chore(agent-core): remove experimental micro compaction

.changeset/remove-micro-compaction.md

@@ -0,0 +1,5 @@
+---
+"@moonshot-ai/kimi-code": patch
+---
+
+Remove the experimental micro compaction feature and its toggle from the experiments panel.

apps/kimi-code/src/tui/commands/config.ts

@@ -1,7 +1,6 @@
 import {
   effectiveModelAlias,
   type ExperimentalFeatureState,
-  type FlagId,
   type ModelAlias,
   type PermissionMode,
   type Session,
@@ -590,7 +589,7 @@ export async function applyExperimentalFeatureChanges(
     return;
   }
 
-  const experimental: Partial<Record<FlagId, boolean>> = {};
+  const experimental: Record<string, boolean> = {};
   for (const change of changes) {
     experimental[change.id] = change.enabled;
   }

docs/en/configuration/config-files.md

@@ -51,8 +51,8 @@ reserved_context_size = 50000
 max_running_tasks = 4
 keep_alive_on_exit = false
 
-[experimental]
-micro_compaction = false
+# [experimental]
+# micro_compaction = false  # disabled: micro compaction has been removed
 
 [[permission.rules]]
 decision = "allow"
@@ -200,13 +200,15 @@ You can also switch models temporarily without touching the config file — by s
 
 `keep_alive_on_exit` can be overridden by the `KIMI_CODE_BACKGROUND_KEEP_ALIVE_ON_EXIT` environment variable, which takes higher priority than `config.toml`.
 
+<!--
 ## `experimental`
 
 `experimental` stores persistent overrides for experimental-feature flags. Currently, `micro_compaction` is the only user-facing entry and defaults to `false`; set it to `true` to enable automatic trimming of older large tool results.
 
 | Field | Type | Default | Description |
 | --- | --- | --- | --- |
 | `micro_compaction` | `boolean` | `false` | Trim older large tool results from context while preserving recent conversation |
+-->
 
 ## `services`
 

docs/en/configuration/env-vars.md

@@ -125,7 +125,6 @@ Switches that control the behavior of subsystems such as telemetry, background t
 | `KIMI_CODE_PLUGIN_MARKETPLACE_URL` | Override the plugin marketplace JSON loaded by `/plugins`; useful for dev loopback servers, staging CDN files, or alternate marketplace directories | `https://code.kimi.com/kimi-code/plugins/marketplace.json`; also accepts `http://`, `file://` URLs, and local paths |
 | `KIMI_CODE_AGENT_SWARM_MAX_CONCURRENCY` | Cap how many AgentSwarm subagents run concurrently during the initial ramp; leave unset for no cap | Positive integer; invalid values fail fast |
 | `KIMI_CODE_EXPERIMENTAL_FLAG` | Enable all registered experimental features for this process | `1`, `true`, `yes`, `on` |
-| `KIMI_CODE_EXPERIMENTAL_MICRO_COMPACTION` | Override [`[experimental].micro_compaction`](./config-files.md#experimental) for this process | Truthy or falsy |
 | `KIMI_SHELL_PATH` | Override the Git Bash path on Windows (used when auto-detection fails) | Absolute path |
 | `KIMI_MODEL_MAX_COMPLETION_TOKENS` | Hard cap on `max_completion_tokens` per LLM step; applies to the `kimi` provider only | Positive integer; `0` or negative disables clamping |
 | `KIMI_MODEL_TEMPERATURE` | Sampling temperature for every request; applies to the `kimi` provider only (global — independent of `KIMI_MODEL_NAME`) | Number, e.g. `0.3` |

docs/zh/configuration/config-files.md

@@ -51,8 +51,8 @@ reserved_context_size = 50000
 max_running_tasks = 4
 keep_alive_on_exit = false
 
-[experimental]
-micro_compaction = false
+# [experimental]
+# micro_compaction = false  # disabled: micro compaction has been removed
 
 [[permission.rules]]
 decision = "allow"
@@ -200,13 +200,15 @@ default_effort = "high"
 
 `keep_alive_on_exit` 可被环境变量 `KIMI_CODE_BACKGROUND_KEEP_ALIVE_ON_EXIT` 覆盖，优先级高于配置文件。
 
+<!--
 ## `experimental`
 
 `experimental` 存放实验功能 flag 的持久化覆盖。目前 `micro_compaction` 是唯一用户可见的字段，默认值为 `false`；如需自动清理较旧的大型工具结果，把它设为 `true`。
 
 | 字段 | 类型 | 默认值 | 说明 |
 | --- | --- | --- | --- |
 | `micro_compaction` | `boolean` | `false` | 清理较旧的大型工具结果内容，同时保留最近对话 |
+-->
 
 ## `services`
 

docs/zh/configuration/env-vars.md

@@ -125,7 +125,6 @@ kimi
 | `KIMI_CODE_PLUGIN_MARKETPLACE_URL` | 覆盖 `/plugins` 加载的 plugin marketplace JSON，适合 dev loopback server、测试 CDN 文件或替换 marketplace 目录 | `https://code.kimi.com/kimi-code/plugins/marketplace.json`；也接受 `http://`、`file://` URL 和本地路径 |
 | `KIMI_CODE_AGENT_SWARM_MAX_CONCURRENCY` | 限制 AgentSwarm 初始提升并发阶段可同时运行的子 Agent 数量；不设置表示不限制 | 正整数；非法值会立即失败 |
 | `KIMI_CODE_EXPERIMENTAL_FLAG` | 在当前进程启用所有已注册的实验功能 | `1`、`true`、`yes`、`on` |
-| `KIMI_CODE_EXPERIMENTAL_MICRO_COMPACTION` | 覆盖当前进程的 [`[experimental].micro_compaction`](./config-files.md#experimental) | 真值或假值 |
 | `KIMI_SHELL_PATH` | Windows 上覆盖 Git Bash 路径（自动探测失败时使用） | 绝对路径 |
 | `KIMI_MODEL_MAX_COMPLETION_TOKENS` | 单步 LLM 请求的 `max_completion_tokens` 硬上限，仅对 `kimi` 供应商生效 | 正整数；`0` 或负数禁用 clamp |
 | `KIMI_MODEL_TEMPERATURE` | 每次请求的采样温度，仅对 `kimi` 供应商生效（全局生效，不依赖 `KIMI_MODEL_NAME`） | 数字，如 `0.3` |

packages/agent-core/src/agent/compaction/micro.ts

@@ -1,10 +1,11 @@
-import type { ContentPart } from '@moonshot-ai/kosong';
+// Micro compaction is disabled; ContentPart is no longer referenced here.
+// import type { ContentPart } from '@moonshot-ai/kosong';
 
 import type { Agent } from '..';
 import type { ContextMessage } from '../context';
 import {
   estimateTokensForContentParts,
-  estimateTokensForMessages,
+  // estimateTokensForMessages, // disabled with micro compaction
 } from '../../utils/tokens';
 
 export interface MicroCompactionConfig {
@@ -47,83 +48,93 @@ export class MicroCompaction {
   }
 
   detect(): void {
-    if (!this.agent.experimentalFlags.enabled('micro_compaction')) return;
-
-    const config = this.config;
-    const { history, lastAssistantAt } = this.agent.context;
-    const cacheAgeMs = lastAssistantAt === null ? null : Date.now() - lastAssistantAt;
-    const cacheMissed = cacheAgeMs !== null && cacheAgeMs >= config.cacheMissedThresholdMs;
-    if (!cacheMissed) return;
-
-    const maxContextTokens = this.agent.config.modelCapabilities.max_context_tokens;
-    const contextTokens = this.agent.context.tokenCountWithPending;
-    const contextUsageRatio =
-      maxContextTokens !== undefined && maxContextTokens > 0
-        ? contextTokens / maxContextTokens
-        : 1;
-    if (contextUsageRatio < config.minContextUsageRatio) return;
-
-    const previousCutoff = this.cutoff;
-    const nextCutoff = Math.max(0, history.length - config.keepRecentMessages);
-    this.apply(nextCutoff);
-    if (previousCutoff !== nextCutoff) {
-      const effect = this.measureEffect(history, nextCutoff);
-      const previousEffect = this.measureEffect(history, previousCutoff);
-      const rawContextTokens = estimateTokensForMessages(history);
-      // Whole-context length before/after this cutoff change, mirroring the
-      // `tokens_before`/`tokens_after` fields on `compaction_finished` so the
-      // two compaction paths can be compared on the same axis.
-      const tokensBefore =
-        rawContextTokens -
-        previousEffect.truncatedToolResultTokensBefore +
-        previousEffect.truncatedToolResultTokensAfter;
-      const tokensAfter =
-        rawContextTokens -
-        effect.truncatedToolResultTokensBefore +
-        effect.truncatedToolResultTokensAfter;
-      this.agent.telemetry.track('micro_compaction_finished', {
-        keep_recent_messages: config.keepRecentMessages,
-        min_content_tokens: config.minContentTokens,
-        cache_missed_threshold_ms: config.cacheMissedThresholdMs,
-        truncated_marker: config.truncatedMarker,
-        min_context_usage_ratio: config.minContextUsageRatio,
-        truncated_tool_result_count: effect.truncatedToolResultCount,
-        truncated_tool_result_tokens_before: effect.truncatedToolResultTokensBefore,
-        truncated_tool_result_tokens_after: effect.truncatedToolResultTokensAfter,
-        tokens_before: tokensBefore,
-        tokens_after: tokensAfter,
-        previous_cutoff: previousCutoff,
-        cutoff: nextCutoff,
-        message_count: history.length,
-        cache_age_ms: cacheAgeMs,
-        thinking_effort: this.agent.config.thinkingEffort,
-      });
-    }
+    // Micro compaction is disabled: the `micro_compaction` experimental flag has
+    // been removed from the registry, so detection is intentionally a no-op.
+    return;
+
+    // Original implementation (disabled):
+    // if (!this.agent.experimentalFlags.enabled('micro_compaction')) return;
+    //
+    // const config = this.config;
+    // const { history, lastAssistantAt } = this.agent.context;
+    // const cacheAgeMs = lastAssistantAt === null ? null : Date.now() - lastAssistantAt;
+    // const cacheMissed = cacheAgeMs !== null && cacheAgeMs >= config.cacheMissedThresholdMs;
+    // if (!cacheMissed) return;
+    //
+    // const maxContextTokens = this.agent.config.modelCapabilities.max_context_tokens;
+    // const contextTokens = this.agent.context.tokenCountWithPending;
+    // const contextUsageRatio =
+    //   maxContextTokens !== undefined && maxContextTokens > 0
+    //     ? contextTokens / maxContextTokens
+    //     : 1;
+    // if (contextUsageRatio < config.minContextUsageRatio) return;
+    //
+    // const previousCutoff = this.cutoff;
+    // const nextCutoff = Math.max(0, history.length - config.keepRecentMessages);
+    // this.apply(nextCutoff);
+    // if (previousCutoff !== nextCutoff) {
+    //   const effect = this.measureEffect(history, nextCutoff);
+    //   const previousEffect = this.measureEffect(history, previousCutoff);
+    //   const rawContextTokens = estimateTokensForMessages(history);
+    //   // Whole-context length before/after this cutoff change, mirroring the
+    //   // `tokens_before`/`tokens_after` fields on `compaction_finished` so the
+    //   // two compaction paths can be compared on the same axis.
+    //   const tokensBefore =
+    //     rawContextTokens -
+    //     previousEffect.truncatedToolResultTokensBefore +
+    //     previousEffect.truncatedToolResultTokensAfter;
+    //   const tokensAfter =
+    //     rawContextTokens -
+    //     effect.truncatedToolResultTokensBefore +
+    //     effect.truncatedToolResultTokensAfter;
+    //   this.agent.telemetry.track('micro_compaction_finished', {
+    //     keep_recent_messages: config.keepRecentMessages,
+    //     min_content_tokens: config.minContentTokens,
+    //     cache_missed_threshold_ms: config.cacheMissedThresholdMs,
+    //     truncated_marker: config.truncatedMarker,
+    //     min_context_usage_ratio: config.minContextUsageRatio,
+    //     truncated_tool_result_count: effect.truncatedToolResultCount,
+    //     truncated_tool_result_tokens_before: effect.truncatedToolResultTokensBefore,
+    //     truncated_tool_result_tokens_after: effect.truncatedToolResultTokensAfter,
+    //     tokens_before: tokensBefore,
+    //     tokens_after: tokensAfter,
+    //     previous_cutoff: previousCutoff,
+    //     cutoff: nextCutoff,
+    //     message_count: history.length,
+    //     cache_age_ms: cacheAgeMs,
+    //     thinking_effort: this.agent.config.thinkingEffort,
+    //   });
+    // }
   }
 
   compact(messages: readonly ContextMessage[]): readonly ContextMessage[] {
-    if (!this.agent.experimentalFlags.enabled('micro_compaction')) return messages;
-
-    const config = this.config;
-    const result: ContextMessage[] = [];
-    let i = 0;
-    for (const msg of messages) {
-      if (
-        i < this.cutoff &&
-        msg.role === 'tool' &&
-        msg.toolCallId !== undefined &&
-        estimateTokensForContentParts(msg.content) >= config.minContentTokens
-      ) {
-        result.push({
-          ...msg,
-          content: [{ type: 'text', text: config.truncatedMarker } satisfies ContentPart],
-        });
-      } else {
-        result.push(msg);
-      }
-      i++;
-    }
-    return result;
+    // Micro compaction is disabled: the `micro_compaction` experimental flag has
+    // been removed from the registry, so messages are always returned unchanged.
+    return messages;
+
+    // Original implementation (disabled):
+    // if (!this.agent.experimentalFlags.enabled('micro_compaction')) return messages;
+    //
+    // const config = this.config;
+    // const result: ContextMessage[] = [];
+    // let i = 0;
+    // for (const msg of messages) {
+    //   if (
+    //     i < this.cutoff &&
+    //     msg.role === 'tool' &&
+    //     msg.toolCallId !== undefined &&
+    //     estimateTokensForContentParts(msg.content) >= config.minContentTokens
+    //   ) {
+    //     result.push({
+    //       ...msg,
+    //       content: [{ type: 'text', text: config.truncatedMarker } satisfies ContentPart],
+    //     });
+    //   } else {
+    //     result.push(msg);
+    //   }
+    //   i++;
+    // }
+    // return result;
   }
 
   private measureEffect(

packages/agent-core/src/flags/registry.ts

@@ -12,14 +12,17 @@ import type { FlagDefinitionInput } from './types';
  * not equal the master switch 'KIMI_CODE_EXPERIMENTAL_FLAG'; `id` must not be 'flag'.
  */
 export const FLAG_DEFINITIONS = [
-  {
-    id: 'micro_compaction',
-    title: 'Micro compaction',
-    description: 'Trim older large tool results from context while keeping recent conversation intact.',
-    env: 'KIMI_CODE_EXPERIMENTAL_MICRO_COMPACTION',
-    default: false,
-    surface: 'core',
-  },
+  // Micro compaction has been disabled and removed: the capability cannot be
+  // enabled via env, config, or the master experimental switch. The entry is
+  // kept here commented out so it can be restored if the feature is revived.
+  // {
+  //   id: 'micro_compaction',
+  //   title: 'Micro compaction',
+  //   description: 'Trim older large tool results from context while keeping recent conversation intact.',
+  //   env: 'KIMI_CODE_EXPERIMENTAL_MICRO_COMPACTION',
+  //   default: false,
+  //   surface: 'core',
+  // },
 ] as const satisfies readonly FlagDefinitionInput[];
 
 /** Literal union of registered flag ids. */

packages/agent-core/src/flags/types.ts

@@ -26,7 +26,10 @@ export type ExperimentalFlagConfig = Partial<Record<FlagId, boolean>>;
 export type ExperimentalFlagSource = 'master-env' | 'env' | 'config' | 'default';
 
 export interface ExperimentalFeatureState {
-  readonly id: FlagId;
+  /** Feature id. Typed as `string` because this is a runtime snapshot that
+   * crosses the SDK/RPC boundary and must remain usable even when no flags are
+   * registered (in which case the internal `FlagId` union collapses to `never`). */
+  readonly id: string;
   readonly title: string;
   readonly description: string;
   readonly surface: FlagSurface;

packages/agent-core/test/agent/basic.test.ts

@@ -9,7 +9,9 @@ it('creates an independent agent with a scoped experimental flag resolver', () =
     experimentalFlags: new FlagResolver({}, FLAG_DEFINITIONS),
   });
 
-  expect(ctx.agent.experimentalFlags.enabled('micro_compaction')).toBe(false);
+  // No experimental flags are currently registered, so the scoped resolver
+  // reports none enabled.
+  expect(ctx.agent.experimentalFlags.enabledIds()).toEqual([]);
 });
 
 it('runs a text-only agent turn from prompt to completion', async () => {

packages/agent-core/test/agent/compaction/compaction-scenarios.test.ts

@@ -349,7 +349,9 @@ describe('compaction — probe tests (high-risk scenarios)', () => {
   // to a recent suffix but still projected through MicroCompaction.compact()
   // with the cutoff computed for the FULL history. The absolute cutoff applied
   // to the shifted suffix can clear recent tool results the summary needs.
-  it.fails('does not clear recent tool results when projecting a shrunk suffix under an active micro-compaction cutoff', () => {
+  // SKIPPED: micro-compaction has been disabled and its flag removed, so this
+  // defect no longer exists.
+  it.skip('does not clear recent tool results when projecting a shrunk suffix under an active micro-compaction cutoff', () => {
     // This defect only exists when micro-compaction is active, so enable the
     // flag explicitly rather than inheriting the ambient KIMI_CODE_EXPERIMENTAL
     // master switch — otherwise the probe's pass/fail flips with the runner's

packages/agent-core/test/agent/compaction/full.test.ts

@@ -324,7 +324,9 @@ describe('FullCompaction', () => {
     ).toBe(false);
   });
 
-  it('micro-compacts old tool results before sending the summary request', async () => {
+  // Micro compaction is disabled; this scenario is skipped because the feature
+  // can no longer be enabled.
+  it.skip('micro-compacts old tool results before sending the summary request', async () => {
     vi.useFakeTimers();
     enableMicroCompactionFlag();
     const ctx = testAgent({
@@ -2227,11 +2229,9 @@ function enableMicroCompactionFlag(): void {
 }
 
 function getMicroCompactionFlagEnv(): string {
-  const flag = FLAG_DEFINITIONS.find((definition) => definition.id === 'micro_compaction');
-  if (flag === undefined) {
-    throw new Error('Missing micro_compaction flag definition.');
-  }
-  return flag.env;
+  // Micro compaction is disabled and its flag has been removed from the registry;
+  // the env var name is kept so the (skipped) test still type-checks.
+  return 'KIMI_CODE_EXPERIMENTAL_MICRO_COMPACTION';
 }
 
 function deferred<T>() {

packages/agent-core/test/agent/compaction/micro.test.ts

@@ -29,15 +29,18 @@ const MINUTE = 60 * 1000;
 const DEFAULT_MARKER = '[Old tool result content cleared]';
 const MICRO_COMPACTION_FLAG_ENV = getMicroCompactionFlagEnv();
 
-describe('MicroCompaction', () => {
+// Micro compaction is disabled and its flag has been removed; the suite is
+// skipped because the feature can no longer be enabled.
+describe.skip('MicroCompaction', () => {
   beforeEach(() => {
     vi.stubEnv(MASTER_ENV, '0');
     vi.stubEnv(MICRO_COMPACTION_FLAG_ENV, '1');
   });
 
-  it('defaults the micro_compaction flag off', () => {
-    expect(new FlagResolver({}, FLAG_DEFINITIONS).enabled('micro_compaction')).toBe(false);
-  });
+  // The micro_compaction flag no longer exists, so there is no default to assert.
+  // it('defaults the micro_compaction flag off', () => {
+  //   expect(new FlagResolver({}, FLAG_DEFINITIONS).enabled('micro_compaction')).toBe(false);
+  // });
 
   it('truncates old tool results after cache miss', () => {
     vi.useFakeTimers();
@@ -966,11 +969,9 @@ function hasMarker(messages: readonly Message[]): boolean {
 }
 
 function getMicroCompactionFlagEnv(): string {
-  const flag = FLAG_DEFINITIONS.find((definition) => definition.id === 'micro_compaction');
-  if (flag === undefined) {
-    throw new Error('Missing micro_compaction flag definition.');
-  }
-  return flag.env;
+  // Micro compaction is disabled and its flag has been removed from the registry;
+  // the env var name is kept so the (skipped) suite still type-checks.
+  return 'KIMI_CODE_EXPERIMENTAL_MICRO_COMPACTION';
 }
 
 function singleTelemetryEvent(