saeeol 1.2.9 → 1.3.1

package/test/session/compaction.test.ts

@@ -0,0 +1,2420 @@
+import { afterEach, describe, expect, mock, test } from "bun:test"
+import { APICallError } from "ai"
+import { Cause, Effect, Exit, Layer, ManagedRuntime } from "effect"
+import * as Stream from "effect/Stream"
+import z from "zod"
+import { Bus } from "../../src/bus"
+import { Config } from "@/config/config"
+import { Agent } from "../../src/agent/agent"
+import { LLM } from "../../src/session/llm"
+import { SessionCompaction } from "../../src/session/compaction"
+import { Token } from "@/util/token"
+import { Instance } from "../../src/project/instance"
+import * as Log from "@saeeol/core/util/log"
+import { Permission } from "../../src/permission"
+import { Plugin } from "../../src/plugin"
+import { provideTmpdirInstance, tmpdir } from "../fixture/fixture"
+import { Session as SessionNs } from "@/session/session"
+import { MessageV2 } from "../../src/session/message-v2"
+import { MessageID, PartID, SessionID } from "../../src/session/schema"
+import { SessionStatus } from "../../src/session/status"
+import { SessionSummary } from "../../src/session/summary"
+import { ModelID, ProviderID } from "../../src/provider/schema"
+import type { Provider } from "@/provider/provider"
+import * as SessionProcessorModule from "../../src/session/processor"
+import { Snapshot } from "../../src/snapshot"
+import { ProviderTest } from "../fake/provider"
+import { testEffect } from "../lib/effect"
+import { CrossSpawnSpawner } from "@saeeol/core/cross-spawn-spawner"
+
+void Log.init({ print: false })
+
+function run<A, E>(fx: Effect.Effect<A, E, SessionNs.Service>) {
+  return Effect.runPromise(fx.pipe(Effect.provide(SessionNs.defaultLayer)))
+}
+
+const svc = {
+  ...SessionNs,
+  create(input?: SessionNs.CreateInput) {
+    return run(SessionNs.Service.use((svc) => svc.create(input)))
+  },
+  messages(input: z.output<typeof SessionNs.MessagesInput.zod>) {
+    return run(SessionNs.Service.use((svc) => svc.messages(input)))
+  },
+  updateMessage<T extends MessageV2.Info>(msg: T) {
+    return run(SessionNs.Service.use((svc) => svc.updateMessage(msg)))
+  },
+  updatePart<T extends MessageV2.Part>(part: T) {
+    return run(SessionNs.Service.use((svc) => svc.updatePart(part)))
+  },
+}
+
+const summary = Layer.succeed(
+  SessionSummary.Service,
+  SessionSummary.Service.of({
+    summarize: () => Effect.void,
+    diff: () => Effect.succeed([]),
+    computeDiff: () => Effect.succeed([]),
+  }),
+)
+
+const ref = {
+  providerID: ProviderID.make("test"),
+  modelID: ModelID.make("test-model"),
+}
+
+afterEach(() => {
+  mock.restore()
+})
+
+function createModel(opts: {
+  context: number
+  output: number
+  input?: number
+  cost?: Provider.Model["cost"]
+  npm?: string
+}): Provider.Model {
+  return {
+    id: "test-model",
+    providerID: "test",
+    name: "Test",
+    limit: {
+      context: opts.context,
+      input: opts.input,
+      output: opts.output,
+    },
+    cost: opts.cost ?? { input: 0, output: 0, cache: { read: 0, write: 0 } },
+    capabilities: {
+      toolcall: true,
+      attachment: false,
+      reasoning: false,
+      temperature: true,
+      input: { text: true, image: false, audio: false, video: false },
+      output: { text: true, image: false, audio: false, video: false },
+    },
+    api: { npm: opts.npm ?? "@ai-sdk/anthropic" },
+    options: {},
+  } as Provider.Model
+}
+
+const wide = () => ProviderTest.fake({ model: createModel({ context: 100_000, output: 32_000 }) })
+
+async function user(sessionID: SessionID, text: string) {
+  const msg = await svc.updateMessage({
+    id: MessageID.ascending(),
+    role: "user",
+    sessionID,
+    agent: "build",
+    model: ref,
+    time: { created: Date.now() },
+  })
+  await svc.updatePart({
+    id: PartID.ascending(),
+    messageID: msg.id,
+    sessionID,
+    type: "text",
+    text,
+  })
+  return msg
+}
+
+async function assistant(sessionID: SessionID, parentID: MessageID, root: string) {
+  const msg: MessageV2.Assistant = {
+    id: MessageID.ascending(),
+    role: "assistant",
+    sessionID,
+    mode: "build",
+    agent: "build",
+    path: { cwd: root, root },
+    cost: 0,
+    tokens: {
+      output: 0,
+      input: 0,
+      reasoning: 0,
+      cache: { read: 0, write: 0 },
+    },
+    modelID: ref.modelID,
+    providerID: ref.providerID,
+    parentID,
+    time: { created: Date.now() },
+    finish: "end_turn",
+  }
+  await svc.updateMessage(msg)
+  return msg
+}
+
+async function summaryAssistant(sessionID: SessionID, parentID: MessageID, root: string, text: string) {
+  const msg: MessageV2.Assistant = {
+    id: MessageID.ascending(),
+    role: "assistant",
+    sessionID,
+    mode: "compaction",
+    agent: "compaction",
+    path: { cwd: root, root },
+    cost: 0,
+    tokens: {
+      output: 0,
+      input: 0,
+      reasoning: 0,
+      cache: { read: 0, write: 0 },
+    },
+    modelID: ref.modelID,
+    providerID: ref.providerID,
+    parentID,
+    summary: true,
+    time: { created: Date.now() },
+    finish: "end_turn",
+  }
+  await svc.updateMessage(msg)
+  await svc.updatePart({
+    id: PartID.ascending(),
+    messageID: msg.id,
+    sessionID,
+    type: "text",
+    text,
+  })
+  return msg
+}
+
+async function lastCompactionPart(sessionID: SessionID) {
+  return (await svc.messages({ sessionID }))
+    .at(-2)
+    ?.parts.find((item): item is MessageV2.CompactionPart => item.type === "compaction")
+}
+
+function fake(
+  input: Parameters<SessionProcessorModule.SessionProcessor.Interface["create"]>[0],
+  result: "continue" | "compact",
+) {
+  const msg = input.assistantMessage
+  return {
+    get message() {
+      return msg
+    },
+    updateToolCall: Effect.fn("TestSessionProcessor.updateToolCall")(() => Effect.succeed(undefined)),
+    completeToolCall: Effect.fn("TestSessionProcessor.completeToolCall")(() => Effect.void),
+    process: Effect.fn("TestSessionProcessor.process")(() => Effect.succeed(result)),
+  } satisfies SessionProcessorModule.SessionProcessor.Handle
+}
+
+function layer(result: "continue" | "compact") {
+  return Layer.succeed(
+    SessionProcessorModule.SessionProcessor.Service,
+    SessionProcessorModule.SessionProcessor.Service.of({
+      create: Effect.fn("TestSessionProcessor.create")((input) => Effect.succeed(fake(input, result))),
+    }),
+  )
+}
+
+function cfg(compaction?: Config.Info["compaction"]) {
+  const base = Config.Info.zod.parse({})
+  return Layer.mock(Config.Service)({
+    get: () => Effect.succeed({ ...base, compaction }),
+  })
+}
+
+function runtime(
+  result: "continue" | "compact",
+  plugin = Plugin.defaultLayer,
+  provider = ProviderTest.fake(),
+  config = Config.defaultLayer,
+) {
+  const bus = Bus.layer
+  return ManagedRuntime.make(
+    Layer.mergeAll(SessionCompaction.layer, bus).pipe(
+      Layer.provide(provider.layer),
+      Layer.provide(SessionNs.defaultLayer),
+      Layer.provide(layer(result)),
+      Layer.provide(Agent.defaultLayer),
+      Layer.provide(plugin),
+      Layer.provide(bus),
+      Layer.provide(config),
+    ),
+  )
+}
+
+const deps = Layer.mergeAll(
+  ProviderTest.fake().layer,
+  layer("continue"),
+  Agent.defaultLayer,
+  Plugin.defaultLayer,
+  Bus.layer,
+  Config.defaultLayer,
+)
+
+const env = Layer.mergeAll(
+  SessionNs.defaultLayer,
+  CrossSpawnSpawner.defaultLayer,
+  SessionCompaction.layer.pipe(Layer.provide(SessionNs.defaultLayer), Layer.provideMerge(deps)),
+)
+
+const it = testEffect(env)
+
+function llm() {
+  const queue: Array<
+    Stream.Stream<LLM.Event, unknown> | ((input: LLM.StreamInput) => Stream.Stream<LLM.Event, unknown>)
+  > = []
+
+  return {
+    push(stream: Stream.Stream<LLM.Event, unknown> | ((input: LLM.StreamInput) => Stream.Stream<LLM.Event, unknown>)) {
+      queue.push(stream)
+    },
+    layer: Layer.succeed(
+      LLM.Service,
+      LLM.Service.of({
+        stream: (input) => {
+          const item = queue.shift() ?? Stream.empty
+          const stream = typeof item === "function" ? item(input) : item
+          return stream.pipe(Stream.mapEffect((event) => Effect.succeed(event)))
+        },
+        raw: () => Effect.die("raw not implemented in test LLM"),
+      }),
+    ),
+  }
+}
+
+function liveRuntime(layer: Layer.Layer<LLM.Service>, provider = ProviderTest.fake(), config = Config.defaultLayer) {
+  const bus = Bus.layer
+  const status = SessionStatus.layer.pipe(Layer.provide(bus))
+  const processor = SessionProcessorModule.SessionProcessor.layer.pipe(Layer.provide(summary))
+  return ManagedRuntime.make(
+    Layer.mergeAll(SessionCompaction.layer.pipe(Layer.provide(processor)), processor, bus, status).pipe(
+      Layer.provide(provider.layer),
+      Layer.provide(SessionNs.defaultLayer),
+      Layer.provide(Snapshot.defaultLayer),
+      Layer.provide(layer),
+      Layer.provide(Permission.defaultLayer),
+      Layer.provide(Agent.defaultLayer),
+      Layer.provide(Plugin.defaultLayer),
+      Layer.provide(status),
+      Layer.provide(bus),
+      Layer.provide(config),
+    ),
+  )
+}
+
+function reply(
+  text: string,
+  capture?: (input: LLM.StreamInput) => void,
+): (input: LLM.StreamInput) => Stream.Stream<LLM.Event, unknown> {
+  return (input) => {
+    capture?.(input)
+    return Stream.make(
+      { type: "start" } satisfies LLM.Event,
+      { type: "text-start", id: "txt-0" } satisfies LLM.Event,
+      { type: "text-delta", id: "txt-0", delta: text, text } as LLM.Event,
+      { type: "text-end", id: "txt-0" } satisfies LLM.Event,
+      {
+        type: "finish-step",
+        finishReason: "stop",
+        rawFinishReason: "stop",
+        response: { id: "res", modelId: "test-model", timestamp: new Date() },
+        providerMetadata: undefined,
+        usage: {
+          inputTokens: 1,
+          outputTokens: 1,
+          totalTokens: 2,
+          inputTokenDetails: {
+            noCacheTokens: undefined,
+            cacheReadTokens: undefined,
+            cacheWriteTokens: undefined,
+          },
+          outputTokenDetails: {
+            textTokens: undefined,
+            reasoningTokens: undefined,
+          },
+        },
+      } satisfies LLM.Event,
+      {
+        type: "finish",
+        finishReason: "stop",
+        rawFinishReason: "stop",
+        totalUsage: {
+          inputTokens: 1,
+          outputTokens: 1,
+          totalTokens: 2,
+          inputTokenDetails: {
+            noCacheTokens: undefined,
+            cacheReadTokens: undefined,
+            cacheWriteTokens: undefined,
+          },
+          outputTokenDetails: {
+            textTokens: undefined,
+            reasoningTokens: undefined,
+          },
+        },
+      } satisfies LLM.Event,
+    )
+  }
+}
+
+function wait(ms = 50) {
+  return new Promise((resolve) => setTimeout(resolve, ms))
+}
+
+function defer() {
+  let resolve!: () => void
+  const promise = new Promise<void>((done) => {
+    resolve = done
+  })
+  return { promise, resolve }
+}
+
+function plugin(ready: ReturnType<typeof defer>) {
+  return Layer.mock(Plugin.Service)({
+    trigger: <Name extends string, Input, Output>(name: Name, _input: Input, output: Output) => {
+      if (name !== "experimental.session.compacting") return Effect.succeed(output)
+      return Effect.sync(() => ready.resolve()).pipe(Effect.andThen(Effect.never), Effect.as(output))
+    },
+    list: () => Effect.succeed([]),
+    init: () => Effect.void,
+  })
+}
+
+function autocontinue(enabled: boolean) {
+  return Layer.mock(Plugin.Service)({
+    trigger: <Name extends string, Input, Output>(name: Name, _input: Input, output: Output) => {
+      if (name !== "experimental.compaction.autocontinue") return Effect.succeed(output)
+      return Effect.sync(() => {
+        ;(output as { enabled: boolean }).enabled = enabled
+        return output
+      })
+    },
+    list: () => Effect.succeed([]),
+    init: () => Effect.void,
+  })
+}
+
+describe("session.compaction.isOverflow", () => {
+  it.live(
+    "returns true when token count exceeds usable context",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const model = createModel({ context: 100_000, output: 32_000 })
+        const tokens = { input: 75_000, output: 5_000, reasoning: 0, cache: { read: 0, write: 0 } }
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(true)
+      }),
+    ),
+  )
+
+  it.live(
+    "returns false when token count within usable context",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const model = createModel({ context: 200_000, output: 32_000 })
+        const tokens = { input: 100_000, output: 10_000, reasoning: 0, cache: { read: 0, write: 0 } }
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(false)
+      }),
+    ),
+  )
+
+  it.live(
+    "includes cache.read in token count",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const model = createModel({ context: 100_000, output: 32_000 })
+        const tokens = { input: 60_000, output: 10_000, reasoning: 0, cache: { read: 10_000, write: 0 } }
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(true)
+      }),
+    ),
+  )
+
+  it.live(
+    "respects input limit for input caps",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const model = createModel({ context: 400_000, input: 272_000, output: 128_000 })
+        const tokens = { input: 271_000, output: 1_000, reasoning: 0, cache: { read: 2_000, write: 0 } }
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(true)
+      }),
+    ),
+  )
+
+  it.live(
+    "returns false when input/output are within input caps",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const model = createModel({ context: 400_000, input: 272_000, output: 128_000 })
+        const tokens = { input: 200_000, output: 20_000, reasoning: 0, cache: { read: 10_000, write: 0 } }
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(false)
+      }),
+    ),
+  )
+
+  it.live(
+    "returns false when output within limit with input caps",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const model = createModel({ context: 200_000, input: 120_000, output: 10_000 })
+        const tokens = { input: 50_000, output: 9_999, reasoning: 0, cache: { read: 0, write: 0 } }
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(false)
+      }),
+    ),
+  )
+
+  // ─── Bug reproduction tests ───────────────────────────────────────────
+  // These tests demonstrate that when limit.input is set, isOverflow()
+  // does not subtract any headroom for the next model response. This means
+  // compaction only triggers AFTER we've already consumed the full input
+  // budget, leaving zero room for the next API call's output tokens.
+  //
+  // Compare: without limit.input, usable = context - output (reserves space).
+  // With limit.input, usable = limit.input (reserves nothing).
+  //
+  // Related issues: #10634, #8089, #11086, #12621
+  // Open PRs: #6875, #12924
+
+  it.live(
+    "BUG: no headroom when limit.input is set — compaction should trigger near boundary but does not",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        // Simulate Claude with prompt caching: input limit = 200K, output limit = 32K
+        const model = createModel({ context: 200_000, input: 200_000, output: 32_000 })
+
+        // We've used 198K tokens total. Only 2K under the input limit.
+        // On the next turn, the full conversation (198K) becomes input,
+        // plus the model needs room to generate output — this WILL overflow.
+        const tokens = { input: 180_000, output: 15_000, reasoning: 0, cache: { read: 3_000, write: 0 } }
+        // count = 180K + 3K + 15K = 198K
+        // usable = limit.input = 200K (no output subtracted!)
+        // 198K > 200K = false → no compaction triggered
+
+        // WITHOUT limit.input: usable = 200K - 32K = 168K, and 198K > 168K = true ✓
+        // WITH limit.input: usable = 200K, and 198K > 200K = false ✗
+
+        // With 198K used and only 2K headroom, the next turn will overflow.
+        // Compaction MUST trigger here.
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(true)
+      }),
+    ),
+  )
+
+  it.live(
+    "BUG: without limit.input, same token count correctly triggers compaction",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        // Same model but without limit.input — uses context - output instead
+        const model = createModel({ context: 200_000, output: 32_000 })
+
+        // Same token usage as above
+        const tokens = { input: 180_000, output: 15_000, reasoning: 0, cache: { read: 3_000, write: 0 } }
+        // count = 198K
+        // usable = context - output = 200K - 32K = 168K
+        // 198K > 168K = true → compaction correctly triggered
+
+        const result = yield* compact.isOverflow({ tokens, model })
+        expect(result).toBe(true) // ← Correct: headroom is reserved
+      }),
+    ),
+  )
+
+  it.live(
+    "BUG: asymmetry — limit.input model allows 30K more usage before compaction than equivalent model without it",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        // Two models with identical context/output limits, differing only in limit.input
+        const withInputLimit = createModel({ context: 200_000, input: 200_000, output: 32_000 })
+        const withoutInputLimit = createModel({ context: 200_000, output: 32_000 })
+
+        // 170K total tokens — well above context-output (168K) but below input limit (200K)
+        const tokens = { input: 166_000, output: 10_000, reasoning: 0, cache: { read: 5_000, write: 0 } }
+
+        const withLimit = yield* compact.isOverflow({ tokens, model: withInputLimit })
+        const withoutLimit = yield* compact.isOverflow({ tokens, model: withoutInputLimit })
+
+        // Both models have identical real capacity — they should agree:
+        expect(withLimit).toBe(true) // should compact (170K leaves no room for 32K output)
+        expect(withoutLimit).toBe(true) // correctly compacts (170K > 168K)
+      }),
+    ),
+  )
+
+  it.live(
+    "returns false when model context limit is 0",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const model = createModel({ context: 0, output: 32_000 })
+        const tokens = { input: 100_000, output: 10_000, reasoning: 0, cache: { read: 0, write: 0 } }
+        expect(yield* compact.isOverflow({ tokens, model })).toBe(false)
+      }),
+    ),
+  )
+
+  it.live(
+    "returns false when compaction.auto is disabled",
+    provideTmpdirInstance(
+      () =>
+        Effect.gen(function* () {
+          const compact = yield* SessionCompaction.Service
+          const model = createModel({ context: 100_000, output: 32_000 })
+          const tokens = { input: 75_000, output: 5_000, reasoning: 0, cache: { read: 0, write: 0 } }
+          expect(yield* compact.isOverflow({ tokens, model })).toBe(false)
+        }),
+      {
+        config: {
+          compaction: { auto: false },
+        },
+      },
+    ),
+  )
+})
+
+describe("session.compaction.create", () => {
+  it.live(
+    "creates a compaction user message and part",
+    provideTmpdirInstance(() =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const ssn = yield* SessionNs.Service
+
+        const info = yield* ssn.create({})
+
+        yield* compact.create({
+          sessionID: info.id,
+          agent: "build",
+          model: ref,
+          auto: true,
+          overflow: true,
+        })
+
+        const msgs = yield* ssn.messages({ sessionID: info.id })
+        expect(msgs).toHaveLength(1)
+        expect(msgs[0].info.role).toBe("user")
+        expect(msgs[0].parts).toHaveLength(1)
+        expect(msgs[0].parts[0]).toMatchObject({
+          type: "compaction",
+          auto: true,
+          overflow: true,
+        })
+      }),
+    ),
+  )
+})
+
+describe("session.compaction.prune", () => {
+  it.live(
+    "compacts old completed tool output",
+    provideTmpdirInstance(
+      (dir) =>
+        Effect.gen(function* () {
+          const compact = yield* SessionCompaction.Service
+          const ssn = yield* SessionNs.Service
+          const info = yield* ssn.create({})
+          const a = yield* ssn.updateMessage({
+            id: MessageID.ascending(),
+            role: "user",
+            sessionID: info.id,
+            agent: "build",
+            model: ref,
+            time: { created: Date.now() },
+          })
+          yield* ssn.updatePart({
+            id: PartID.ascending(),
+            messageID: a.id,
+            sessionID: info.id,
+            type: "text",
+            text: "first",
+          })
+          const b: MessageV2.Assistant = {
+            id: MessageID.ascending(),
+            role: "assistant",
+            sessionID: info.id,
+            mode: "build",
+            agent: "build",
+            path: { cwd: dir, root: dir },
+            cost: 0,
+            tokens: {
+              output: 0,
+              input: 0,
+              reasoning: 0,
+              cache: { read: 0, write: 0 },
+            },
+            modelID: ref.modelID,
+            providerID: ref.providerID,
+            parentID: a.id,
+            time: { created: Date.now() },
+            finish: "end_turn",
+          }
+          yield* ssn.updateMessage(b)
+          yield* ssn.updatePart({
+            id: PartID.ascending(),
+            messageID: b.id,
+            sessionID: info.id,
+            type: "tool",
+            callID: crypto.randomUUID(),
+            tool: "bash",
+            state: {
+              status: "completed",
+              input: {},
+              output: "x".repeat(200_000),
+              title: "done",
+              metadata: {},
+              time: { start: Date.now(), end: Date.now() },
+            },
+          })
+          for (const text of ["second", "third"]) {
+            const msg = yield* ssn.updateMessage({
+              id: MessageID.ascending(),
+              role: "user",
+              sessionID: info.id,
+              agent: "build",
+              model: ref,
+              time: { created: Date.now() },
+            })
+            yield* ssn.updatePart({
+              id: PartID.ascending(),
+              messageID: msg.id,
+              sessionID: info.id,
+              type: "text",
+              text,
+            })
+          }
+
+          yield* compact.prune({ sessionID: info.id })
+
+          const msgs = yield* ssn.messages({ sessionID: info.id })
+          const part = msgs.flatMap((msg) => msg.parts).find((part) => part.type === "tool")
+          expect(part?.type).toBe("tool")
+          expect(part?.state.status).toBe("completed")
+          if (part?.type === "tool" && part.state.status === "completed") {
+            expect(part.state.time.compacted).toBeNumber()
+          }
+        }),
+
+      {
+        config: {
+          compaction: { prune: true },
+        },
+      },
+    ),
+  )
+
+  it.live(
+    "skips protected skill tool output",
+    provideTmpdirInstance((dir) =>
+      Effect.gen(function* () {
+        const compact = yield* SessionCompaction.Service
+        const ssn = yield* SessionNs.Service
+        const info = yield* ssn.create({})
+        const a = yield* ssn.updateMessage({
+          id: MessageID.ascending(),
+          role: "user",
+          sessionID: info.id,
+          agent: "build",
+          model: ref,
+          time: { created: Date.now() },
+        })
+        yield* ssn.updatePart({
+          id: PartID.ascending(),
+          messageID: a.id,
+          sessionID: info.id,
+          type: "text",
+          text: "first",
+        })
+        const b: MessageV2.Assistant = {
+          id: MessageID.ascending(),
+          role: "assistant",
+          sessionID: info.id,
+          mode: "build",
+          agent: "build",
+          path: { cwd: dir, root: dir },
+          cost: 0,
+          tokens: {
+            output: 0,
+            input: 0,
+            reasoning: 0,
+            cache: { read: 0, write: 0 },
+          },
+          modelID: ref.modelID,
+          providerID: ref.providerID,
+          parentID: a.id,
+          time: { created: Date.now() },
+          finish: "end_turn",
+        }
+        yield* ssn.updateMessage(b)
+        yield* ssn.updatePart({
+          id: PartID.ascending(),
+          messageID: b.id,
+          sessionID: info.id,
+          type: "tool",
+          callID: crypto.randomUUID(),
+          tool: "skill",
+          state: {
+            status: "completed",
+            input: {},
+            output: "x".repeat(200_000),
+            title: "done",
+            metadata: {},
+            time: { start: Date.now(), end: Date.now() },
+          },
+        })
+        for (const text of ["second", "third"]) {
+          const msg = yield* ssn.updateMessage({
+            id: MessageID.ascending(),
+            role: "user",
+            sessionID: info.id,
+            agent: "build",
+            model: ref,
+            time: { created: Date.now() },
+          })
+          yield* ssn.updatePart({
+            id: PartID.ascending(),
+            messageID: msg.id,
+            sessionID: info.id,
+            type: "text",
+            text,
+          })
+        }
+
+        yield* compact.prune({ sessionID: info.id })
+
+        const msgs = yield* ssn.messages({ sessionID: info.id })
+        const part = msgs.flatMap((msg) => msg.parts).find((part) => part.type === "tool")
+        expect(part?.type).toBe("tool")
+        if (part?.type === "tool" && part.state.status === "completed") {
+          expect(part.state.time.compacted).toBeUndefined()
+        }
+      }),
+    ),
+  )
+})
+
+describe("session.compaction.process", () => {
+  test("throws when parent is not a user message", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const reply = await assistant(session.id, msg.id, tmp.path)
+        const rt = runtime("continue")
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          await expect(
+            rt.runPromise(
+              SessionCompaction.Service.use((svc) =>
+                svc.process({
+                  parentID: reply.id,
+                  messages: msgs,
+                  sessionID: session.id,
+                  auto: false,
+                }),
+              ),
+            ),
+          ).rejects.toThrow(`Compaction parent must be a user message: ${reply.id}`)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("publishes compacted event on continue", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const msgs = await svc.messages({ sessionID: session.id })
+        const done = defer()
+        let seen = false
+        const rt = runtime("continue", Plugin.defaultLayer, wide())
+        let unsub: (() => void) | undefined
+        try {
+          unsub = await rt.runPromise(
+            Bus.Service.use((svc) =>
+              svc.subscribeCallback(SessionCompaction.Event.Compacted, (evt) => {
+                if (evt.properties.sessionID !== session.id) return
+                seen = true
+                done.resolve()
+              }),
+            ),
+          )
+
+          const result = await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: msg.id,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          await Promise.race([
+            done.promise,
+            wait(500).then(() => {
+              throw new Error("timed out waiting for compacted event")
+            }),
+          ])
+          expect(result).toBe("continue")
+          expect(seen).toBe(true)
+        } finally {
+          unsub?.()
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("marks summary message as errored on compact result", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const rt = runtime("compact", Plugin.defaultLayer, wide())
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const result = await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: msg.id,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const summary = (await svc.messages({ sessionID: session.id })).find(
+            (msg) => msg.info.role === "assistant" && msg.info.summary,
+          )
+
+          expect(result).toBe("stop")
+          expect(summary?.info.role).toBe("assistant")
+          if (summary?.info.role === "assistant") {
+            expect(summary.info.finish).toBe("error")
+            expect(JSON.stringify(summary.info.error)).toContain("Session too large to compact")
+          }
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("adds synthetic continue prompt when auto is enabled", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const rt = runtime("continue", Plugin.defaultLayer, wide())
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const result = await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: msg.id,
+                messages: msgs,
+                sessionID: session.id,
+                auto: true,
+              }),
+            ),
+          )
+
+          const all = await svc.messages({ sessionID: session.id })
+          const last = all.at(-1)
+
+          expect(result).toBe("continue")
+          expect(last?.info.role).toBe("user")
+          expect(last?.parts[0]).toMatchObject({
+            type: "text",
+            synthetic: true,
+            metadata: { compaction_continue: true },
+          })
+          if (last?.parts[0]?.type === "text") {
+            expect(last.parts[0].text).toContain("Continue if you have next steps")
+          }
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("persists tail_start_id for retained recent turns", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "first")
+        const keep = await user(session.id, "second")
+        await user(session.id, "third")
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = runtime(
+          "continue",
+          Plugin.defaultLayer,
+          wide(),
+          cfg({ tail_turns: 2, preserve_recent_tokens: 10_000 }),
+        )
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const part = await lastCompactionPart(session.id)
+          expect(part?.type).toBe("compaction")
+          expect(part?.tail_start_id).toBe(keep.id)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("shrinks retained tail to fit preserve token budget", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "first")
+        await user(session.id, "x".repeat(2_000))
+        const keep = await user(session.id, "tiny")
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = runtime("continue", Plugin.defaultLayer, wide(), cfg({ tail_turns: 2, preserve_recent_tokens: 100 }))
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const part = await lastCompactionPart(session.id)
+          expect(part?.type).toBe("compaction")
+          expect(part?.tail_start_id).toBe(keep.id)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("falls back to full summary when even one recent turn exceeds preserve token budget", async () => {
+    await using tmp = await tmpdir({ git: true })
+    const stub = llm()
+    let captured = ""
+    stub.push(
+      reply("summary", (input) => {
+        captured = JSON.stringify(input.messages)
+      }),
+    )
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "first")
+        await user(session.id, "y".repeat(2_000))
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 1, preserve_recent_tokens: 20 }))
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const part = await lastCompactionPart(session.id)
+          expect(part?.type).toBe("compaction")
+          expect(part?.tail_start_id).toBeUndefined()
+          expect(captured).toContain("yyyy")
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("falls back to full summary when retained tail media exceeds preserve token budget", async () => {
+    await using tmp = await tmpdir({ git: true })
+    const stub = llm()
+    let captured = ""
+    stub.push(
+      reply("summary", (input) => {
+        captured = JSON.stringify(input.messages)
+      }),
+    )
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "older")
+        const recent = await user(session.id, "recent image turn")
+        await svc.updatePart({
+          id: PartID.ascending(),
+          messageID: recent.id,
+          sessionID: session.id,
+          type: "file",
+          mime: "image/png",
+          filename: "big.png",
+          url: `data:image/png;base64,${"a".repeat(4_000)}`,
+        })
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 1, preserve_recent_tokens: 100 }))
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const part = await lastCompactionPart(session.id)
+          expect(part?.type).toBe("compaction")
+          expect(part?.tail_start_id).toBeUndefined()
+          expect(captured).toContain("recent image turn")
+          expect(captured).toContain("Attached image/png: big.png")
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("retains a split turn suffix when a later message fits the preserve token budget", async () => {
+    await using tmp = await tmpdir({ git: true })
+    const stub = llm()
+    let captured = ""
+    stub.push(
+      reply("summary", (input) => {
+        captured = JSON.stringify(input.messages)
+      }),
+    )
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "older")
+        const recent = await user(session.id, "recent turn")
+        const large = await assistant(session.id, recent.id, tmp.path)
+        await svc.updatePart({
+          id: PartID.ascending(),
+          messageID: large.id,
+          sessionID: session.id,
+          type: "text",
+          text: "z".repeat(2_000),
+        })
+        const keep = await assistant(session.id, recent.id, tmp.path)
+        await svc.updatePart({
+          id: PartID.ascending(),
+          messageID: keep.id,
+          sessionID: session.id,
+          type: "text",
+          text: "keep tail",
+        })
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 1, preserve_recent_tokens: 100 }))
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const part = await lastCompactionPart(session.id)
+          expect(part?.type).toBe("compaction")
+          expect(part?.tail_start_id).toBe(keep.id)
+          expect(captured).toContain("zzzz")
+          expect(captured).not.toContain("keep tail")
+
+          const filtered = MessageV2.filterCompacted(MessageV2.stream(session.id))
+          expect(filtered[0]?.info.id).toBe(keep.id)
+          expect(filtered.map((msg) => msg.info.id)).not.toContain(large.id)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("allows plugins to disable synthetic continue prompt", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const rt = runtime("continue", autocontinue(false), wide())
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const result = await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: msg.id,
+                messages: msgs,
+                sessionID: session.id,
+                auto: true,
+              }),
+            ),
+          )
+
+          const all = await svc.messages({ sessionID: session.id })
+          const last = all.at(-1)
+
+          expect(result).toBe("continue")
+          expect(last?.info.role).toBe("assistant")
+          expect(
+            all.some(
+              (msg) =>
+                msg.info.role === "user" &&
+                msg.parts.some(
+                  (part) =>
+                    part.type === "text" && part.synthetic && part.text.includes("Continue if you have next steps"),
+                ),
+            ),
+          ).toBe(false)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("replays the prior user turn on overflow when earlier context exists", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "root")
+        const replay = await user(session.id, "image")
+        await svc.updatePart({
+          id: PartID.ascending(),
+          messageID: replay.id,
+          sessionID: session.id,
+          type: "file",
+          mime: "image/png",
+          filename: "cat.png",
+          url: "https://example.com/cat.png",
+        })
+        const msg = await user(session.id, "current")
+        const rt = runtime("continue", Plugin.defaultLayer, wide())
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const result = await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: msg.id,
+                messages: msgs,
+                sessionID: session.id,
+                auto: true,
+                overflow: true,
+              }),
+            ),
+          )
+
+          const last = (await svc.messages({ sessionID: session.id })).at(-1)
+
+          expect(result).toBe("continue")
+          expect(last?.info.role).toBe("user")
+          expect(last?.parts.some((part) => part.type === "file")).toBe(false)
+          expect(
+            last?.parts.some((part) => part.type === "text" && part.text.includes("Attached image/png: cat.png")),
+          ).toBe(true)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("falls back to overflow guidance when no replayable turn exists", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "earlier")
+        const msg = await user(session.id, "current")
+
+        const rt = runtime("continue", Plugin.defaultLayer, wide())
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const result = await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: msg.id,
+                messages: msgs,
+                sessionID: session.id,
+                auto: true,
+                overflow: true,
+              }),
+            ),
+          )
+
+          const last = (await svc.messages({ sessionID: session.id })).at(-1)
+
+          expect(result).toBe("continue")
+          expect(last?.info.role).toBe("user")
+          if (last?.parts[0]?.type === "text") {
+            expect(last.parts[0].text).toContain("previous request exceeded the provider's size limit")
+          }
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("stops quickly when aborted during retry backoff", async () => {
+    const stub = llm()
+    const ready = defer()
+    stub.push(
+      Stream.fromAsyncIterable(
+        {
+          async *[Symbol.asyncIterator]() {
+            yield { type: "start" } as LLM.Event
+            throw new APICallError({
+              message: "boom",
+              url: "https://example.com/v1/chat/completions",
+              requestBodyValues: {},
+              statusCode: 503,
+              responseHeaders: { "retry-after-ms": "10000" },
+              responseBody: '{"error":"boom"}',
+              isRetryable: true,
+            })
+          },
+        },
+        (err) => err,
+      ),
+    )
+
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const msgs = await svc.messages({ sessionID: session.id })
+        const abort = new AbortController()
+        const rt = liveRuntime(stub.layer, wide())
+        let off: (() => void) | undefined
+        let run: Promise<"continue" | "stop"> | undefined
+        try {
+          off = await rt.runPromise(
+            Bus.Service.use((svc) =>
+              svc.subscribeCallback(SessionStatus.Event.Status, (evt) => {
+                if (evt.properties.sessionID !== session.id) return
+                if (evt.properties.status.type !== "retry") return
+                ready.resolve()
+              }),
+            ),
+          )
+
+          run = rt
+            .runPromiseExit(
+              SessionCompaction.Service.use((svc) =>
+                svc.process({
+                  parentID: msg.id,
+                  messages: msgs,
+                  sessionID: session.id,
+                  auto: false,
+                }),
+              ),
+              { signal: abort.signal },
+            )
+            .then((exit) => {
+              if (Exit.isFailure(exit)) {
+                if (Cause.hasInterrupts(exit.cause) && abort.signal.aborted) return "stop"
+                throw Cause.squash(exit.cause)
+              }
+              return exit.value
+            })
+
+          await Promise.race([
+            ready.promise,
+            wait(5000).then(() => {
+              throw new Error("timed out waiting for retry status")
+            }),
+          ])
+
+          const start = Date.now()
+          abort.abort()
+          const result = await Promise.race([
+            run.then((value) => ({ kind: "done" as const, value, ms: Date.now() - start })),
+            wait(2000).then(() => ({ kind: "timeout" as const })),
+          ])
+
+          expect(result.kind).toBe("done")
+          if (result.kind === "done") {
+            expect(result.value).toBe("stop")
+            expect(result.ms).toBeLessThan(2000)
+          }
+        } finally {
+          off?.()
+          abort.abort()
+          await rt.dispose()
+          await run?.catch(() => undefined)
+        }
+      },
+    })
+  })
+
+  test("does not leave a summary assistant when aborted before processor setup", async () => {
+    const ready = defer()
+
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const msgs = await svc.messages({ sessionID: session.id })
+        const abort = new AbortController()
+        const rt = runtime("continue", plugin(ready), wide())
+        let run: Promise<"continue" | "stop"> | undefined
+        try {
+          run = rt
+            .runPromiseExit(
+              SessionCompaction.Service.use((svc) =>
+                svc.process({
+                  parentID: msg.id,
+                  messages: msgs,
+                  sessionID: session.id,
+                  auto: false,
+                }),
+              ),
+              { signal: abort.signal },
+            )
+            .then((exit) => {
+              if (Exit.isFailure(exit)) {
+                if (Cause.hasInterrupts(exit.cause) && abort.signal.aborted) return "stop"
+                throw Cause.squash(exit.cause)
+              }
+              return exit.value
+            })
+
+          await Promise.race([
+            ready.promise,
+            wait(1000).then(() => {
+              throw new Error("timed out waiting for compaction hook")
+            }),
+          ])
+
+          abort.abort()
+          expect(await run).toBe("stop")
+
+          const all = await svc.messages({ sessionID: session.id })
+          expect(all.some((msg) => msg.info.role === "assistant" && msg.info.summary)).toBe(false)
+        } finally {
+          abort.abort()
+          await rt.dispose()
+          await run?.catch(() => undefined)
+        }
+      },
+    })
+  })
+
+  test("does not allow tool calls while generating the summary", async () => {
+    const stub = llm()
+    stub.push(
+      Stream.make(
+        { type: "start" } satisfies LLM.Event,
+        { type: "tool-input-start", id: "call-1", toolName: "_noop" } satisfies LLM.Event,
+        { type: "tool-call", toolCallId: "call-1", toolName: "_noop", input: {} } satisfies LLM.Event,
+        {
+          type: "finish-step",
+          finishReason: "tool-calls",
+          rawFinishReason: "tool_calls",
+          response: { id: "res", modelId: "test-model", timestamp: new Date() },
+          providerMetadata: undefined,
+          usage: {
+            inputTokens: 1,
+            outputTokens: 1,
+            totalTokens: 2,
+            inputTokenDetails: {
+              noCacheTokens: undefined,
+              cacheReadTokens: undefined,
+              cacheWriteTokens: undefined,
+            },
+            outputTokenDetails: {
+              textTokens: undefined,
+              reasoningTokens: undefined,
+            },
+          },
+        } satisfies LLM.Event,
+        {
+          type: "finish",
+          finishReason: "tool-calls",
+          rawFinishReason: "tool_calls",
+          totalUsage: {
+            inputTokens: 1,
+            outputTokens: 1,
+            totalTokens: 2,
+            inputTokenDetails: {
+              noCacheTokens: undefined,
+              cacheReadTokens: undefined,
+              cacheWriteTokens: undefined,
+            },
+            outputTokenDetails: {
+              textTokens: undefined,
+              reasoningTokens: undefined,
+            },
+          },
+        } satisfies LLM.Event,
+      ),
+    )
+
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const msg = await user(session.id, "hello")
+        const rt = liveRuntime(stub.layer, wide())
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: msg.id,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const summary = (await svc.messages({ sessionID: session.id })).find(
+            (item) => item.info.role === "assistant" && item.info.summary,
+          )
+
+          expect(summary?.info.role).toBe("assistant")
+          expect(summary?.parts.some((part) => part.type === "tool")).toBe(false)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("summarizes only the head while keeping recent tail out of summary input", async () => {
+    const stub = llm()
+    let captured = ""
+    stub.push(
+      reply("summary", (input) => {
+        captured = JSON.stringify(input.messages)
+      }),
+    )
+
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "older context")
+        await user(session.id, "keep this turn")
+        await user(session.id, "and this one too")
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = liveRuntime(stub.layer, wide())
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          expect(captured).toContain("older context")
+          expect(captured).not.toContain("keep this turn")
+          expect(captured).not.toContain("and this one too")
+          expect(captured).not.toContain("What did we do so far?")
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("anchors repeated compactions with the previous summary", async () => {
+    const stub = llm()
+    let captured = ""
+    stub.push(reply("summary one"))
+    stub.push(
+      reply("summary two", (input) => {
+        captured = JSON.stringify(input.messages)
+      }),
+    )
+
+    await using tmp = await tmpdir({ git: true })
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "older context")
+        await user(session.id, "keep this turn")
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = liveRuntime(stub.layer, wide())
+        try {
+          let msgs = await svc.messages({ sessionID: session.id })
+          let parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          await user(session.id, "latest turn")
+          await SessionCompaction.create({
+            sessionID: session.id,
+            agent: "build",
+            model: ref,
+            auto: false,
+          })
+
+          msgs = MessageV2.filterCompacted(MessageV2.stream(session.id))
+          parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          expect(captured).toContain("<previous-summary>")
+          expect(captured).toContain("summary one")
+          expect(captured.match(/summary one/g)?.length).toBe(1)
+          expect(captured).toContain("## Constraints & Preferences")
+          expect(captured).toContain("## Progress")
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("keeps recent pre-compaction turns across repeated compactions", async () => {
+    const stub = llm()
+    stub.push(reply("summary one"))
+    stub.push(reply("summary two"))
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        const u1 = await user(session.id, "one")
+        const u2 = await user(session.id, "two")
+        const u3 = await user(session.id, "three")
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = liveRuntime(stub.layer, wide(), cfg({ tail_turns: 2, preserve_recent_tokens: 10_000 }))
+        try {
+          let msgs = await svc.messages({ sessionID: session.id })
+          let parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const u4 = await user(session.id, "four")
+          await SessionCompaction.create({
+            sessionID: session.id,
+            agent: "build",
+            model: ref,
+            auto: false,
+          })
+
+          msgs = MessageV2.filterCompacted(MessageV2.stream(session.id))
+          parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const filtered = MessageV2.filterCompacted(MessageV2.stream(session.id))
+          const ids = filtered.map((msg) => msg.info.id)
+
+          expect(ids).not.toContain(u1.id)
+          expect(ids).not.toContain(u2.id)
+          expect(ids).toContain(u3.id)
+          expect(ids).toContain(u4.id)
+          expect(filtered.some((msg) => msg.info.role === "assistant" && msg.info.summary)).toBe(true)
+          expect(
+            filtered.some((msg) => msg.info.role === "user" && msg.parts.some((part) => part.type === "compaction")),
+          ).toBe(true)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+
+  test("ignores previous summaries when sizing the retained tail", async () => {
+    await using tmp = await tmpdir()
+    await Instance.provide({
+      directory: tmp.path,
+      fn: async () => {
+        const session = await svc.create({})
+        await user(session.id, "older")
+        const keep = await user(session.id, "keep this turn")
+        const keepReply = await assistant(session.id, keep.id, tmp.path)
+        await svc.updatePart({
+          id: PartID.ascending(),
+          messageID: keepReply.id,
+          sessionID: session.id,
+          type: "text",
+          text: "keep reply",
+        })
+
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+        const firstCompaction = (await svc.messages({ sessionID: session.id })).at(-1)?.info.id
+        expect(firstCompaction).toBeTruthy()
+        await summaryAssistant(session.id, firstCompaction!, tmp.path, "summary ".repeat(800))
+
+        const recent = await user(session.id, "recent turn")
+        const recentReply = await assistant(session.id, recent.id, tmp.path)
+        await svc.updatePart({
+          id: PartID.ascending(),
+          messageID: recentReply.id,
+          sessionID: session.id,
+          type: "text",
+          text: "recent reply",
+        })
+
+        await SessionCompaction.create({
+          sessionID: session.id,
+          agent: "build",
+          model: ref,
+          auto: false,
+        })
+
+        const rt = runtime("continue", Plugin.defaultLayer, wide(), cfg({ tail_turns: 2, preserve_recent_tokens: 500 }))
+        try {
+          const msgs = await svc.messages({ sessionID: session.id })
+          const parent = msgs.at(-1)?.info.id
+          expect(parent).toBeTruthy()
+          await rt.runPromise(
+            SessionCompaction.Service.use((svc) =>
+              svc.process({
+                parentID: parent!,
+                messages: msgs,
+                sessionID: session.id,
+                auto: false,
+              }),
+            ),
+          )
+
+          const part = await lastCompactionPart(session.id)
+          expect(part?.type).toBe("compaction")
+          expect(part?.tail_start_id).toBe(keep.id)
+        } finally {
+          await rt.dispose()
+        }
+      },
+    })
+  })
+})
+
+describe("util.token.estimate", () => {
+  test("estimates tokens from text (4 chars per token)", () => {
+    const text = "x".repeat(4000)
+    expect(Token.estimate(text)).toBe(1000)
+  })
+
+  test("estimates tokens from larger text", () => {
+    const text = "y".repeat(20_000)
+    expect(Token.estimate(text)).toBe(5000)
+  })
+
+  test("returns 0 for empty string", () => {
+    expect(Token.estimate("")).toBe(0)
+  })
+})
+
+describe("SessionNs.getUsage", () => {
+  test("normalizes standard usage to token format", () => {
+    const model = createModel({ context: 100_000, output: 32_000 })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        inputTokenDetails: {
+          noCacheTokens: undefined,
+          cacheReadTokens: undefined,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      },
+    })
+
+    expect(result.tokens.input).toBe(1000)
+    expect(result.tokens.output).toBe(500)
+    expect(result.tokens.reasoning).toBe(0)
+    expect(result.tokens.cache.read).toBe(0)
+    expect(result.tokens.cache.write).toBe(0)
+  })
+
+  test("extracts cached tokens to cache.read", () => {
+    const model = createModel({ context: 100_000, output: 32_000 })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        inputTokenDetails: {
+          noCacheTokens: 800,
+          cacheReadTokens: 200,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      },
+    })
+
+    expect(result.tokens.input).toBe(800)
+    expect(result.tokens.cache.read).toBe(200)
+  })
+
+  test("handles anthropic cache write metadata", () => {
+    const model = createModel({ context: 100_000, output: 32_000 })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        inputTokenDetails: {
+          noCacheTokens: undefined,
+          cacheReadTokens: undefined,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      },
+      metadata: {
+        anthropic: {
+          cacheCreationInputTokens: 300,
+        },
+      },
+    })
+
+    expect(result.tokens.cache.write).toBe(300)
+  })
+
+  test("subtracts cached tokens for anthropic provider", () => {
+    const model = createModel({ context: 100_000, output: 32_000 })
+    // AI SDK v6 normalizes inputTokens to include cached tokens for all providers
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        inputTokenDetails: {
+          noCacheTokens: 800,
+          cacheReadTokens: 200,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      },
+      metadata: {
+        anthropic: {},
+      },
+    })
+
+    expect(result.tokens.input).toBe(800)
+    expect(result.tokens.cache.read).toBe(200)
+  })
+
+  test("separates reasoning tokens from output tokens", () => {
+    const model = createModel({ context: 100_000, output: 32_000 })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        inputTokenDetails: {
+          noCacheTokens: undefined,
+          cacheReadTokens: undefined,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: 400,
+          reasoningTokens: 100,
+        },
+      },
+    })
+
+    expect(result.tokens.input).toBe(1000)
+    expect(result.tokens.output).toBe(400)
+    expect(result.tokens.reasoning).toBe(100)
+    expect(result.tokens.total).toBe(1500)
+  })
+
+  test("does not double count reasoning tokens in cost", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 0,
+        output: 15,
+        cache: { read: 0, write: 0 },
+      },
+    })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 0,
+        outputTokens: 1_000_000,
+        totalTokens: 1_000_000,
+        inputTokenDetails: {
+          noCacheTokens: undefined,
+          cacheReadTokens: undefined,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: 750_000,
+          reasoningTokens: 250_000,
+        },
+      },
+    })
+
+    expect(result.tokens.output).toBe(750_000)
+    expect(result.tokens.reasoning).toBe(250_000)
+    expect(result.cost).toBe(15)
+  })
+
+  test("handles undefined optional values gracefully", () => {
+    const model = createModel({ context: 100_000, output: 32_000 })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 0,
+        outputTokens: 0,
+        totalTokens: 0,
+        inputTokenDetails: {
+          noCacheTokens: undefined,
+          cacheReadTokens: undefined,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      },
+    })
+
+    expect(result.tokens.input).toBe(0)
+    expect(result.tokens.output).toBe(0)
+    expect(result.tokens.reasoning).toBe(0)
+    expect(result.tokens.cache.read).toBe(0)
+    expect(result.tokens.cache.write).toBe(0)
+    expect(Number.isNaN(result.cost)).toBe(false)
+  })
+
+  test("calculates cost correctly", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: {
+          noCacheTokens: undefined,
+          cacheReadTokens: undefined,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      },
+    })
+
+    expect(result.cost).toBe(3 + 1.5)
+  })
+  test("uses openrouter provider cost when available", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: { noCacheTokens: undefined, cacheReadTokens: undefined, cacheWriteTokens: undefined },
+        outputTokenDetails: { textTokens: undefined, reasoningTokens: undefined },
+      },
+      metadata: {
+        openrouter: {
+          usage: {
+            cost: 0.42, // Provider-reported cost should be used instead of calculated
+          },
+        },
+      },
+    })
+
+    // Should use the provider cost (0.42) instead of calculated cost (4.5)
+    expect(result.cost).toBe(0.42)
+  })
+
+  test("falls back to calculated cost when openrouter cost is not available", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: { noCacheTokens: undefined, cacheReadTokens: undefined, cacheWriteTokens: undefined },
+        outputTokenDetails: { textTokens: undefined, reasoningTokens: undefined },
+      },
+      metadata: {
+        openrouter: {
+          usage: {
+            // cost is undefined
+          },
+        },
+      },
+    })
+
+    // Should fall back to calculated cost
+    expect(result.cost).toBe(3 + 1.5)
+  })
+
+  test("falls back to calculated cost when openrouter metadata is empty", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: { noCacheTokens: undefined, cacheReadTokens: undefined, cacheWriteTokens: undefined },
+        outputTokenDetails: { textTokens: undefined, reasoningTokens: undefined },
+      },
+      metadata: {
+        openrouter: {},
+      },
+    })
+
+    // Should fall back to calculated cost
+    expect(result.cost).toBe(3 + 1.5)
+  })
+
+  test("uses upstreamInferenceCost for Saeeol provider", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const provider = { id: "saeeol" } as Provider.Info
+    const result = SessionNs.getUsage({
+      model,
+      provider,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: { noCacheTokens: undefined, cacheReadTokens: undefined, cacheWriteTokens: undefined },
+        outputTokenDetails: { textTokens: undefined, reasoningTokens: undefined },
+      },
+      metadata: {
+        openrouter: {
+          usage: {
+            cost: 0.01, // OpenRouter 5% fee
+            costDetails: {
+              upstreamInferenceCost: 0.2, // Actual inference cost
+            },
+          },
+        },
+      },
+    })
+
+    // Should use upstreamInferenceCost for Saeeol provider (BYOK)
+    expect(result.cost).toBe(0.2)
+  })
+
+  test("uses regular cost for OpenRouter provider", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const provider = { id: "openrouter" } as Provider.Info
+    const result = SessionNs.getUsage({
+      model,
+      provider,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: { noCacheTokens: undefined, cacheReadTokens: undefined, cacheWriteTokens: undefined },
+        outputTokenDetails: { textTokens: undefined, reasoningTokens: undefined },
+      },
+      metadata: {
+        openrouter: {
+          usage: {
+            cost: 0.5, // Regular OpenRouter cost
+            costDetails: {
+              upstreamInferenceCost: 0.45,
+            },
+          },
+        },
+      },
+    })
+
+    // Should use regular cost for OpenRouter provider
+    expect(result.cost).toBe(0.5)
+  })
+
+  test("falls back to regular cost when provider is not specified", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: { noCacheTokens: undefined, cacheReadTokens: undefined, cacheWriteTokens: undefined },
+        outputTokenDetails: { textTokens: undefined, reasoningTokens: undefined },
+      },
+      metadata: {
+        openrouter: {
+          usage: {
+            cost: 0.3,
+            costDetails: {
+              upstreamInferenceCost: 0.25,
+            },
+          },
+        },
+      },
+    })
+
+    // Should use regular cost when provider is not specified
+    expect(result.cost).toBe(0.3)
+  })
+
+  test("uses regular cost when upstreamInferenceCost is missing for Saeeol", () => {
+    const model = createModel({
+      context: 100_000,
+      output: 32_000,
+      cost: {
+        input: 3,
+        output: 15,
+        cache: { read: 0.3, write: 3.75 },
+      },
+    })
+    const provider = { id: "saeeol" } as Provider.Info
+    const result = SessionNs.getUsage({
+      model,
+      provider,
+      usage: {
+        inputTokens: 1_000_000,
+        outputTokens: 100_000,
+        totalTokens: 1_100_000,
+        inputTokenDetails: { noCacheTokens: undefined, cacheReadTokens: undefined, cacheWriteTokens: undefined },
+        outputTokenDetails: { textTokens: undefined, reasoningTokens: undefined },
+      },
+      metadata: {
+        openrouter: {
+          usage: {
+            cost: 0.01,
+            // costDetails is missing
+          },
+        },
+      },
+    })
+
+    // When upstream cost is missing for Saeeol, fall back to regular cost field
+    expect(result.cost).toBe(0.01)
+  })
+
+  // Tests for Anthropic Messages / OpenAI Responses / Vercel AI Gateway cost extraction
+  // live in test/saeeol/provider-cost.test.ts (saeeol_change).
+
+  test.each(["@ai-sdk/anthropic", "@ai-sdk/amazon-bedrock", "@ai-sdk/google-vertex/anthropic"])(
+    "computes total from components for %s models",
+    (npm) => {
+      const model = createModel({ context: 100_000, output: 32_000, npm })
+      // AI SDK v6: inputTokens includes cached tokens for all providers
+      const usage = {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        inputTokenDetails: {
+          noCacheTokens: 800,
+          cacheReadTokens: 200,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      }
+      if (npm === "@ai-sdk/amazon-bedrock") {
+        const result = SessionNs.getUsage({
+          model,
+          usage,
+          metadata: {
+            bedrock: {
+              usage: {
+                cacheWriteInputTokens: 300,
+              },
+            },
+          },
+        })
+
+        // inputTokens (1000) includes cache, so adjusted = 1000 - 200 - 300 = 500
+        expect(result.tokens.input).toBe(500)
+        expect(result.tokens.cache.read).toBe(200)
+        expect(result.tokens.cache.write).toBe(300)
+        // total = adjusted (500) + output (500) + cacheRead (200) + cacheWrite (300)
+        expect(result.tokens.total).toBe(1500)
+        return
+      }
+
+      const result = SessionNs.getUsage({
+        model,
+        usage,
+        metadata: {
+          anthropic: {
+            cacheCreationInputTokens: 300,
+          },
+        },
+      })
+
+      // inputTokens (1000) includes cache, so adjusted = 1000 - 200 - 300 = 500
+      expect(result.tokens.input).toBe(500)
+      expect(result.tokens.cache.read).toBe(200)
+      expect(result.tokens.cache.write).toBe(300)
+      // total = adjusted (500) + output (500) + cacheRead (200) + cacheWrite (300)
+      expect(result.tokens.total).toBe(1500)
+    },
+  )
+
+  test("extracts cache write tokens from vertex metadata key", () => {
+    const model = createModel({ context: 100_000, output: 32_000, npm: "@ai-sdk/google-vertex/anthropic" })
+    const result = SessionNs.getUsage({
+      model,
+      usage: {
+        inputTokens: 1000,
+        outputTokens: 500,
+        totalTokens: 1500,
+        inputTokenDetails: {
+          noCacheTokens: 800,
+          cacheReadTokens: 200,
+          cacheWriteTokens: undefined,
+        },
+        outputTokenDetails: {
+          textTokens: undefined,
+          reasoningTokens: undefined,
+        },
+      },
+      metadata: {
+        vertex: {
+          cacheCreationInputTokens: 300,
+        },
+      },
+    })
+
+    expect(result.tokens.input).toBe(500)
+    expect(result.tokens.cache.read).toBe(200)
+    expect(result.tokens.cache.write).toBe(300)
+  })
+})