From 8a917f8e1e6615259e3e8f21cda082b90155b118 Mon Sep 17 00:00:00 2001
From: Daniel Saldarriaga <dsaldarriaga@prevalentware.com>
Date: Thu, 20 Aug 2026 20:21:38 +0200
Subject: [PATCH] fix: keep state mutations on one file

---
 CONTRIBUTING.md        |   4 +-
 dist/server.js         |  27 +++----
 package.json           |   4 +-
 src/state.ts           |  19 +++--
 test/server-v2.test.ts |  76 +++++++++++-------
 test/server.test.ts    | 172 ++++++++++++++++++++++-------------------
 test/state.test.ts     |  18 +++++
 7 files changed, 182 insertions(+), 138 deletions(-)

diff --git a/CONTRIBUTING.md b/CONTRIBUTING.md
index 89d8941..c709ede 100644
--- a/CONTRIBUTING.md
+++ b/CONTRIBUTING.md
@@ -19,8 +19,8 @@ Useful scripts:
 
 | Script | What it does |
 | --- | --- |
-| `bun run test` | Run the unit test suite serially |
-| `bun run test:coverage` | Run the serial test suite with a coverage report |
+| `bun run test` | Run the unit test suite |
+| `bun run test:coverage` | Run the test suite with a coverage report |
 | `bun run lint` | ESLint over the repo |
 | `bun run typecheck` | TypeScript `--noEmit` check |
 | `bun run build` | Bundle `src/server.ts` into `dist/` |
diff --git a/dist/server.js b/dist/server.js
index f70626d..2fbca64 100644
--- a/dist/server.js
+++ b/dist/server.js
@@ -242,8 +242,7 @@ function parseStateText(raw, file) {
 function decodeState(value) {
   return Schema.decodeUnknown(StateSchema)(value).pipe(Effect.map(mutableState), Effect.map(normalizeState), Effect.mapError((cause) => new StateDecodeError({ cause })));
 }
-function readStateEffect() {
-  const file = statePath();
+function readStateEffect(file = statePath()) {
   return Effect.tryPromise({
     try: () => readFile(file, "utf8"),
     catch: (cause) => new StateReadError({ cause })
@@ -252,10 +251,9 @@ function readStateEffect() {
     catch: (cause) => new StateDecodeError({ cause })
   })), Effect.flatMap(decodeState), Effect.catchAll((error) => error._tag === "StateReadError" && isMissingStateFile(error.cause) ? Effect.succeed(emptyState()) : Effect.fail(error)));
 }
-function writeStateEffect(state) {
+function writeStateEffect(state, file = statePath()) {
   return Effect.tryPromise({
     try: async () => {
-      const file = statePath();
       await mkdir(dirname2(file), { recursive: true, mode: 448 });
       await atomicWriteFile(file, JSON.stringify(state, null, 2) + `
 `);
@@ -277,15 +275,18 @@ function enqueueMutation(operation) {
   return current;
 }
 async function mutate(fn) {
-  return enqueueMutation(() => Effect.runPromise(Effect.gen(function* () {
-    const state = yield* readStateEffect();
-    const result = yield* Effect.tryPromise({
-      try: () => Promise.resolve(fn(state)),
-      catch: (cause) => cause instanceof Error ? cause : new Error(String(cause))
-    });
-    yield* writeStateEffect(state);
-    return result;
-  })));
+  return enqueueMutation(() => {
+    const file = statePath();
+    return Effect.runPromise(Effect.gen(function* () {
+      const state = yield* readStateEffect(file);
+      const result = yield* Effect.tryPromise({
+        try: () => Promise.resolve(fn(state)),
+        catch: (cause) => cause instanceof Error ? cause : new Error(String(cause))
+      });
+      yield* writeStateEffect(state, file);
+      return result;
+    }));
+  });
 }
 function validateObjective(objective) {
   const value = objective.trim();
diff --git a/package.json b/package.json
index 0cc5913..f515ffa 100644
--- a/package.json
+++ b/package.json
@@ -48,8 +48,8 @@
     "ci:version": "bun scripts/resolve-ci-version.ts",
     "lint": "eslint .",
     "pack:dry-run": "npm pack --dry-run",
-    "test": "bun test --concurrent --max-concurrency 1",
-    "test:coverage": "bun test --concurrent --max-concurrency 1 --coverage",
+    "test": "bun test",
+    "test:coverage": "bun test --coverage",
     "typecheck": "tsc --noEmit",
     "prepublishOnly": "bun run test && bun run build"
   },
diff --git a/src/state.ts b/src/state.ts
index bb810ec..63b68c2 100644
--- a/src/state.ts
+++ b/src/state.ts
@@ -320,8 +320,7 @@ function decodeState(value: unknown) {
   )
 }
 
-function readStateEffect() {
-  const file = statePath()
+function readStateEffect(file = statePath()) {
   return Effect.tryPromise({
     try: () => readFile(file, "utf8"),
     catch: (cause) => new StateReadError({ cause }),
@@ -339,10 +338,9 @@ function readStateEffect() {
   )
 }
 
-function writeStateEffect(state: State) {
+function writeStateEffect(state: State, file = statePath()) {
   return Effect.tryPromise({
     try: async () => {
-      const file = statePath()
       await mkdir(dirname(file), { recursive: true, mode: 0o700 })
       // atomicWriteFile writes to a same-directory temp file, fsyncs it, then
       // renames it into place: the final path is only ever replaced by a
@@ -387,19 +385,20 @@ function enqueueMutation<T>(operation: () => Promise<T>) {
 }
 
 async function mutate<T>(fn: (state: State) => T | Promise<T>) {
-  return enqueueMutation(() =>
-    Effect.runPromise(
+  return enqueueMutation(() => {
+    const file = statePath()
+    return Effect.runPromise(
       Effect.gen(function* () {
-        const state = yield* readStateEffect()
+        const state = yield* readStateEffect(file)
         const result = yield* Effect.tryPromise({
           try: () => Promise.resolve(fn(state)),
           catch: (cause) => (cause instanceof Error ? cause : new Error(String(cause))),
         })
-        yield* writeStateEffect(state)
+        yield* writeStateEffect(state, file)
         return result
       }),
-    ),
-  )
+    )
+  })
 }
 
 export function validateObjective(objective: string) {
diff --git a/test/server-v2.test.ts b/test/server-v2.test.ts
index 0151ff2..7103ed2 100644
--- a/test/server-v2.test.ts
+++ b/test/server-v2.test.ts
@@ -180,6 +180,21 @@ async function createGoalViaV2Tool(mock: MockContext, objective: string, agent =
 }
 
 let dir = ""
+const setupDisposers: Array<() => void | Promise<void>> = []
+
+async function setupPlugin(...args: Parameters<typeof plugin.setup>) {
+  const cleanup = await plugin.setup(...args)
+  let disposed = false
+  const dispose = async () => {
+    if (disposed) return
+    disposed = true
+    const index = setupDisposers.indexOf(dispose)
+    if (index >= 0) setupDisposers.splice(index, 1)
+    await cleanup()
+  }
+  setupDisposers.push(dispose)
+  return dispose
+}
 
 beforeEach(async () => {
   dir = await mkdtemp(join(tmpdir(), "opencode-goal-plugin-v2-"))
@@ -187,6 +202,7 @@ beforeEach(async () => {
 })
 
 afterEach(async () => {
+  for (const dispose of setupDisposers.splice(0).reverse()) await dispose()
   delete process.env.OPENCODE_GOAL_STATE_PATH
   await rm(dir, { recursive: true, force: true })
 })
@@ -199,7 +215,7 @@ test("default export exposes both V1 server and V2 setup", () => {
 
 test("V2 setup registers goal tools with JSON Schema inputs, codemode:false, and {content} executors", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
 
   expect(mock.tools.map((tool) => tool.name).sort()).toEqual(TOOL_NAMES)
 
@@ -230,7 +246,7 @@ test("V2 setup registers goal tools with JSON Schema inputs, codemode:false, and
 
 test("V2 list_all_goals returns goals from other sessions", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await goalTool(mock, "create_goal").execute(
     { objective: "first V2 session goal" },
     toolContext("ses_first"),
@@ -253,7 +269,7 @@ test("V2 list_all_goals returns goals from other sessions", async () => {
 test("V2 create_goal recovers from a zero-filled state file", async () => {
   await writeFile(process.env.OPENCODE_GOAL_STATE_PATH!, "\u0000\u0000", "utf8")
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
 
   const created = await createGoalViaV2Tool(mock, "recover V2 state")
 
@@ -265,7 +281,7 @@ test("V2 create_goal recovers from a zero-filled state file", async () => {
 
 test("V2 create_goal reuses the same active objective without reinitializing state", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await goalTool(mock, "create_goal").execute(
     { objective: "finish V2 safely", token_budget: 100 },
     toolContext(),
@@ -292,7 +308,7 @@ test("V2 create_goal reuses the same active objective without reinitializing sta
 
 test("V2 setup registers the /goal command via command transform", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
 
   const command = mock.commandDraft.get("goal")
   expect(command).toBeDefined()
@@ -307,7 +323,7 @@ test("V2 setup registers the /goal command via command transform", async () => {
 
 test("V2 setup skips command registration when register_command is false", async () => {
   const mock = makeMockContext({ auto_continue: false, register_command: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
 
   expect(mock.commandDraft.get("goal")).toBeUndefined()
   expect(mock.disposals).not.toContain("command.transform")
@@ -318,7 +334,7 @@ test("V2 setup skips command registration when register_command is false", async
 
 test("V2 session context hook injects the goal-mode system reminder", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
 
   const contextHook = mock.hooks["context"]!
   expect(contextHook).toBeTypeOf("function")
@@ -336,7 +352,7 @@ test("V2 session context hook injects the goal-mode system reminder", async () =
 
 test("V2 setup registers tool execute hooks", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
 
   expect(mock.hooks["execute.before"]).toBeTypeOf("function")
   expect(mock.hooks["execute.after"]).toBeTypeOf("function")
@@ -355,7 +371,7 @@ test("V2 setup registers tool execute hooks", async () => {
 
 test("V2 events account usage and checkpoints from step/usage events", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "account usage from events")
 
   // Step events drive per-step token sums and checkpoints.
@@ -426,7 +442,7 @@ test("V2 events account usage and checkpoints from step/usage events", async ()
 
 test("V2 step accounting excludes steps observed before goal creation", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
 
   mock.stream.push({
     type: "session.step.ended",
@@ -462,7 +478,7 @@ test("V2 step accounting excludes steps observed before goal creation", async ()
 
 test("V2 step and session sources do not double-count when session usage arrives first", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "reconcile usage sources")
 
   mock.stream.push({
@@ -494,7 +510,7 @@ test("V2 step and session sources do not double-count when session usage arrives
 
 test("V2 failed steps account usage and replace stale assistant progress", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "survive a failed model step")
 
   mock.stream.push({
@@ -529,7 +545,7 @@ test("V2 failed steps account usage and replace stale assistant progress", async
 
 test("V2 idle event triggers auto-continue via ctx.session.prompt", async () => {
   const mock = makeMockContext({ auto_continue: true, min_continue_interval_seconds: 0, max_auto_turns: 5 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "auto-continue from idle events")
 
   mock.stream.push({ type: "session.idle", created: Date.now(), data: { sessionID: "ses_v2" } })
@@ -548,7 +564,7 @@ test("V2 idle event triggers auto-continue via ctx.session.prompt", async () =>
 
 test("V2 idle continuation waits for a running child session", async () => {
   const mock = makeMockContext({ min_continue_interval_seconds: 1 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "wait for delegated work")
 
   mock.stream.push({ type: "session.created", created: 100, data: { sessionID: "child", parentID: "ses_v2" } })
@@ -566,7 +582,7 @@ test("V2 idle continuation waits for a running child session", async () => {
 
 test("V2 idle auto-continue is suppressed for plan-agent goals", async () => {
   const mock = makeMockContext({ auto_continue: true, min_continue_interval_seconds: 0, max_auto_turns: 5 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "plan-mode goal must stay paused", "plan")
 
   mock.stream.push({ type: "session.idle", created: Date.now(), data: { sessionID: "ses_v2" } })
@@ -582,7 +598,7 @@ test("V2 idle auto-continue is suppressed for plan-agent goals", async () => {
 
 test("V2 cleanup disposes registrations and stops the event consumer", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "cleanup lifecycle")
 
   mock.stream.end()
@@ -604,7 +620,7 @@ test("V2 cleanup disposes registrations and stops the event consumer", async ()
 
 test("V2 session.error schedules bounded recovery without a phantom failure", async () => {
   const mock = makeMockContext({ auto_continue: true, min_continue_interval_seconds: 0, max_auto_turns: 5 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "recover from a transport error")
 
   mock.stream.push({
@@ -630,7 +646,7 @@ test("V2 idle after a started pending attempt counts one unresolved failure and
     max_auto_turns: 5,
     max_prompt_failures: 1,
   })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "detect a no response")
 
   mock.stream.push({ type: "session.idle", created: 1, data: { sessionID: "ses_v2" } })
@@ -659,7 +675,7 @@ test("V2 persists the attempt before the prompt resolves so a later busy can cor
       resolvePrompt = resolve
     })
   }
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "correlate a racing busy")
 
   void mock.stream.push({ type: "session.idle", created: 1, data: { sessionID: "ses_v2" } })
@@ -686,7 +702,7 @@ test("V2 persists the attempt before the prompt resolves so a later busy can cor
 
 test("V2 retry status cancels scheduled transport recovery", async () => {
   const mock = makeMockContext({ auto_continue: true, min_continue_interval_seconds: 0 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "let the native retry win")
 
   mock.stream.push({
@@ -706,7 +722,7 @@ test("V2 retry status cancels scheduled transport recovery", async () => {
 
 test("V2 successful tool progress cancels no-pending transport recovery", async () => {
   const mock = makeMockContext({ auto_continue: true, min_continue_interval_seconds: 0 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "cancel recovery via tool progress")
 
   mock.stream.push({
@@ -755,7 +771,7 @@ test("V2 successful tool progress cancels no-pending transport recovery", async
 
 test("V2 assistant progress cancels no-pending transport recovery", async () => {
   const mock = makeMockContext({ auto_continue: true, min_continue_interval_seconds: 0 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "cancel recovery via assistant progress")
 
   mock.stream.push({
@@ -789,7 +805,7 @@ test("V2 assistant progress cancels no-pending transport recovery", async () =>
 
 test("V2 watchdog rescues a busy active goal without consuming auto-turn budgets", async () => {
   const mock = makeMockContext({ auto_continue: false, max_turn_time: 0.02, max_prompt_failures: 5, max_auto_turns: 1 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "watchdog should not eat the budget")
 
   mock.stream.push({ type: "session.status", created: Date.now(), data: { sessionID: "ses_v2", status: { type: "busy" } } })
@@ -807,7 +823,7 @@ test("V2 non-transport prompt errors do not count toward the ceiling or retry",
   mock.session.prompt = async () => {
     throw new Error("invalid provider configuration")
   }
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "non-transport must be ignored")
 
   mock.stream.push({ type: "session.idle", created: Date.now(), data: { sessionID: "ses_v2" } })
@@ -823,7 +839,7 @@ test("V2 non-transport prompt errors do not count toward the ceiling or retry",
 
 test("V2 a native retry status suppresses a later session.error until busy ends the episode", async () => {
   const mock = makeMockContext({ auto_continue: true, min_continue_interval_seconds: 0, max_prompt_failures: 3 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "native retry must win")
 
   // retry arrives before the transport error; the error is suppressed while
@@ -857,7 +873,7 @@ test("V2 a native retry status suppresses a later session.error until busy ends
 
 test("V2 watchdog no-response counts a failure on idle even with auto_continue false", async () => {
   const mock = makeMockContext({ auto_continue: false, max_turn_time: 0.02, max_prompt_failures: 5, max_auto_turns: 1 })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "watchdog no response with auto-continue disabled")
 
   mock.stream.push({ type: "session.status", created: Date.now(), data: { sessionID: "ses_v2", status: { type: "busy" } } })
@@ -885,7 +901,7 @@ test("V2 watchdog no-response counts a failure on idle even with auto_continue f
 
 test("V2 delayed tool output from a prior turn cannot clear a newer pending attempt", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "correlate tool progress to the attempt")
 
   // The tool call starts while attempt A is pending; the before hook captures
@@ -937,7 +953,7 @@ test("V2 dispose during an in-flight prompt rolls back the reserved attempt on r
     })
     throw new Error("network down")
   }
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "dispose mid-flight rejection")
 
   void mock.stream.push({ type: "session.idle", created: 1, data: { sessionID: "ses_v2" } })
@@ -967,7 +983,7 @@ test("V2 commits an accepted prompt when its recovery timer is canceled in fligh
       resolvePrompt = resolve
     })
   }
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "commit accepted recovery")
 
   mock.stream.push({
@@ -999,7 +1015,7 @@ test("V2 commits an accepted prompt when its recovery timer is canceled in fligh
 
 test("V2 completed tool failures do not clear retry state", async () => {
   const mock = makeMockContext({ auto_continue: false })
-  const cleanup = await plugin.setup(mock as never)
+  const cleanup = await setupPlugin(mock as never)
   await createGoalViaV2Tool(mock, "keep failed tools from masking recovery")
 
   await reserveContinuation("ses_v2", 10, 0)
diff --git a/test/server.test.ts b/test/server.test.ts
index 791d0ec..8e89f8d 100644
--- a/test/server.test.ts
+++ b/test/server.test.ts
@@ -40,6 +40,15 @@ async function waitForContinuation(calls: unknown[]) {
 }
 
 let dir = ""
+const serverDisposers: Array<() => Promise<void>> = []
+
+async function setupServer(...args: Parameters<typeof plugin.server>) {
+  const hooks = await plugin.server(...args)
+  serverDisposers.push(async () => {
+    await hooks.dispose?.()
+  })
+  return hooks
+}
 
 beforeEach(async () => {
   dir = await mkdtemp(join(tmpdir(), "opencode-goal-plugin-"))
@@ -47,13 +56,14 @@ beforeEach(async () => {
 })
 
 afterEach(async () => {
+  for (const dispose of serverDisposers.splice(0).reverse()) await dispose()
   delete process.env.OPENCODE_GOAL_STATE_PATH
   await rm(dir, { recursive: true, force: true })
 })
 
 test("server plugin exposes Codex-style goal tools", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -99,7 +109,7 @@ test("server plugin exposes Codex-style goal tools", async () => {
 })
 
 test("list_all_goals returns goals from other sessions", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     { client: { session: { promptAsync: async () => {} } } } as never,
     { auto_continue: false },
   )
@@ -125,7 +135,7 @@ test("list_all_goals returns goals from other sessions", async () => {
 })
 
 test("set goal lets the agent formulate the goal objective", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -148,7 +158,7 @@ test("set goal lets the agent formulate the goal objective", async () => {
 })
 
 test("create_goal reuses the same active objective without mutating state", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     { client: { session: { promptAsync: async () => {} } } } as never,
     { auto_continue: false },
   )
@@ -182,7 +192,7 @@ test("create_goal reuses the same active objective without mutating state", asyn
 })
 
 test("create_goal starts a fresh goal when the matching prior goal is closed", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     { client: { session: { promptAsync: async () => {} } } } as never,
     { auto_continue: false },
   )
@@ -202,7 +212,7 @@ test("create_goal starts a fresh goal when the matching prior goal is closed", a
 })
 
 test("concurrent matching create_goal calls converge on one goal", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     { client: { session: { promptAsync: async () => {} } } } as never,
     { auto_continue: false },
   )
@@ -219,7 +229,7 @@ test("concurrent matching create_goal calls converge on one goal", async () => {
 })
 
 test("duplicate limited goals retain the safety stop notice", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     { client: { session: { promptAsync: async () => {} } } } as never,
     { auto_continue: false },
   )
@@ -235,7 +245,7 @@ test("duplicate limited goals retain the safety stop notice", async () => {
 })
 
 test("server plugin registers goal as a desktop/web command by default", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -266,7 +276,7 @@ test("server plugin registers goal as a desktop/web command by default", async (
 
 test("system transform is byte-stable across the complete goal lifecycle", async () => {
   setSystemTime(new Date(100_000))
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -451,7 +461,7 @@ OpenCode goal mode policy:
 })
 
 test("compaction autocontinue is disabled while a goal is active", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -472,7 +482,7 @@ test("compaction autocontinue is disabled while a goal is active", async () => {
 })
 
 test("goal objective can be edited and history can be reported", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -500,7 +510,7 @@ test("goal objective can be edited and history can be reported", async () => {
 })
 
 test("goal status tool pauses and resumes a goal", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -525,7 +535,7 @@ test("goal status tool pauses and resumes a goal", async () => {
 })
 
 test("server plugin does not overwrite an existing goal command", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -551,7 +561,7 @@ test("server plugin does not overwrite an existing goal command", async () => {
 })
 
 test("server plugin can disable desktop/web command registration", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -571,7 +581,7 @@ test("server plugin can disable desktop/web command registration", async () => {
 })
 
 test("update goal can close as unmet with a blocker", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -597,7 +607,7 @@ test("update goal can close as unmet with a blocker", async () => {
 })
 
 test("message transform prefers exact step token usage", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -640,7 +650,7 @@ test("message transform prefers exact step token usage", async () => {
 })
 
 test("message transform excludes session usage observed before goal work", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     { client: { session: { promptAsync: async () => {} } } } as never,
     { auto_continue: false },
   )
@@ -675,7 +685,7 @@ test("message transform excludes session usage observed before goal work", async
 
 test("per-prompt chat hook recovers from an empty state file", async () => {
   await writeFile(process.env.OPENCODE_GOAL_STATE_PATH!, "", "utf8")
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -692,7 +702,7 @@ test("per-prompt chat hook recovers from an empty state file", async () => {
 })
 
 test("message transform records assistant checkpoints", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -725,7 +735,7 @@ test("message transform records assistant checkpoints", async () => {
 
 test("compaction hook preserves active goal context", async () => {
   setSystemTime(new Date(100_000))
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -775,7 +785,7 @@ Preserve the goal objective, status, elapsed time, budget usage, latest checkpoi
 
 test("idle event auto-continues active goals when enabled", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -799,7 +809,7 @@ test("idle event auto-continues active goals when enabled", async () => {
 
 test("session status idle event auto-continues active goals", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -822,7 +832,7 @@ test("session status idle event auto-continues active goals", async () => {
 
 test("turn watchdog retries a busy active goal without consuming continuation budgets", async () => {
   const calls: { body?: { agent?: string; parts?: { text?: string }[] } }[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -879,7 +889,7 @@ test("turn watchdog retries a busy active goal without consuming continuation bu
 
 test("turn watchdog resets when another busy turn starts", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -910,7 +920,7 @@ test("turn watchdog resets when another busy turn starts", async () => {
 
 test("turn watchdog cancels on idle, retry, deletion, and dispose", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -961,7 +971,7 @@ test("turn watchdog cancels on idle, retry, deletion, and dispose", async () =>
 
 test("turn watchdog does not inject while tasks are active, the goal is paused, or the turn is restricted", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1027,7 +1037,7 @@ test("turn watchdog does not inject while tasks are active, the goal is paused,
 
 test("turn watchdog transport failures share the prompt-failure ceiling without charging auto-turns", async () => {
   const logs: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         app: { log: async (input: unknown) => logs.push(input) },
@@ -1080,7 +1090,7 @@ test("turn watchdog transport failures share the prompt-failure ceiling without
 
 test("running task defers idle auto-continue", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1111,7 +1121,7 @@ test("running task defers idle auto-continue", async () => {
 
 test("running task deferral does not record repeated assistant messages as no-progress", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1155,7 +1165,7 @@ test("running task deferral does not record repeated assistant messages as no-pr
 })
 
 test("low-output tool-call messages do not pause an active goal without continuations", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1206,7 +1216,7 @@ test("auto-continue pauses only after a low-progress continuation turn", async (
       { type: "step-finish", tokens: { input: 10, output: 200 } },
     ],
   }
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1255,7 +1265,7 @@ test("auto-continue pauses only after a low-progress continuation turn", async (
 
 test("terminal task waits for orchestrator assistant turn before goal continuation", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1300,7 +1310,7 @@ test("terminal task waits for orchestrator assistant turn before goal continuati
 
 test("terminal-only task output defers until orchestrator reconciles it", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1353,7 +1363,7 @@ test("terminal-only task output defers until orchestrator reconciles it", async
 
 test("synthetic terminal task message defers until orchestrator reconciles it", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1391,7 +1401,7 @@ test("synthetic terminal task message defers until orchestrator reconciles it",
 
 test("live child session status blocks goal continuation when task launch was missed", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1416,7 +1426,7 @@ test("live child session status blocks goal continuation when task launch was mi
 
 test("idle live child session uses bounded deferral when task launch was missed", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1443,7 +1453,7 @@ test("idle live child session uses bounded deferral when task launch was missed"
 
 test("idle live child bounded retry does not inject while parent session is busy", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1479,7 +1489,7 @@ test("idle live child bounded retry does not inject while parent session is busy
 test("tracked running child absent from live children stops blocking after grace period", async () => {
   const calls: unknown[] = []
   let children = [{ id: "task_1" }]
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1510,7 +1520,7 @@ test("tracked running child absent from live children stops blocking after grace
 
 test("task deferral can be disabled with config", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1537,7 +1547,7 @@ test("task deferral can be disabled with config", async () => {
 
 test("auto-continue failures pause after configured retry limit", async () => {
   const logs: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         app: {
@@ -1566,7 +1576,7 @@ test("auto-continue failures pause after configured retry limit", async () => {
 
 test("set_goal from the plan agent records a paused goal instead of an active one", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1596,7 +1606,7 @@ test("set_goal from the plan agent records a paused goal instead of an active on
 })
 
 test("create_goal from the plan agent records a paused goal", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1626,7 +1636,7 @@ test("create_goal from the plan agent records a paused goal", async () => {
 })
 
 test("plan-created goal cannot resume from plan but resumes from build", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1659,7 +1669,7 @@ test("plan-created goal cannot resume from plan but resumes from build", async (
 })
 
 test("update_goal_objective cannot activate a goal from the plan agent", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1689,7 +1699,7 @@ test("update_goal_objective cannot activate a goal from the plan agent", async (
 
 test("idle continuation is blocked when the latest assistant turn ran under plan", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1726,7 +1736,7 @@ test("idle continuation is blocked when the latest assistant turn ran under plan
 
 test("build resume of a plan-created goal restores auto-continue pinned to build", async () => {
   const calls: { body?: { agent?: string } }[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1760,7 +1770,7 @@ test("build resume of a plan-created goal restores auto-continue pinned to build
 
 test("idle continuation is suppressed and pauses the goal after a plan-mode prompt", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1793,7 +1803,7 @@ test("idle continuation is suppressed and pauses the goal after a plan-mode prom
 
 test("auto-continue pins the continuation prompt to the recorded agent", async () => {
   const calls: { body?: { agent?: string } }[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1819,7 +1829,7 @@ test("auto-continue pins the continuation prompt to the recorded agent", async (
 })
 
 test("system reminder remains invariant after a plan-mode prompt", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1853,7 +1863,7 @@ test("system reminder remains invariant after a plan-mode prompt", async () => {
 })
 
 test("allow_goal_execution_from_plan restores active goal creation from plan", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1876,7 +1886,7 @@ test("allow_goal_execution_from_plan restores active goal creation from plan", a
 })
 
 test("restricted_agents option extends plan-mode protection to custom agents", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1901,7 +1911,7 @@ test("restricted_agents option extends plan-mode protection to custom agents", a
 test("idle handler skips overlapping continuations for the same session", async () => {
   let release: (() => void) | undefined
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -1931,7 +1941,7 @@ test("idle handler skips overlapping continuations for the same session", async
 
 test("auto-continue retries are bounded: three failed attempts, no fourth", async () => {
   const logs: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         app: { log: async (input: unknown) => logs.push(input) },
@@ -1962,7 +1972,7 @@ test("auto-continue retries are bounded: three failed attempts, no fourth", asyn
 
 test("failed continuation retries wait for the configured minimum interval", async () => {
   const logs: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         app: { log: async (input: unknown) => logs.push(input) },
@@ -2005,7 +2015,7 @@ test("recognized transport error strings accumulate as continuation failures", a
     "Provider response headers timed out after 10000ms",
   ]
   for (const [index, message] of errors.entries()) {
-    const hooks = await plugin.server(
+    const hooks = await setupServer(
       {
         client: {
           session: {
@@ -2038,7 +2048,7 @@ test("recognized transport error strings accumulate as continuation failures", a
 })
 
 test("failed tool output does not reset prompt failures; successful tool output does", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2075,7 +2085,7 @@ test("failed tool output does not reset prompt failures; successful tool output
 
 test("duplicate idle events before any busy never count a failure or send a duplicate", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2112,7 +2122,7 @@ test("duplicate idle events before any busy never count a failure or send a dupl
 
 test("paired idle events after a busy count exactly one unresolved failure and pause at the ceiling", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2156,7 +2166,7 @@ test("paired idle events after a busy count exactly one unresolved failure and p
 })
 
 test("concurrent session.error transport events count at most one failure per pending attempt", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2215,7 +2225,7 @@ test("concurrent session.error transport events count at most one failure per pe
 
 test("auto_continue false never schedules a retry after a transport event", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2252,7 +2262,7 @@ test("a repeated old assistant message cannot hide a no-response failure", async
     info: { id: "msg_old", role: "assistant", sessionID: "ses_old_message" },
     parts: [{ type: "text", text: "Earlier progress" }],
   }
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2288,7 +2298,7 @@ test("a repeated old assistant message cannot hide a no-response failure", async
 test("non-transport prompt errors do not count toward the ceiling or auto-retry", async () => {
   const logs: unknown[] = []
   let calls = 0
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         app: { log: async (input: unknown) => logs.push(input) },
@@ -2321,7 +2331,7 @@ test("non-transport prompt errors do not count toward the ceiling or auto-retry"
 
 test("session.error without a pending attempt schedules recovery without a phantom failure", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2362,7 +2372,7 @@ test("session.error without a pending attempt schedules recovery without a phant
 
 test("restart resolves a persisted started pending attempt at the next idle", async () => {
   const firstCalls: unknown[] = []
-  const hooks1 = await plugin.server(
+  const hooks1 = await setupServer(
     {
       client: {
         session: {
@@ -2389,7 +2399,7 @@ test("restart resolves a persisted started pending attempt at the next idle", as
   // A fresh instance reads the same persisted state: the started=true pending
   // attempt must be resolvable by the next idle after the restart.
   const calls2: unknown[] = []
-  const hooks2 = await plugin.server(
+  const hooks2 = await setupServer(
     {
       client: {
         session: {
@@ -2415,7 +2425,7 @@ test("restart resolves a persisted started pending attempt at the next idle", as
 })
 
 test("persisted started=false pending attempts go stale after restart", async () => {
-  const hooks1 = await plugin.server(
+  const hooks1 = await setupServer(
     {
       client: {
         session: {
@@ -2451,7 +2461,7 @@ test("persisted started=false pending attempts go stale after restart", async ()
   await hooks1.dispose?.()
 
   const calls: unknown[] = []
-  const hooks2 = await plugin.server(
+  const hooks2 = await setupServer(
     {
       client: {
         session: {
@@ -2477,7 +2487,7 @@ test("persisted started=false pending attempts go stale after restart", async ()
 
 test("a locally delivered unstarted attempt never becomes a false no-response failure", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2521,7 +2531,7 @@ test("a locally delivered unstarted attempt never becomes a false no-response fa
 
 test("a built-in retry status cancels scheduled transport recovery", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2557,7 +2567,7 @@ test("a built-in retry status cancels scheduled transport recovery", async () =>
 
 test("a native retry status suppresses a later session.error until busy or idle ends the episode", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2608,7 +2618,7 @@ test("a native retry status suppresses a later session.error until busy or idle
 
 test("an error during a native retry episode does not fail the pending attempt", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2660,7 +2670,7 @@ test("an error during a native retry episode does not fail the pending attempt",
 
 test("assistant progress cancels no-pending transport recovery", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2702,7 +2712,7 @@ test("assistant progress cancels no-pending transport recovery", async () => {
 
 test("successful tool progress cancels no-pending transport recovery", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2736,7 +2746,7 @@ test("successful tool progress cancels no-pending transport recovery", async ()
 
 test("interrupted connection messages are not classified as transport recovery", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2768,7 +2778,7 @@ test("interrupted connection messages are not classified as transport recovery",
 })
 
 test("tool progress honors completed states and never resets on failed or incomplete tools", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2826,7 +2836,7 @@ test("tool progress honors completed states and never resets on failed or incomp
 })
 
 test("delayed tool output from a prior turn cannot clear a newer pending attempt", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2880,7 +2890,7 @@ test("delayed tool output from a prior turn cannot clear a newer pending attempt
 
 test("watchdog rescues at most once per busy episode", async () => {
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2922,7 +2932,7 @@ test("watchdog rescues at most once per busy episode", async () => {
 test("a busy that races prompt resolution correlates to the persisted attempt", async () => {
   let resolvePrompt: (() => void) | undefined
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -2968,7 +2978,7 @@ test("a busy that races prompt resolution correlates to the persisted attempt",
 test("dispose prevents an in-flight continuation from scheduling retries or committing turns", async () => {
   let resolvePrompt: (() => void) | undefined
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
@@ -3007,7 +3017,7 @@ test("dispose while a prompt is in flight rolls back on rejection without a fail
   let resolvePrompt: (() => void) | undefined
   const logs: unknown[] = []
   const calls: unknown[] = []
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         app: { log: async (input: unknown) => logs.push(input) },
@@ -3049,7 +3059,7 @@ test("dispose while a prompt is in flight rolls back on rejection without a fail
 })
 
 test("the public goal tool result never exposes internal pending attempt fields", async () => {
-  const hooks = await plugin.server(
+  const hooks = await setupServer(
     {
       client: {
         session: {
diff --git a/test/state.test.ts b/test/state.test.ts
index 433b0c9..4ba8147 100644
--- a/test/state.test.ts
+++ b/test/state.test.ts
@@ -55,6 +55,24 @@ test("creates, reads, pauses, resumes, completes, and clears a goal", async () =
   expect(await getGoal("ses_1")).toBeNull()
 })
 
+test("a mutation writes back to the state path it read", async () => {
+  const firstPath = process.env.OPENCODE_GOAL_STATE_PATH!
+  const secondPath = join(dir, "other-goals.json")
+  await createGoal("ses_path", "original objective", null)
+
+  const update = updateGoalObjective("ses_path", "updated objective")
+  queueMicrotask(() => {
+    process.env.OPENCODE_GOAL_STATE_PATH = secondPath
+  })
+  await update
+
+  process.env.OPENCODE_GOAL_STATE_PATH = firstPath
+  expect((await getGoal("ses_path"))?.objective).toBe("updated objective")
+  process.env.OPENCODE_GOAL_STATE_PATH = secondPath
+  expect(await getGoal("ses_path")).toBeNull()
+  process.env.OPENCODE_GOAL_STATE_PATH = firstPath
+})
+
 test("lists public goals across sessions by most recent update", async () => {
   expect(await getAllGoals()).toEqual({ goals: [], total: 0, truncated: false })