storage is `workflow/runs.json`, with inner `{version:2,runs:[...]}` inside the host's S10 v1
-wrapper. Runs admitted under the bounds below keep the text within the 8 MiB payload as the file
+wrapper. `workflow/runs.version` holds the token of the last committed store write: a running stage's
+one-second wait reads that token (36 bytes) and reuses the owner's snapshot, so its per-second cost
+does not grow with the store; only a changed token, startup recovery and every write read the store
+itself. Runs admitted under the bounds below keep the text within the 8 MiB payload as the file
 service carries it (the JSON string, where escapes can double): a start projects the store with
 every active run's record, the stage identities, `agent_id` copies, results and logs it can gain
 later, plus the new record; if that exceeds the payload it first expires the oldest settled
 terminal runs and otherwise returns `limit_exceeded` without persisting. The retained store,
 therefore, keeps each read, write and `workflow.status` reply within one bridge frame; a record
-committed before these bounds keeps its original size. Invalid, v1,
+committed before these bounds keeps its original size. The token is written in its own transaction
+after the store replace, so a crash between them leaves it naming older text; recovery reads the
+store and does not depend on it. A failed token read or write is a storage failure, a failed store
+access drops the snapshot so committed bytes stay visible, and a poll reaches the frame with the
+token, never the store. Invalid, v1,
 unknown-version or missing-marker records return `invalid_data` and preserve their bytes. There is
 no migration. After a crash, unfinished runs become interrupted/lost without relaunch; a lost
 stage's running turn or pending approval is left untouched. Only orderly unload stops stages under
@@ -126,7 +133,7 @@ changes that accepted input/terminal-fact contract.
 
 ```sh
 node runtime/node_modules/typescript/bin/tsc -p runtime/plugins/workflow/tsconfig.json
-node --test runtime/plugins/workflow/workflow.test.mjs
+cargo test --locked -p dim-app-server --test transport_flow node_workflow -- --ignored
 ```
 
 Actual Node/App Server flows are in
@@ -134,10 +141,10 @@ Actual Node/App Server flows are in
 `DIM_TEST_NODE`, `DIM_TEST_JS_RUNTIME` and `DIM_TEST_RUTIS_SDK` settings and disposable homes.
 They exercise stage Sessions, published reports, resubmission across row/runtime replacement, crashes and failed
 record replacement. Native retention flows use actual Session deletion, file read/replace failures,
-fixed-row replacement and the existing one-minute maintenance timer. The acceptance mapping below
-separates native coverage from SDK coverage.
+fixed-row replacement and the real maintenance callback. The fixture advances Node interval timers
+without waiting one minute per pass; stage I/O, SQLite publication and filesystem faults remain real.
 The read-failure retry case needs a valid startup restore before making the run
-file unreadable. It retains two real timer passes; corrupting that file before
+file unreadable. It advances both maintenance passes; corrupting that file before
 row activation would exercise startup failure instead of retention recovery.
 The existing [#126](https://github.com/arcships/dim/issues/126) idle-unload work must land before
 shipping repeated stage creation. A launch whose reply is lost before its Session id is persisted
@@ -145,22 +152,18 @@ can leave an unidentified Session; it is not relaunched or recovered by a new re
 
 ## Acceptance coverage
 
-The [SDK flows](workflow.test.mjs) execute the actual modules through their declared ports.
 The [native flows](../../../crates/dim-app-server/tests/transport_flow/workflow.rs) use real
-Node rows, Settings, file authority, S2 calls, publication and Session history. This maps to
-[Workflow Appendix D](../../../docs/design/plugins/Workflow.md#appendix-d-acceptance):
+Node rows, Settings, file authority, S2 calls, publication and Session history.
 
-| Items | Evidence |
+| Public behavior | Evidence |
 | --- | --- |
-| 1 | SDK definition, template, parameter, frozen-run and whole-step overflow flows; native source identities, generic Settings writes and model-visible diagnostics |
-| 2, 5 | Native start + wait batch: report wakes wait, adoption follows the batch result, and interrupting the starter leaves accepted run work alive; native starter closure makes its report undeliverable |
-| 3, 4 | Native held, consumed/approval and completed-turn stop cases; SDK applied-before-consumption and concurrent-withdrawal cases; repeated stop and no-report assertions |
-| 6 | Native project default-disabled/client enablement, editing a running definition and the next run reading the edit |
-| 7 | Native running-stage Node crash, row/runtime report replacement, complete host restart and Agents-index reload; no stage relaunch |
-| 8–11 | Native live first-admission narrowing with the sibling launch gated, cap persistence before siblings, starter loss, non-interactive/default-profile stage Sessions with and without `workflow` in the mount list, ReadOnly denial and handler refusal when it is declared |
-| 12 | Native retention keeps the latest 100 terminal records, newer active/pending work and retained references; SDK flows prove that older active/pending records also survive the terminal limit and that a start expires older settled records for frame space; user forks and external Sessions survive owned-parent deletion. Real deletion/read/atomic-replace failures preserve ownership, minute maintenance retries without active runs or definitions, and Node restart reconciles deletion before record removal. Native terminal/report-settlement triggers continue with an unreadable definition source; native source CAS and invalid-file preservation are also covered |
-| 13 | Native resubmission across row/runtime replacement with the append held, before/after-input crashes, failed atomic replacement, a resubmitted report written once and host restart; SDK rejected, failed, ambiguous, reloading and publication-unknown replies |
-| 14 | SDK definition, user-document, run-store and listing payload flows: over-1-MiB definitions are rejected, a user save over the 8 MiB document is rejected without writing, quote/backslash/control-character results and logs are measured in their committed JSON form and a run's results stop at 512 KiB, `agent_id` copies are charged, an over-payload active admission returns `limit_exceeded` without persisting, a start expires older settled records to fit, the listing is sent once and stops at its payload with `truncated`, and the store and status replies stay within 8 MiB. Native inner-v2 records remain decodable |
+| Definition and parameter admission, source identities, project enablement, source-file preservation and whole-parallel-step prompt rejection | `node_workflow_sources_stages_policy_and_report_use_the_real_package`, `node_workflow_project_enablement_invalid_sources_and_tool_policy` |
+| Live first-admission policy, starter loss, frozen authority and native non-interactive stage Sessions | `node_workflow_captures_live_first_admission_and_survives_starter_and_index_changes` |
+| Held, consumed and completed work settles on stop and reload | `node_workflow_stop_and_reload_settle_held_consumed_and_completed_work` |
+| Reports wake wait and adopt after the start batch commits | `node_workflow_report_wakes_wait_and_is_adopted_after_its_start_batch` |
+| Crash, row/runtime replacement, complete-host restart and uncertain report publication | `node_workflow_reconciles_crashes_failed_records_and_resubmitted_reports`, `node_workflow_reports_recover_after_complete_host_restart`, `node_workflow_background_report_resubmits_after_row_and_runtime_replacement` |
+| Invalid inner records and failed change-token writes preserve committed runs | `node_workflow_rejects_invalid_inner_records_without_changing_bytes`, `node_workflow_failed_token_write_keeps_the_committed_run_readable` |
+| Retention preserves live references, user forks and external Sessions, and retries real deletion/read/record faults | The three `node_workflow_retention_*` flows |
 
 Scope cleanup also exercises a failing parallel reader while another native launch is admitted:
 the original task joins its siblings before its file scope can close. No additional lock or owner

package/plugins/workflow/store.ts

@@ -1,13 +1,12 @@
+import { randomUUID } from 'node:crypto'
 import { z } from 'zod'
-import type { FileScope, PermissionPolicy } from '../../../npm/plugin.js'
+import type { FileLease, FileScope, PermissionPolicy } from '../../../npm/plugin.js'
 import { definitionSchema, issues, logEntries, logMessageBytes, resultBytes, stageReasonBytes, stages, WorkflowError } from './definition.ts'
 
 const id = z.string().min(1)
 const timestamp = z.iso.datetime({ precision: 3 })
-const policy = z.strictObject({
+const policy = z.object({
   preset: z.enum(['read-only', 'workspace-write', 'full-access']),
-  fileAccess: z.enum(['read-only', 'workspace-read-write', 'unrestricted']).optional(),
-  toolApproval: z.enum(['ask', 'bypass']).optional(),
   allowedTools: z.array(z.string()), disallowedTools: z.array(z.string()),
 }) satisfies z.ZodType<PermissionPolicy>
 const address = z.strictObject({ session_id: id, branch_id: id })
@@ -139,19 +138,62 @@ export function projectedBytes(runs: Run[]): number {
   return total
 }
 
-/** The only decoder for the committed inner-v2 run document. */
+/** The only decoder for the committed inner-v2 run document.
+ *
+ * A poll reads `workflow/runs.version` instead of the store: it holds the
+ * token of the last committed store write, and the owner re-reads and decodes
+ * the store only when that token differs from the snapshot's. The token is
+ * derived state; startup and every write read the store itself, and a failed
+ * access drops the snapshot, so a committed replace whose token write failed
+ * stays visible to the next read. */
 export class RunStore {
   readonly scope: FileScope
+  private snapshot: { token: string | null; runs: Run[] } | undefined
+
   constructor(scope: FileScope) { this.scope = scope }
 
-  async read(): Promise<Run[]> { return this.access(runs => runs, false) }
+  async read(): Promise<Run[]> {
+    const token = await this.signal()
+    if (this.snapshot?.token === token) return this.snapshot.runs
+    const runs = await this.access(runs => runs, false)
+    this.snapshot = { token, runs }
+    return runs
+  }
+
   async update<T>(change: (runs: Run[]) => T): Promise<T> { return this.access(change, true) }
 
-  private async access<T>(change: (runs: Run[]) => T, write: boolean): Promise<T> {
+  private signal(): Promise<string | null> {
+    return this.locked('workflow/runs.version', async file => {
+      const reply = await file.read()
+      if (!reply.ok) throw new WorkflowError(reply.error)
+      return reply.value
+    })
+  }
+
+  private commit(): Promise<string> {
+    const token = randomUUID()
+    return this.locked('workflow/runs.version', async file => {
+      const reply = await file.replace(token)
+      if (!reply.ok) throw new WorkflowError(reply.error)
+      return token
+    })
+  }
+
+  private async locked<T>(path: string, work: (file: FileLease) => Promise<T>): Promise<T> {
     let value!: T
     let problem: unknown
-    const outcome = await this.scope.withLock('workflow/runs.json', async file => {
-      try {
+    const outcome = await this.scope.withLock(path, async file => {
+      try { value = await work(file) } catch (error) { problem = error }
+    })
+    if (!outcome.ok) throw new WorkflowError(outcome.error)
+    if (problem !== undefined) throw problem
+    return value
+  }
+
+  private async access<T>(change: (runs: Run[]) => T, write: boolean): Promise<T> {
+    let committed: { token: string; runs: Run[] } | undefined
+    try {
+      const value = await this.locked('workflow/runs.json', async file => {
         const text = await file.read()
         if (!text.ok) throw new WorkflowError(text.error)
         let runs: Run[]
@@ -160,18 +202,19 @@ export class RunStore {
           try { runs = storedSchema.parse(JSON.parse(text.value)).runs }
           catch { throw new WorkflowError('invalid_data') }
         }
-        value = change(runs)
+        const changed = change(runs)
         if (write) {
           const reply = await file.replace(JSON.stringify({ version: 2, runs }))
           if (!reply.ok) throw new WorkflowError(reply.error)
+          committed = { token: await this.commit(), runs }
         }
-      } catch (error) {
-        problem = error
-      }
-      return null
-    })
-    if (!outcome.ok) throw new WorkflowError(outcome.error)
-    if (problem !== undefined) throw problem
-    return value
+        return changed
+      })
+      if (committed) this.snapshot = committed
+      return value
+    } catch (error) {
+      this.snapshot = undefined
+      throw error
+    }
   }
 }

package/release-manifest.json

@@ -1,7 +1,7 @@
 {
   "schemaVersion": 1,
-  "version": "0.1.0-alpha.16",
-  "sourceCommit": "7a232d0ebe939c4f9cc7140e368a53567b76b7d8",
+  "version": "0.1.0-alpha.17",
+  "sourceCommit": "10b7088064ea7e3add35743ce35eba9ea374e7a5",
   "platform": "darwin-x64",
   "toolchain": {
     "rust": "1.98.1"
@@ -10,8 +10,8 @@
   "files": {
     "LICENSE": "39100788d88143503514bc617369f1d1c7c0265c3f8742171e6af9701b0511b0",
     "NOTICE": "dd8e025eb8f6672bef1d021ea302602d6c3402856f277d48d4ed3c1a5f0c9e4d",
-    "default-plugins.json": "cb0f1baef72b42bdf9491cfa50065a5de4fe24ce0033a1c825ad2ba69f2d0351",
-    "bin/dimcode": "70eb60ca30debdd2e226d5c2914da6a56f6e67cc6fe6d6d6ca00eff8fad4b03d",
+    "default-plugins.json": "b5be1a8552248fbe87346960163e62cf1eb8f6346a74a801afa98d20113467d4",
+    "bin/dimcode": "bbeb81fb9b8bc26ca3e2447036592629025ea0c193c02f2f1cf68a2f821a4bc6",
     "licenses/grok-build/LICENSE": "116f7778b9802e569b7fa3a532b17bd80eb13c67837def01eed093d4ea472f28",
     "licenses/grok-build/NOTICE": "8db60f5ffc55a9845974915b1c5bfc7804a9b531ee0ffc6af786b3c0031c71e0",
     "plugins/goal/branch.ts": "37b1d033d48ad26fbf254090ccdaafb4bb4dc4f4146136f8e522be0fcc62d70c",
@@ -23,8 +23,8 @@
     "plugins/goal/session.ts": "75945194cb051f6e949049afc7a12fde34e63a6f1d924f20d7c669342bcf4d46",
     "plugins/goal/settings.ts": "2414bd311c11ac9f4a5bd69b8b46f8fe980dfd60b9a6adbf3bf15507a55819f5",
     "plugins/plugin-authoring/dim-plugin.json": "d43d6bd83f3b8ac05192f3ec8cadfe39500ccd541af336108ad1504c14f6ded8",
-    "plugins/scheduler/README.md": "60163f9f6be2bf1b839f1a9ef1fcc4a0a0a04fd98acc9169a211b359844c176c",
-    "plugins/scheduler/data.ts": "cf45347945ac609ad87795b8f03b4c10b6e0cec4bb5353dc5d345b6c4e5936f8",
+    "plugins/scheduler/README.md": "5730b50a4d8e1a6c386ae7f7100ad9b61f145125f8e2fe308c26d7749ab725d3",
+    "plugins/scheduler/data.ts": "8d95f8154556d815f41e61717be5a029afc74890d9a9b432c3e44b1bfb0f8d5c",
     "plugins/scheduler/dim-plugin.json": "c6b3bb2d5b6d3cef382567d9d8f473ec5db36b7a074e9ecc80b709c558e82d32",
     "plugins/scheduler/engine.ts": "9e224c008949fbaf6b8733df2219cbc1dcd29326bcf21b73b6f2ade3db0299d1",
     "plugins/scheduler/history.ts": "709f5ba2b1885ba9a28c527855c706444a6a2a02c16598182f2534dd33b1f635",
@@ -34,17 +34,10 @@
     "plugins/scheduler/package.json": "de006697139ef0bd04e90202ffeb6358458bffd8dea7f62cfa439235b80152e0",
     "plugins/scheduler/requests.ts": "9460a51c58800c8d3e4c2e2e45b4132c4f665988ea63a10cd7fdd7bc064376e8",
     "plugins/scheduler/session.ts": "26a9f013ce9e52c476123d3a69c3583f3c060e5ae8d7f1d3efe503a21cc33edc",
-    "plugins/scheduler/settings.ts": "35e7986d10a2171e9401dda386689b7029ae5b117a62a819b6a6d1c737234ec6",
+    "plugins/scheduler/settings.ts": "ddafc111c7e6f6b86ddd47b7f4ec4d20b4c2f313b48712430e4e55a07af321a9",
     "plugins/scheduler/store.ts": "796e072e3596e81d750140b0826a298198cbc4fa474c00bccbd30f22e95e41c2",
     "plugins/scheduler/windows.ts": "4edbd2036750b7d7225df4a977e239fea179ebad42b1cdc2a585c02ba31c1783",
-    "plugins/session-title/README.md": "37fc84524e2ffd3103bb3f5873d415ce5ef4e32cec9c27caa455739f8e9bc12b",
-    "plugins/session-title/dim-plugin.json": "7d1a61a1eda6f233309395a50e8983bd244a6098681a5d3071eda73300ec4a2c",
-    "plugins/session-title/index.ts": "bc35ec91bfb3f9e14a812bbb92800349ffbee663791686537f892721dd370b40",
-    "plugins/session-title/package-lock.json": "e6f6b25573bbafeb77a19b38934c38ec515839bff15b185b40731a4e3c2a372b",
-    "plugins/session-title/package.json": "5efa959d56ea8c1926a93bf9bc80d3ab64bdd36d99d6bb406f55bc20dce4a131",
-    "plugins/session-title/part.ts": "646e0721b5c2f0962db73e426dc7c00634d89e23bce2fca9984a5456df3db5aa",
-    "plugins/session-title/settings.ts": "a7c600d6cd67c9e02f5582dbd2c558995683f49461c2dd8b0cbc8f514d56a43a",
-    "plugins/workflow/README.md": "cce53d09504923f7235e69eb001a7f869bce5d58b11aeaa29c1f12d0c7bf2e85",
+    "plugins/workflow/README.md": "d771ff56f72d99b3dc13373ad29e7be530826fd27d72b9276e573cc8e689c187",
     "plugins/workflow/catalog.ts": "f84131352323bd6d7607a521bf50427ab2bfa2d4f1be3651d544c20f25429cc7",
     "plugins/workflow/definition.ts": "4b2b90c9027cdf0f04120dee137f53caae9ead96a4702f86bc6badf00d069aee",
     "plugins/workflow/dim-plugin.json": "ee3b26ced63defe5a258ff7831c5941c7c8351984ecf31dd52c1fe002b737f37",
@@ -56,12 +49,11 @@
     "plugins/workflow/requests.ts": "226b0f83e05ee1ca52d57fc2938d1be0d7133a4585912dc8e5294eb045777674",
     "plugins/workflow/session.ts": "00b97f5eacc4d76acc8bbcad3e38aba9a50b69b95d9e612a9c588c8dada350d8",
     "plugins/workflow/settings.ts": "c906571b25c34d781170dc70fa954924a3240260f40b622944398d0496bbb484",
-    "plugins/workflow/store.ts": "2bcdf6e954da2b27e299375d4a057935e53995a8ee721aca801c38d4c731c042",
+    "plugins/workflow/store.ts": "6193ddff16a98b6d8fe36918c12db35471527745912e2e5fe71d3e6cba4f4049",
     "licenses/grok-build/xai-ratatui-inline/NOTICE": "0b2ba5c4dcc49e4e17be08fe2980968fc7bce31e41ad5af09bb80097f61483dd",
     "licenses/grok-build/xai-ratatui-textarea/NOTICE": "5d310dde613653911ea026b5cb66367600f5a704d299fa0afedbf1d3efd5d493",
     "plugins/goal/node_modules/.package-lock.json": "96b3a00b4ac808cf7c3105ebbc0abc2a9a0b980f89140e275fca7e6cfbeb88d3",
     "plugins/scheduler/node_modules/.package-lock.json": "0e81b809be53fce3e194017e7d34e8935dd4d7a85af1f90a5810627304b46356",
-    "plugins/session-title/node_modules/.package-lock.json": "1476e74ec17e24726f9c124d1eff69a63aaa7767d9c0a671f0379a1d1ab7b72a",
     "plugins/workflow/node_modules/.package-lock.json": "3716d0f89cc04eda67b76c126708b242459beb13ddaeba4c3f8b3e112e250736",
     "plugins/goal/node_modules/zod/LICENSE": "3f1189b28e3866e0d979968d466b78f813f76827cfdca1fbb124cc0a5c8841f8",
     "plugins/goal/node_modules/zod/README.md": "0ae96d90800b51cee45fe028e14474db1e58ea6479e4d9ed02de28b53c27500e",
@@ -207,8 +199,6 @@
     "plugins/scheduler/node_modules/zod/v4-mini/index.d.ts": "92f42fea7c90ec3f2f623a923d8798d2a316cb4ff93d6399084065b9214365de",
     "plugins/scheduler/node_modules/zod/v4-mini/index.js": "92f42fea7c90ec3f2f623a923d8798d2a316cb4ff93d6399084065b9214365de",
     "plugins/scheduler/node_modules/zod/v4-mini/package.json": "3e3125b5a537a24c3632c1c5c09640ce7bcb3935901b127ad8ae3c37a27f133f",
-    "plugins/session-title/node_modules/@arcships/rutis/README.md": "765eddfcf43a3069d5e67168b83cbfe85bc298e3b1927fcbcdc0f87a144278eb",
-    "plugins/session-title/node_modules/@arcships/rutis/package.json": "c803303b7d7dbf9e3b60d192758ea6c4d8b6853efed43e86075b96850d010116",
     "plugins/workflow/node_modules/@arcships/rutis/README.md": "765eddfcf43a3069d5e67168b83cbfe85bc298e3b1927fcbcdc0f87a144278eb",
     "plugins/workflow/node_modules/@arcships/rutis/package.json": "c803303b7d7dbf9e3b60d192758ea6c4d8b6853efed43e86075b96850d010116",
     "plugins/workflow/node_modules/yaml/browser/index.js": "5ae196b7abe55bed1d4d11f7eeba63aeda0c49fb73eb3c7359e3d6b9a080696b",
@@ -1114,13 +1104,6 @@
     "plugins/scheduler/node_modules/zod/v4/mini/schemas.d.cts": "1a8564333cf4f7ab1940d80791b75d862c05481d5d2b709966fa53ddf8ca7344",
     "plugins/scheduler/node_modules/zod/v4/mini/schemas.d.ts": "f67a25ded944912bd4621b68df17c7c70a52aae117278c076c0aabbc32a3abba",
     "plugins/scheduler/node_modules/zod/v4/mini/schemas.js": "51b3a7869df2d3a1bd98acb9d5b6a61bbccc7653d7a3d8fca7f0e0db7e84e284",
-    "plugins/session-title/node_modules/@arcships/rutis/src/advanced.d.ts": "9b75feb7a68b662561ae05f907680c8f2b6f691c7d0358525411767be5bed546",
-    "plugins/session-title/node_modules/@arcships/rutis/src/advanced.mjs": "d498fca3d835840657f9bce22cd94631e68185b5c828b94b1f1cb358eac98b59",
-    "plugins/session-title/node_modules/@arcships/rutis/src/index.d.ts": "58f83cebc736219cc3da67afd4e05760a9555524cd94c364d831b540bf0b3892",
-    "plugins/session-title/node_modules/@arcships/rutis/src/index.mjs": "3b0ee7d43ec527e6da8ea0c09b3e0361c51e1bbd4c8e141bde1e59cf38c43101",
-    "plugins/session-title/node_modules/@arcships/rutis/src/testing.d.ts": "c5ca82ed4e39a9f8a6c774e0a0bb6e70daa0ba7397e0748648770881b1032413",
-    "plugins/session-title/node_modules/@arcships/rutis/src/testing.mjs": "6afe5359dc1a28734aab932a98eb449b0b1af8974205e42917b9c0bc9f1ba6ef",
-    "plugins/session-title/node_modules/@arcships/rutis/src/values.mjs": "d668102fd296e2021cc735177bc6d377257e2edfcaa6bbd04da55f1783006f8d",
     "plugins/workflow/node_modules/@arcships/rutis/src/advanced.d.ts": "9b75feb7a68b662561ae05f907680c8f2b6f691c7d0358525411767be5bed546",
     "plugins/workflow/node_modules/@arcships/rutis/src/advanced.mjs": "d498fca3d835840657f9bce22cd94631e68185b5c828b94b1f1cb358eac98b59",
     "plugins/workflow/node_modules/@arcships/rutis/src/index.d.ts": "58f83cebc736219cc3da67afd4e05760a9555524cd94c364d831b540bf0b3892",

package/plugins/session-title/README.md

@@ -1,76 +0,0 @@
-# SessionTitle
-
-Optional Session TypeScript package implementing the [SessionTitle contract](../../../docs/design/plugins/SessionTitle.md).
-Its one row registers the Session part (the `part` export) with `partRegistry`, and each Session
-mounts a copy. The copy registers `session.rename` with the Session's `sessionMethods` and
-reads its Session through `sessionIdentity`, `sessionConfig` and `sessionLifecycle`; the existing
-process services own history, settings, model execution and title persistence.
-
-## Loading
-
-Install the shared runtime and this package's pinned SDK dependencies:
-
-```sh
-npm ci --prefix runtime --ignore-scripts
-npm ci --prefix runtime/plugins/session-title --ignore-scripts
-(cd runtime && npm install --no-save --package-lock=false --ignore-scripts --install-links ./plugins/session-title)
-```
-
-These commands set up the shared runtime and this package in the checkout for `--session-rows` embedding and the package's tests. `./dev` needs none of them: it stages the default plugins, with their own dependencies, into `staging/dev-plugins`.
-
-The plugin directory carries [dim-plugin.json](dim-plugin.json) and is a default plugin listed in
-[runtime/default-plugins.json](../../default-plugins.json), which PluginManager loads (`--defaults`).
-An embedding host may still build a row from the manifest's `code` fields (`entry`, `inject`); the
-module imports `inject` from the same manifest and provides nothing.
-The `dim-app-server` binary takes such rows with `--session-rows FILE` (a JSON array whose
-file location is the anchor); see the [App Server README](../../../crates/dim-app-server/README.md#session-typescript-embedding).
-
-The Session part's required named dependencies are `sessions`, `settings`, `sessionMetadata` and
-`auxiliaryModel`. An unavailable provider keeps the part Pending, including rename. Native Sessions
-continue under existing assembly rules. Events missed while the part is unavailable are not replayed;
-the next adopted operator input can trigger an attempt using the first adopted input in history.
-
-## Behavior
-
-Each Session's copy listens to its Session's `session/{sessionId}` channel. An interactive leader's own
-`main` adoption schedules owned asynchronous work and returns from the listener. A fixed published history prefix supplies only the first adopted
-operator input's text, bounded to 2000 Unicode code points. Images, raw blocks, deleted inputs,
-later input text and tool/assistant/system content are excluded. An image-only first input
-produces no automatic title and is remembered for this load.
-
-Each attempt first reads the validated user-only `session-title` Settings document; disabled
-generation reads neither metadata nor history. Defaults are
-`enabled: true`, `maxLength: 80` (10–200 code points) and `model: null`. The native backend
-stores it in `user/session-title`; the file backend uses the user file `session-title.json`.
-Use existing `settings.read`/`settings.update` with the returned revision. Project scope is
-rejected. Business edits affect the next attempt or rename without reloading the row.
-
-With `model: null`, the latest adopted `context/configured` in the fixed prefix supplies the
-model. A prefix without one fails the attempt; creation configuration and Provider defaults
-are not substitutes. The request omits `maxOutputTokens` and uses S4's existing model limit.
-Generation has a 30-second budget and three failed attempts per loaded generation. The package
-emits outcome/duration diagnostics to stderr. The supplied C4 host currently discards runtime
-stderr; it does not expose these diagnostics to clients.
-
-`session.rename({sessionId, title})` trims and validates a manual title; `null` clears it.
-Manual conflicts allow three retries. An automatic result uses the revision read before
-generation, so a concurrent manual write wins. Existing SessionStore notices and list fields
-announce committed titles. Titles add no Timeline facts or main-model context.
-
-Closing cancels and joins the original auxiliary operation through S4's existing reference.
-After `setTitle` handover, SessionStore settles the mutation; a lost reply permits no repeat
-or rollback. Generation takes no Session activity hold. Clearing permits a new attempt on
-the next adopted operator input, still using the original first input.
-
-## Verification
-
-```sh
-node runtime/node_modules/typescript/bin/tsc -p runtime/plugins/session-title/tsconfig.json
-node --test runtime/plugins/session-title/session-title.test.mjs
-```
-
-The SDK flows execute the actual plugin over its declared ports. The real Node/App Server flow is
-`node_session_title_generation_rename_settings_restore_fork_and_close` in
-[session_title.rs](../../../crates/dim-app-server/tests/transport_flow/session_title.rs), run with
-the same `DIM_TEST_NODE`, `DIM_TEST_JS_RUNTIME` and `DIM_TEST_RUTIS_SDK` settings as public Node CI.
-It uses disposable stores and existing Prompt adoption, Provider, Settings, S2 and metadata bindings.

package/plugins/session-title/dim-plugin.json

@@ -1,11 +0,0 @@
-{
-  "name": "session-title",
-  "description": "Titles Sessions from their first exchange",
-  "code": {
-    "entry": "index.ts",
-    "inject": [
-      "partRegistry",
-      "settings"
-    ]
-  }
-}

package/plugins/session-title/index.ts

@@ -1,13 +0,0 @@
-import { definePlugin } from '@arcships/rutis'
-import type { PartRegistry, Settings } from '../../../npm/plugin.d.ts'
-import manifest from './dim-plugin.json' with { type: 'json' }
-import part from './part.ts'
-import { settings as declaration } from './settings.ts'
-
-export default definePlugin({
-  name: 'SessionTitle', inject: manifest.code.inject,
-  async apply(ctx) {
-    await ctx.use<Settings>('settings').register(declaration)
-    await ctx.use<PartRegistry>('partRegistry').registerDefinition('session', 'SessionTitle', part)
-  },
-})

package/plugins/session-title/node_modules/.package-lock.json

@@ -1,18 +0,0 @@
-{
-  "name": "@dimcode/session-title",
-  "version": "0.1.0",
-  "lockfileVersion": 3,
-  "requires": true,
-  "packages": {
-    "node_modules/@arcships/rutis": {
-      "version": "0.19.3",
-      "resolved": "https://registry.npmjs.org/@arcships/rutis/-/rutis-0.19.3.tgz",
-      "integrity": "sha512-mpeVQ2tIZFyiKibgmRtyEhFu0NW3xam/OMUWSkddXIj8a/QqyW3AiE37GKlb4ihpHtuRx1dIz9HyCYHgvnecJw==",
-      "license": "MIT",
-      "engines": {
-        "bun": ">=1.4",
-        "node": ">=22"
-      }
-    }
-  }
-}

package/plugins/session-title/node_modules/@arcships/rutis/README.md

@@ -1,32 +0,0 @@
-# @arcships/rutis
-
-Write [rutis](https://github.com/arcships/rutis) plugins in TypeScript or JavaScript.
-
-```ts
-import { definePlugin } from '@arcships/rutis'
-
-export default definePlugin({
-  inject: ['llm'],                              // services it uses
-  config: { type: 'object', properties: { city: { type: 'string' } } },
-  apply(ctx, config) {
-    ctx.provide('weather', new Weather(ctx.use('llm'), config.city))
-    return () => {}                             // cleanup
-  },
-})
-```
-
-What `ctx.provide` provides reaches rutis by itself, its methods the object's functions, called as `async`. `provides` is optional: `provides: { weather: { unit: 'sync' } }` marks methods `sync` and tells `rutis-host check` what the plugin provides before it runs.
-
-Test it on a real host (`rutis-host test`; `@arcships/rutis-runtime` and `@arcships/rutis-host` as dev dependencies, which `rutis-host new` adds); the plugin is the default export of its module:
-
-```ts
-import { load } from '@arcships/rutis/testing'
-
-const t = await load(plugin, { config: { city: 'Oslo' }, services: { llm: { ask: async () => 'sunny' } } })
-assert.equal(await t.service('weather').today(), 'sunny in Oslo')
-await t.unload()
-```
-
-This package has no dependencies. Events (`ctx.on`, and `ctx.emit` / `parallel` / `serial` / `waterfall` / `bail` / `filter`) are one bus across every plugin, runtime and linked node, dispatched by the host; in a test too (`host()` puts several plugins on one bus). The host installs the runtime that runs the plugin (`@arcships/rutis-runtime` for Node, `@arcships/rutis-bun` for Bun). Start a project with `npx @arcships/rutis-host new <name> --lang node` (or `--lang bun`).
-
-Guide (Chinese): [docs/guide/typescript-plugin.md](https://github.com/arcships/rutis/blob/main/docs/guide/typescript-plugin.md).