billion-context-dsh 0.2.12 → 0.2.14

package/README.en.md

@@ -3,7 +3,7 @@
 [English](./README.en.md) | [中文](./README.md)
 
 > **⚠️ Beta notice — not for production use**
-> This project (**v0.2.12**) is a work-in-progress beta. The [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) itself is also in **public beta**. **Do not use either in engineering / production environments** — expect breaking changes and rough edges.
+> This project (**v0.2.14**) is a work-in-progress beta. The [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) itself is also in **public beta**. **Do not use either in engineering / production environments** — expect breaking changes and rough edges.
 
 <p align="center">
 <strong>Built with gratitude on top of these projects</strong> — please give them a ⭐:
@@ -202,9 +202,9 @@ This project reuses `acp-kernel`'s compression core and `billion-context-pi`'s d
 | `modelContextLimit` | auto-detected (fallback `128000`) | Context window used for the kernel's pressure decisions; an explicit value wins and skips the probe |
 | `autoModelContextLimit` | `true` | Probe the model's real window from the model API (`agent.ctx.llm.resolveModelInfo`); fall back to the default on failure, the `/acp` command shows the window source (the `acp_status` model tool carries no window info). A failed probe is surfaced in the host log and the `/acp` panel (`restart to re-probe`) — the failure is cached like a success, so fixing the gateway requires a restart or an explicit `modelContextLimit` before the probe retries |
 | `nudgeMinContextLimitPct` | kernel default `0.45` | Nudge window lower bound (usage fraction) — validation only; the growth-driven trigger has no percentage floor — same default as billion-context-pi |
-| `nudgeMaxContextLimitPct` | engine default `0.70` (kernel/pi default `0.75`) | Over-limit line: above this the nudge fires regardless of growth — deliberately below the host compaction-basic 80% auto-compaction line so the forced nudge fires first; an explicit value wins |
-| `nudgeEmergencyThresholdPct` | engine default `0.85` (kernel/pi default `0.95`) | Emergency nudge (bypasses the per-turn dedup) — lowered from `0.95`: at 95% the model has no room to act and the 80% auto-compaction line shadows it; an explicit value wins |
-| `coreOverrides` | — | Any other acp-kernel `Config` override (billion-context-pi's `coreOverrides` escape hatch) |
+| `nudgeMaxContextLimitPct` | engine default `0.70` (kernel/pi default `0.75`) | Over-limit line: above this the nudge fires regardless of growth — deliberately below the host compaction-basic 80% auto-compaction line so the forced nudge fires first; an explicit value wins (a same-name key in `coreOverrides.nudge` outranks it — see below) |
+| `nudgeEmergencyThresholdPct` | engine default `0.85` (kernel/pi default `0.95`) | Emergency nudge (bypasses the per-turn dedup) — lowered from `0.95`: at 95% the model has no room to act and the 80% auto-compaction line shadows it; an explicit value wins (a same-name key in `coreOverrides.nudge` outranks it — see below) |
+| `coreOverrides` | — | Any other acp-kernel `Config` override (billion-context-pi's `coreOverrides` escape hatch). Merge order: kernel defaults → top-level pct knobs → `coreOverrides.nudge` lands last — same-name keys take its value |
 | `autoTools` | `true` | Register the four model tools on `ctx.tools` |
 | `autoCommand` | `true` | Register the `/acp` command on `ctx.commands` |
 | `autoNudge` | `true` | Inject the nudge into `agent/pre-step` |

package/README.md

@@ -3,7 +3,7 @@
 [中文](./README.md) | [English](./README.en.md)
 
 > **⚠️ 测试版声明——请勿用于生产环境**
-> 本项目（**v0.2.12**）仍处于开发中的测试版。[DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) 本身也处于**公开测试版**阶段。**请勿将两者用于工程化 / 生产环境**——预期会有破坏性变更与粗糙之处。
+> 本项目（**v0.2.14**）仍处于开发中的测试版。[DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) 本身也处于**公开测试版**阶段。**请勿将两者用于工程化 / 生产环境**——预期会有破坏性变更与粗糙之处。
 
 <p align="center">
 <strong>衷心感谢以下项目——请给它们一个 ⭐：</strong>
@@ -202,9 +202,9 @@ DSH 的每个模型请求都派生自其 append-only 会话日志（*surface*）
 | `modelContextLimit` | 自动探测（回退 `128000`） | 用于内核压力决策的上下文窗口；显式配置时优先且跳过探测 |
 | `autoModelContextLimit` | `true` | 从模型 API 自动探测真实窗口（`agent.ctx.llm.resolveModelInfo`）；探测失败回退默认值，`/acp` 命令展示窗口来源（模型工具 `acp_status` 不含窗口信息）。探测失败会在宿主日志与 `/acp` 面板提示（`restart to re-probe`）——失败结果同样被缓存，修复网关后需重启或显式设置 `modelContextLimit` 才会重新探测 |
 | `nudgeMinContextLimitPct` | 内核默认 `0.45` | Nudge 窗口下界（用量占比）——仅作配置校验，增长路径的触发没有百分比下限——与 billion-context-pi 相同的默认值 |
-| `nudgeMaxContextLimitPct` | engine 默认 `0.70`（内核/pi 默认 `0.75`） | 过限线：超过此值则无论增长与否都触发 nudge——刻意低于宿主 compaction-basic 的 80% 自动压缩线，保证强制 nudge 先触发；显式配置优先 |
-| `nudgeEmergencyThresholdPct` | engine 默认 `0.85`（内核/pi 默认 `0.95`） | 紧急 nudge（绕过每轮去重）——从 `0.95` 下调：95% 时模型已无操作空间且会被 80% 自动压缩线遮蔽；显式配置优先 |
-| `coreOverrides` | — | 任何其他 acp-kernel `Config` 覆盖（billion-context-pi 的 `coreOverrides` 逃生口） |
+| `nudgeMaxContextLimitPct` | engine 默认 `0.70`（内核/pi 默认 `0.75`） | 过限线：超过此值则无论增长与否都触发 nudge——刻意低于宿主 compaction-basic 的 80% 自动压缩线，保证强制 nudge 先触发；显式配置优先（`coreOverrides.nudge` 同名键优先级更高，见下） |
+| `nudgeEmergencyThresholdPct` | engine 默认 `0.85`（内核/pi 默认 `0.95`） | 紧急 nudge（绕过每轮去重）——从 `0.95` 下调：95% 时模型已无操作空间且会被 80% 自动压缩线遮蔽；显式配置优先（`coreOverrides.nudge` 同名键优先级更高，见下） |
+| `coreOverrides` | — | 任何其他 acp-kernel `Config` 覆盖（billion-context-pi 的 `coreOverrides` 逃生口）。合并顺序：内核默认 → 顶层 pct 配置 → `coreOverrides.nudge` 最后落地——同名键以它为准 |
 | `autoTools` | `true` | 在 `ctx.tools` 注册四个模型工具 |
 | `autoCommand` | `true` | 在 `ctx.commands` 注册 `/acp` 命令 |
 | `autoNudge` | `true` | 当内核建议时向 `agent/pre-step` 注入 nudge |

package/dist/index.d.ts

@@ -30,6 +30,7 @@ import { CompactionEngine, type CompactionAgentContext, type CompactionResult, t
 import { type CompressionCore } from 'acp-kernel';
 import type { Agent } from '@deepseek-ai/dsh-agent';
 import { AcpStateStore } from './state.ts';
+import { type ToolEnvironment } from './tools.ts';
 import { type AcpPrompts, type ResolvedPrompts } from './prompts.ts';
 import { type AcpWindow } from './window.ts';
 export { AcpStateStore } from './state.ts';
@@ -59,7 +60,8 @@ export interface AcpConfig {
      * kernel injects a nudge regardless of growth or cadence. Engine default
      * 0.70 (deliberately BELOW the kernel/billion-context-pi default 0.75 and
      * the host compaction-basic auto-compaction line 0.80, so the forced nudge
-     * always fires first); an explicit value wins.
+     * always fires first); an explicit value wins over this default — a
+     * same-name key in `coreOverrides.nudge` wins over both (it merges last).
      */
     readonly nudgeMaxContextLimitPct?: number;
     /**
@@ -69,7 +71,11 @@ export interface AcpConfig {
      * compaction-basic line shadows it in standard/code/cordis modes).
      */
     readonly nudgeEmergencyThresholdPct?: number;
-    /** Any other acp-kernel Config override (billion-context-pi's `coreOverrides` escape hatch). */
+    /**
+     * Any other acp-kernel Config override (billion-context-pi's `coreOverrides`
+     * escape hatch). Merge order per section: kernel defaults → the engine pct
+     * knobs above → these keys land LAST, so a same-name key here wins.
+     */
     readonly coreOverrides?: Partial<import('acp-kernel').Config>;
     /**
      * Custom token-count function for the kernel's internal estimation.
@@ -107,6 +113,14 @@ export declare class AcpCompactionEngine extends CompactionEngine {
     readonly config: AcpConfig;
     /** Resolved prompt templates (validated at construction — fail-fast on template typos). */
     readonly prompts: ResolvedPrompts;
+    /**
+     * The environment wired into tools / command / nudge. Exposed so tests (and
+     * introspection) can assert the forwarding actually happened: the config
+     * chain user config → this.config → env → kernelConfigFor is all OPTIONAL
+     * fields, so a dropped forwarding line fails typecheck silently and would
+     * revive lost-config bugs with every unit test green.
+     */
+    readonly env: ToolEnvironment;
     private readonly lastNudgeTurn;
     /** Successful compress call ids awaiting their tool/result so the pair can be hidden. */
     private readonly compressCallIdsToHide;

package/dist/index.js

@@ -3502,7 +3502,7 @@ var AcpStateStore = class {
 };
 
 // src/tools.ts
-import { defineTool } from "@deepseek-ai/dsh-tools";
+import { defineTool, ToolArgsError } from "@deepseek-ai/dsh-tools";
 
 // src/config.ts
 function kernelConfigFor(input) {
@@ -3511,8 +3511,12 @@ function kernelConfigFor(input) {
   if (input.nudgeMaxContextLimitPct !== void 0) nudgePatch.maxContextLimitPct = input.nudgeMaxContextLimitPct;
   if (input.nudgeEmergencyThresholdPct !== void 0) nudgePatch.emergencyThresholdPct = input.nudgeEmergencyThresholdPct;
   const overrides = { ...input.coreOverrides };
-  if (Object.keys(nudgePatch).length > 0) {
-    overrides.nudge = { ...defaultConfig(input.modelContextLimit).nudge, ...nudgePatch };
+  if (Object.keys(nudgePatch).length > 0 || input.coreOverrides?.nudge) {
+    overrides.nudge = {
+      ...defaultConfig(input.modelContextLimit).nudge,
+      ...nudgePatch,
+      ...input.coreOverrides?.nudge
+    };
   }
   return defaultConfig(input.modelContextLimit, overrides);
 }
@@ -3851,6 +3855,16 @@ var compressParameters = {
   // intentionally NOT `required: true` — a required property would reject the
   // wrapped shape before `handleCompress` can see it. The tool description
   // still tells the model content is mandatory.
+  //
+  // The items fields are the opposite case: startSeq/endSeq/summary MUST be
+  // `required: true`. Without that, a model call that omits `summary` (only
+  // startSeq/endSeq/topic present) passed schema validation and failed late
+  // inside the kernel with "Summary is empty" — and live sessions showed the
+  // model retrying the identical broken call in a loop. With the fields
+  // required, the same call is rejected at the schema gate with
+  // `missing required property "content[0].summary"`, which tells the model
+  // exactly which field to add (same pattern as decompress's required
+  // blockId / search_context's required query).
   arguments: { type: "json", description: "Tolerated wrapped-arguments form (model-generated); unwrapped in handleCompress. Prefer passing content directly." },
   topic: { type: "string", description: "Fallback topic for entries without their own." },
   content: {
@@ -3860,18 +3874,20 @@ var compressParameters = {
       type: "object",
       properties: {
         startSeq: {
+          required: true,
           oneOf: [
             { type: "integer", description: "First surface seq of the range." },
             { type: "string", description: "Seq as text; a trailing #callId fragment is ignored." }
           ]
         },
         endSeq: {
+          required: true,
           oneOf: [
             { type: "integer", description: "Inclusive last surface seq of the range." },
             { type: "string", description: "Seq as text; a trailing #callId fragment is ignored." }
           ]
         },
-        summary: { type: "string", description: "Complete technical summary replacing the range; keep paths, decisions, values verbatim. Minimum 50 characters." },
+        summary: { type: "string", required: true, description: "Complete technical summary replacing the range; keep paths, decisions, values verbatim. Minimum 50 characters." },
         topic: { type: "string", description: "Short label (3-5 words) for this range." }
       },
       additionalProperties: false
@@ -3942,6 +3958,18 @@ function unwrapEnvelope(args) {
   if (typeof inner !== "object" || inner === null || Array.isArray(inner)) return args;
   return { ...args, ...inner };
 }
+function validateContentItems(content) {
+  const violations = [];
+  content.forEach((item, index) => {
+    const path = `content[${index}]`;
+    if (item.startSeq === void 0) violations.push(`missing required property "${path}.startSeq"`);
+    if (item.endSeq === void 0) violations.push(`missing required property "${path}.endSeq"`);
+    if (typeof item.summary !== "string" || item.summary.trim().length === 0) {
+      violations.push(`missing required property "${path}.summary"`);
+    }
+  });
+  if (violations.length > 0) throw new ToolArgsError(violations);
+}
 async function handleCompress(env, args, exec) {
   const agent = requireAgent(exec);
   const session = agent.session;
@@ -3963,6 +3991,7 @@ async function handleCompress(env, args, exec) {
     };
   }
   args = unwrapped;
+  validateContentItems(args.content);
   const ranges = [];
   const alreadyCompressedNotes = [];
   for (const range of args.content) {
@@ -4438,7 +4467,9 @@ var DEFAULT_CONFIG = {
   // kernel/billion-context-pi 0.75/0.95. 0.95 leaves no room to act before
   // the API rejects, and the host's compaction-basic line (thresholdRatio
   // 0.80) shadows it in standard/code/cordis modes; 0.70 keeps the forced
-  // over-limit nudge ahead of that 80% line. Explicit values always win.
+  // over-limit nudge ahead of that 80% line. Explicit values always win
+  // against these defaults — `coreOverrides` merges last and beats them on
+  // same-name keys.
   nudgeMaxContextLimitPct: 0.7,
   nudgeEmergencyThresholdPct: 0.85
 };
@@ -4454,6 +4485,14 @@ var AcpCompactionEngine = class extends CompactionEngine {
   config;
   /** Resolved prompt templates (validated at construction — fail-fast on template typos). */
   prompts;
+  /**
+   * The environment wired into tools / command / nudge. Exposed so tests (and
+   * introspection) can assert the forwarding actually happened: the config
+   * chain user config → this.config → env → kernelConfigFor is all OPTIONAL
+   * fields, so a dropped forwarding line fails typecheck silently and would
+   * revive lost-config bugs with every unit test green.
+   */
+  env;
   lastNudgeTurn = /* @__PURE__ */ new Map();
   /** Successful compress call ids awaiting their tool/result so the pair can be hidden. */
   compressCallIdsToHide = /* @__PURE__ */ new Set();
@@ -4479,6 +4518,7 @@ var AcpCompactionEngine = class extends CompactionEngine {
       prompts: this.prompts,
       compressCallIdsToHide: this.compressCallIdsToHide
     };
+    this.env = env;
     const tools = ctx.get("tools");
     if (tools !== void 0) {
       for (const tool of makeTools(env)) tools.register(tool);