dsh-loop-engine-allinone 0.7.3 → 0.7.5

package/CHANGELOG.md

@@ -8,14 +8,85 @@ adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
 Nothing yet.
 
+## [0.7.5]
+
+### Added
+
+- **Coverage is a ratchet now, and the ratchet is at 100%.** `scripts/coverage-gate.mjs`
+  replaces `vitest.config.ts`'s `thresholds.perFile`, which had never passed on this
+  branch: files the change touches must be at 100% per metric, every other file may not
+  fall below `.coverage-baseline.json`. The gate reads the istanbul `json` report, diffs
+  against the base commit, and refuses a lowered floor in the same change that needs it
+  (the baseline is regenerated deliberately, in its own commit, with a reason).
+- The 38 files the old threshold flagged are all at per-file 100% now (1,613 → **1,817
+  tests**, 96 files); `.coverage-baseline.json` was rewritten to 119 × 100%, so "do not
+  regress" now means the same thing as "must be full".
+
+### Fixed
+
+- **A profile install no longer writes its managed block into `web`.** With
+  `config.profile` omitted, `resolvePatchPath()` defaulted to `web`; a market install into
+  the Desktop app's `desktop` profile therefore disabled `agent-loop` in a profile that
+  had no plugin, while `desktop` kept the base loop in the agent-factory slot and every
+  per-session engine switch answered `router-unmounted` ("the loop engine router is not
+  ready"). The profile is now resolved from the launcher's `profileContext`, then from the
+  directory the package was installed into, and only then from `web`. An explicit
+  `profile:` entry still wins (needed for a `link:`/source checkout).
+- **Two preset-authoring tests could flake on a loaded runner.** They polled for one second while the plugin's own roster wait is bounded at two; a docs-only push (run `39118609901`) failed both. The helper now polls for the full eight-second budget, which is what the assertion means. (Tests only.)
+- **Coverage no longer depends on the host's home directory.** `model-registry.ts` sat at
+  93.94% on a clean CI runner while reading 100% locally, because the local machine has a
+  real `~/.kimi-code/config.toml` and the tests silently relied on it. The spec now points
+  `KIMI_CODE_HOME` at a fixture for the discovered arm and at an empty directory for the
+  "not installed" arm, so both run everywhere. (Tests only; runtime behaviour unchanged.)
+
+### Changed
+
+- Coverage-only source edits are comments at genuinely unreachable arms (`/* v8 ignore
+  start|stop */` plus a reason: POSIX-only ternaries, registry-constant branches, the
+  production-layout early return). Runtime behaviour is unchanged.
+
+## [0.7.4]
+
+### Fixed
+
+- **`0.7.3` could not be installed on DeepSeek Harness `0.2.1`.** `dsh` refuses a
+  plugin whose declared peers the running version does not satisfy, and two things
+  broke at once. The union range ended at `<0.2.1-0`, so every `0.2.1-rc.N` host fell
+  outside it; and `@deepseek-ai/dsh-invariants` — declared as a hard peer — was never
+  published for that line and is not part of the `0.2.1` runtime at all. Both are
+  fixed: the last segment is now `>=0.2.0-rc.1 <0.2.2-0` (one range covers `0.2.0` and
+  `0.2.1`, which kept the same API shape), and the invariants peer is marked
+  `optional`, which is what it always was in practice — the `./invariant` companion
+  imports it for types only and the built bundle contains no runtime import of it.
+
+### Added
+
+- **The `0.2.1` line is a supported generation**, verified end to end: an isolated
+  `.compat/021/` dependency set plus `vitest.config.compat021.ts` run the same 1,613
+  specs, `tsconfig.compat021.json` typechecks the sources against `0.2.1`'s published
+  declarations, and a real `dsh web` on `0.2.1-rc.1` installs the plugin, writes its
+  managed block once (byte-identical on reboot), publishes all ten engine presets,
+  and serves the client bundle.
+- `npm run test:compat` and `npm run compat:install` cover all four generations, and
+  `npm run typecheck:compat` is the `0.2.1`-against-isolated-declarations check.
+
+### Changed
+
+- `devDependencies` moved to the `0.2.1` line; `schemastery` to `3.18.5`,
+  `cordis` to `4.0.5`.
+- The test suite mounts the JSONL durability backend through the new
+  `tests/helpers/jsonl-persistence.ts`. `0.2.1` replaced that package's default
+  export with named `{ name, apply }` exports, so the direct default import the specs
+  used resolved to `undefined` and cordis rejected the mount with
+  `invalid plugin, expect function or object with an "apply" method`.
+
 ## [0.7.3]
 
 ### Fixed
 
 - **ZCode's path probe only worked on the author's machine.** The resolver and the
   probe script both hardcoded `D:\Programs\ZCode\...` (and the probe also
-  `D:\Programs
-pm-global`), so on any other host the "well-known location" check
+  `D:\Programs\npm-global`), so on any other host the "well-known location" check
   silently never matched and a user with ZCode installed somewhere normal was told it
   was not installed. Both now derive candidates from the environment:
   `%ZCODE_HOME%` → `%LOCALAPPDATA%\Programs\ZCode` → `%ProgramFiles%\ZCode`, and

package/README.md

@@ -3,7 +3,7 @@
 **English** · [简体中文](README.zh.md)
 
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
-[![Harness: 0.1.5 | 0.1.7 | 0.2.0](https://img.shields.io/badge/harness-0.1.5%20%7C%200.1.7%20%7C%200.2.0-6f42c1)](#version-compatibility)
+[![Harness: 0.1.5 | 0.1.7 | 0.2.0 | 0.2.1](https://img.shields.io/badge/harness-0.1.5%20%7C%200.1.7%20%7C%200.2.0%20%7C%200.2.1-6f42c1)](#version-compatibility)
 
 **dsh-loop-engine-allinone** is an all-in-one agent loop engine switcher for **[DeepSeek Harness](https://github.com/deepseek-ai)** (`dsh`). It enables per-session engine switching across **10 hosted engines** (Claude Code, Codex, Pi, Kimi Code, Antigravity, ZCode, Qoder CN, Hermes Agent, Grok Build, MiniMax Code) alongside the built-in `in-process` loop.
 
@@ -57,25 +57,32 @@ dsh plugin --profile web add ./dsh-loop-engine-allinone
 
 `web` is the profile the harness's own CLI documents (`dsh web` is short for
 `dsh --profile web`), so it is the right default. But the Desktop app boots
-`desktop`, a CI deployment may boot `headless`, and you may have made your own. In
-that case:
-
-1. install with that profile's name — `dsh plugin --profile desktop add
-   dsh-loop-engine-allinone`;
-2. **and** set the same name in the composition entry:
-
-   ```yaml
-   - id: loop-engine
-     config:
-       profile: desktop
-   ```
-
-Step 2 is the one that bites. The field defaults to `web`, so omitting it makes the
-plugin write its managed block into `profiles/web/cordis.patch.yml` while the profile
-you actually boot never reads it: the plugin mounts, but the base `agent-loop` row is
-never disabled, the router cannot take the agent-factory slot, and **no hosted engine
-is reachable** — with nothing logged. This release also warns when it detects that
-shape, naming the field to set.
+`desktop`, a CI deployment may boot `headless`, and you may have made your own.
+
+**Installing with `dsh plugin --profile desktop add dsh-loop-engine-allinone` is
+enough**: the plugin resolves the profile it is actually serving — the launcher's
+`profileContext` when the generation provides it, otherwise the profile directory
+it was installed into (the nearest ancestor with a `dsh.profile` manifest). Only a
+`link:`/source checkout has neither, and there the composition entry still wins:
+
+```yaml
+- id: loop-engine
+  config:
+    profile: desktop      # only needed for a linked checkout / global install
+```
+
+Before 0.7.5 the field defaulted to `web` and nothing else was consulted. A market
+install into `desktop` then wrote the managed block into `profiles/web/`, the
+`desktop` base `agent-loop` kept the agent-factory slot, and every per-session
+engine switch answered `router-unmounted` ("the loop engine router is not ready")
+— with nothing logged. The same omission could disable `agent-loop` in a `web`
+profile that had no plugin at all. 0.7.5 fixes both ends: it finds the right
+profile, and it warns when the resolved profile has no `package.json` beside it.
+
+If 0.7.4 already wrote the managed block into `web` while you did not install the
+plugin there, delete that block (from `# -- dsh-loop-engine managed block --` through
+`# -- /dsh-loop-engine managed block --`, plus the preset block) before booting
+`web`; otherwise that profile has no agent factory.
 
 ### Installation notes
 
@@ -85,9 +92,35 @@ shape, naming the field to set.
   delete the managed `loop-engine` block** from that profile's `cordis.patch.yml`. The
   block outlives the plugin, and while it is present the profile has no agent factory,
   so no session can be created.
-- **pnpm 10+** may block dependency build scripts (`ERR_PNPM_IGNORED_BUILDS` naming
-  `esbuild`, `@google/genai`, `protobufjs`). Run `pnpm approve-builds`, or add an
-  `allowBuilds` entry, and retry — only the installing project can grant this.
+- **One-time script approval.** pnpm 10+ refuses to run dependency install scripts
+  until the installing project approves them, so a fresh `dsh plugin add` can stop at
+  `ERR_PNPM_IGNORED_BUILDS` naming `@google/genai` and `protobufjs`. **This is not a
+  missing dependency and there is nothing to install by hand** — the whole tree,
+  including the bundled Pi engine, is already on disk; pnpm is only asking whether to
+  execute two scripts those packages ship:
+
+  | Package | Script | What it actually does |
+  |---|---|---|
+  | `@google/genai` | `preinstall` | `echo 'preinstall: no-op'` |
+  | `protobufjs` | `postinstall` | reads its own `package.json`, prints a version-scheme warning |
+
+  Both are no-ops, so **either answer is safe**: approving runs them, and
+  `allowBuilds: false` skips them. Verified for the second case — with both set to
+  `false`, `@google/genai` loads (145 exports), `protobufjs` loads, and Pi's
+  `dist/cli.js` is present. They arrive only through the bundled Pi engine;
+  Claude Code and Codex ship no install scripts, so this prompt never appears
+  because of them.
+
+  The question is asked **once per profile** — the answer is written into that
+  profile's `pnpm-workspace.yaml` and later installs and upgrades do not ask again.
+  To pre-answer it before the first install, add to the profile's
+  `pnpm-workspace.yaml`:
+
+  ```yaml
+  allowBuilds:
+    '@google/genai': false
+    protobufjs: false
+  ```
 - **Harness from source**: bridge the profile's harness peers with `file:` shims, or
   session resume fails with `agent-presets: refusing to compose an unscoped context`.
   Steps in [docs/source-checkout.md](docs/source-checkout.md).
@@ -193,11 +226,16 @@ Both tables enforce compile-time exhaustiveness checks — incomplete engine reg
 
 A single release seamlessly supports multiple DeepSeek Harness generations via runtime detection:
 
-| Plugin Release | Supported DeepSeek Harness Line |
+| Plugin release | Supported DeepSeek Harness |
 |---|---|
-| `0.2.0-rcX` | `0.2.0-rc.1`, `0.2.0-rc.2` |
-| `0.1.7-rcX` | `0.1.7-rc.1`, `0.1.7-rc.2` |
-| `0.1.5-rcX` | `0.1.5-rc.1`, `0.1.5-rc.2` |
+| `0.7.4`+ | `0.2.0-rc.1` … `0.2.1-rc.1` |
+| `0.7.0` – `0.7.3` | `0.1.5-rc.1` … `0.2.0-rc.2` |
+
+The union range is `>=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0`.
+`0.2.0` and `0.2.1` share the last segment because the harness kept the same API shape
+across them, so one range covers both. `dsh`'s compatibility check reads these peers
+and refuses to load a plugin that misses one, which is what made `0.7.3` uninstallable
+on `0.2.1`.
 
 See [docs/compatibility.md](docs/compatibility.md) for harness upgrade checklists and API surface diffs.
 

package/README.zh.md

@@ -3,7 +3,7 @@
 [English](README.md) · **简体中文**
 
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
-[![Harness: 0.1.5 | 0.1.7 | 0.2.0](https://img.shields.io/badge/harness-0.1.5%20%7C%200.1.7%20%7C%200.2.0-6f42c1)](#版本兼容性)
+[![Harness: 0.1.5 | 0.1.7 | 0.2.0 | 0.2.1](https://img.shields.io/badge/harness-0.1.5%20%7C%200.1.7%20%7C%200.2.0%20%7C%200.2.1-6f42c1)](#版本兼容性)
 
 **dsh-loop-engine-allinone** 是专为 [DeepSeek Harness](https://github.com/deepseek-ai)（`dsh`）打造的 Agent 循环引擎全功能切换插件。支持**按会话独立切换** 10 家主流外部引擎与 Harness 原生 `in-process` 循环。
 
@@ -79,9 +79,30 @@ dsh plugin --profile web add ./dsh-loop-engine-allinone
 - **卸载**：`dsh plugin --profile web remove dsh-loop-engine-allinone`，**然后手动删掉**
   该 profile `cordis.patch.yml` 里的 `loop-engine` 受管块。块会比插件活得久，而它还在时
   profile 里没有任何 agent factory，会话一条也建不出来。
-- **pnpm 10+** 可能拦下依赖的 build script（`ERR_PNPM_IGNORED_BUILDS`，列出 `esbuild`、
-  `@google/genai`、`protobufjs`）。执行 `pnpm approve-builds`，或加一条 `allowBuilds`，然后重试
-  ——只有安装方能授予这个权限。
+- **一次性脚本授权**。pnpm 10+ 默认不执行依赖包自带的安装脚本，要由安装方批准，
+  所以首次 `dsh plugin add` 可能停在 `ERR_PNPM_IGNORED_BUILDS`，点名 `@google/genai` 与
+  `protobufjs`。**这不是缺依赖，也没有任何东西要你手动装**——整棵树（含捆绑的 Pi 引擎）
+  已经落盘，pnpm 只是在问要不要执行这两个包自带的脚本：
+
+  | 包 | 脚本 | 实际做的事 |
+  |---|---|---|
+  | `@google/genai` | `preinstall` | `echo 'preinstall: no-op'` |
+  | `protobufjs` | `postinstall` | 读自己的 `package.json`，打印一行版本方案警告 |
+
+  两个都是空操作，**所以批准或拒绝都安全**：批准就执行，`allowBuilds: false` 就跳过。
+  拒绝这条路径已实测——两者设为 `false` 后，`@google/genai` 正常加载（145 个导出）、
+  `protobufjs` 正常加载、Pi 的 `dist/cli.js` 也在。这两个包**只由捆绑的 Pi 引擎带进来**；
+  Claude Code 与 Codex 不带安装脚本，永远不会因为它们弹这个窗。
+
+  **每个 profile 只问一次**——答案写进该 profile 的 `pnpm-workspace.yaml`，
+  之后的安装与升级都不再问。想在首次安装前就预先回答，往该 profile 的
+  `pnpm-workspace.yaml` 加：
+
+  ```yaml
+  allowBuilds:
+    '@google/genai': false
+    protobufjs: false
+  ```
 - **用源码启动 harness** 时，要用 `file:` shim 把 profile 的 harness peer 包桥接到 checkout 源码，
   否则恢复会话会报 `agent-presets: refusing to compose an unscoped context`。
   步骤见 [docs/source-checkout.md](docs/source-checkout.md)。
@@ -181,11 +202,14 @@ dsh plugin --profile web add ./dsh-loop-engine-allinone
 
 单一版本通过动态探测，全面兼容 DeepSeek Harness 多代版本：
 
-| 插件版本系列 | 兼容的 DeepSeek Harness 版本 |
+| 插件版本 | 兼容的 DeepSeek Harness |
 |---|---|
-| `0.2.0-rcX` | `0.2.0-rc.1`, `0.2.0-rc.2` |
-| `0.1.7-rcX` | `0.1.7-rc.1`, `0.1.7-rc.2` |
-| `0.1.5-rcX` | `0.1.5-rc.1`, `0.1.5-rc.2` |
+| `0.7.4` 起 | `0.2.0-rc.1` … `0.2.1-rc.1` |
+| `0.7.0` – `0.7.3` | `0.1.5-rc.1` … `0.2.0-rc.2` |
+
+并集范围是 `>=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0`。
+`0.2.0` 与 `0.2.1` 共用末段，因为 harness 在这两代之间没有改 API 形状，一段范围就够了。
+`dsh` 的兼容性判定读的就是这些 peer，差一条就拒绝加载——这正是 `0.7.3` 在 `0.2.1` 上装不上的原因。
 
 版本号 `<harness 线>-rcN` 中的 `rcN` 为本插件对应分支的迭代发布号。详见 [docs/compatibility.md](docs/compatibility.md)。
 

package/lib/index.js

@@ -92,7 +92,8 @@ var __callDispose = (stack, error, hasError) => {
 import { existsSync as existsSync8, readFileSync as readFileSync5 } from "node:fs";
 import { mkdir as mkdir2, readFile as readFile6, rename as rename2, writeFile as writeFile2 } from "node:fs/promises";
 import { randomUUID as randomUUID3 } from "node:crypto";
-import { dirname as dirname7, join as join22 } from "node:path";
+import { basename, dirname as dirname7, join as join22 } from "node:path";
+import { fileURLToPath as fileURLToPath2 } from "node:url";
 import z12 from "@deepseek-ai/schemastery";
 import { resolveDshHome as resolveDshHome2 } from "@deepseek-ai/dsh-home-paths";
 
@@ -8548,8 +8549,10 @@ var AgyAgent = class extends CliHostedAgent {
         stdin: agyStdinPrompt(prompt),
         map: (document) => mapAgyDocument(document, accept, run)
       },
+      /* v8 ignore start -- the mapper pushes frames straight to `accept` rather than returning them, so this collector is never called; the run signature requires one */
       () => {
       }
+      /* v8 ignore stop */
     );
     if (result.parsed === 0) {
       accept({
@@ -8887,7 +8890,9 @@ function mapStreamJsonDocument(document, accept, run) {
       const subtype = typeof frame2.subtype === "string" ? frame2.subtype : void 0;
       const isError = frame2.is_error === true || subtype !== void 0 && subtype !== "success";
       if (isError) {
-        const message = typeof frame2.result === "string" && frame2.result !== "" ? frame2.result : typeof frame2.error === "string" && frame2.error !== "" ? frame2.error : `the engine ended the run with "${subtype ?? "an error"}"`;
+        let message = `the engine ended the run with "${subtype ?? "an error"}"`;
+        if (typeof frame2.result === "string" && frame2.result !== "") message = frame2.result;
+        else if (typeof frame2.error === "string" && frame2.error !== "") message = frame2.error;
         run.reportedError = message;
         accept({ kind: "error", message });
       }
@@ -8984,8 +8989,10 @@ var QoderclicnAgent = class extends CliHostedAgent {
     };
     const result = await runNdjson(
       { transport, map: (document) => mapDocument(document, accept, run) },
+      /* v8 ignore start -- the mapper pushes frames straight to `accept` rather than returning them, so this collector is never called; the run signature requires one */
       () => {
       }
+      /* v8 ignore stop */
     );
     if (result.parsed === 0) {
       this.sink.accept({
@@ -9172,9 +9179,10 @@ function resolveOnPath(name2, env = process.env) {
   return void 0;
 }
 function npmPackageEntry(prefix, pkg, entry) {
+  const slash = pkg.indexOf("/");
   const scoped = pkg.startsWith("@");
-  const scope = scoped ? pkg.slice(0, pkg.indexOf("/")) : "";
-  const name2 = scoped ? pkg.slice(pkg.indexOf("/") + 1) : pkg;
+  const scope = scoped && slash > 0 ? pkg.slice(0, slash) : "";
+  const name2 = scoped ? pkg.slice(slash + 1) : pkg;
   if (name2 === "" || scoped && scope === "") return void 0;
   const relative = scoped ? join10("node_modules", scope, name2, entry) : join10("node_modules", name2, entry);
   for (const candidate of [join10(prefix, relative), join10(prefix, "lib", relative)]) {
@@ -11394,12 +11402,16 @@ var Config11 = z12.object({
   // Derived from the registry's own capability declaration rather than restated:
   // a permission mode an engine declares is a mode the entry accepts, so the two
   // cannot drift.
+  /* v8 ignore start -- the registry row always declares its permission modes, so the empty fallback is a type-level guard */
   permissionMode: z12.union((ENGINE_REGISTRY["claude-code"].capabilities.permissionModes ?? []).map((mode) => z12.const(mode))),
+  /* v8 ignore stop */
   env: z12.dict(z12.string()),
   model: z12.string(),
   disposeGraceMs: z12.number(),
   maxTurns: z12.number(),
+  /* v8 ignore start -- the registry row always declares them, so the empty fallback is a type-level guard */
   sandboxMode: z12.union((ENGINE_REGISTRY.codex.capabilities.permissionModes ?? []).map((mode) => z12.const(mode))),
+  /* v8 ignore stop */
   approvalPolicy: z12.union(CODEX_APPROVAL_POLICIES.map((policy) => z12.const(policy))),
   piProvider: z12.string(),
   piThinking: z12.string(),
@@ -11437,15 +11449,37 @@ var Config11 = z12.object({
     showEngineBadge: LOOP_ENGINE_SHOW_ENGINE_BADGE_SCHEMA
   }
 });
-function resolvePatchPath(config) {
+function resolvePatchPath(config, fallbackProfile = "web") {
   if (config.patchPath !== void 0 && config.patchPath !== "") return config.patchPath;
   return join22(
     resolveDshHome2(),
     "profiles",
-    config.profile ?? "web",
+    config.profile ?? fallbackProfile,
     config.patchFilename ?? "cordis.patch.yml"
   );
 }
+function inferProfileName(entry = fileURLToPath2(import.meta.url)) {
+  let dir = dirname7(entry);
+  for (; ; ) {
+    const manifest = join22(dir, "package.json");
+    if (existsSync8(manifest)) {
+      try {
+        const parsed = JSON.parse(readFileSync5(manifest, "utf8"));
+        if (parsed.dsh?.profile !== void 0) return basename(dir);
+      } catch {
+      }
+    }
+    const parent = dirname7(dir);
+    if (parent === dir) return void 0;
+    dir = parent;
+  }
+}
+function resolveProfileName(ctx, config) {
+  if (config.profile !== void 0 && config.profile !== "") return config.profile;
+  const fromContext = ctx.get("profileContext")?.name;
+  if (typeof fromContext === "string" && fromContext !== "") return fromContext;
+  return inferProfileName() ?? "web";
+}
 function resolveSummarizerTarget(config) {
   const provider = config.summarizerProvider;
   const model = config.summarizerModel;
@@ -11497,18 +11531,18 @@ function readPatchFileSync(path) {
     throw error;
   }
 }
-function warnIfProfileLooksWrong(ctx, config, patchPath) {
+function warnIfProfileLooksWrong(ctx, profileName, patchPath) {
   const profileDir = dirname7(patchPath);
   if (existsSync8(join22(profileDir, "package.json"))) return;
-  const named = config.profile ?? "web";
   ctx.logger.warn(
-    `loop-engine: profile "${named}" has no package.json beside "${patchPath}". If you boot a different profile, set \`profile: <that-name>\` in this plugin's composition entry - otherwise its managed block is written here, the profile you actually boot never reads it, and no hosted engine will be reachable.`
+    `loop-engine: profile "${profileName}" has no package.json beside "${patchPath}". If you boot a different profile, set \`profile: <that-name>\` in this plugin's composition entry - otherwise its managed block is written here, the profile you actually boot never reads it, and no hosted engine will be reachable.`
   );
 }
 function apply(ctx, config) {
   const summarizer = resolveSummarizerTarget(config);
-  const patchPath = resolvePatchPath(config);
-  warnIfProfileLooksWrong(ctx, config, patchPath);
+  const profileName = resolveProfileName(ctx, config);
+  const patchPath = resolvePatchPath(config, profileName);
+  warnIfProfileLooksWrong(ctx, profileName, patchPath);
   const patchText = readPatchFileSync(patchPath);
   const legacyEngine = legacyBlockEngineOf(patchText);
   const needsWrite = hasLegacyManagedBlock(patchText) || !hasManagedBlock(patchText);
@@ -11783,9 +11817,11 @@ function apply(ctx, config) {
 export {
   Config11 as Config,
   apply,
+  inferProfileName,
   inject,
   name,
   resolvePatchPath,
+  resolveProfileName,
   resolveSummarizerTarget,
   syncManagedBlock,
   writePatchFile,

package/lib/types/index.d.ts

@@ -152,8 +152,44 @@ type ConfigInput = Omit<Config, 'engine' | 'showInComposer' | 'showEngineBadge'>
  * entry is the engine knobs alone.
  */
 export declare const Config: z<ConfigInput, Config>;
-/** Resolve the managed patch file from configuration, defaulting to the web profile. */
-export declare function resolvePatchPath(config: Config): string;
+/**
+ * Resolve the managed patch file from configuration, defaulting to the web profile.
+ * @param config - composition entry; `patchPath` wins, then `profile` + `patchFilename`.
+ * @param fallbackProfile - profile name to use when the entry names none (the
+ *   running profile, discovered by {@link resolveProfileName}); `web` when unknown.
+ * @returns the absolute patch file path.
+ */
+export declare function resolvePatchPath(config: Config, fallbackProfile?: string): string;
+/**
+ * Infer the profile a plugin copy was installed into from its own location.
+ *
+ * A profile is the nearest ancestor directory holding a `package.json` with a
+ * `dsh.profile` field. Walking up (rather than assuming
+ * `<home>/profiles/<name>/node_modules`) survives pnpm's real path
+ * (`node_modules/.pnpm/<pkg>@<ver>/node_modules/<pkg>/lib`) and nested installs,
+ * and returns undefined when the plugin runs from a checkout, a global install,
+ * or a linked source tree - none of which belongs to a profile.
+ * @param entry - module path to start from; the resolved module by default.
+ * @returns the profile name, or undefined when the copy is not inside one.
+ */
+export declare function inferProfileName(entry?: string): string | undefined;
+/**
+ * The profile this process is actually serving.
+ *
+ * Precedence: the composition entry's `profile` (an explicit deployment choice),
+ * then the harness's `profileContext.name` (0.2.x launchers), then the install
+ * location, and only then `web` (the historical default, correct for a global
+ * install and for `dsh web` before profiles had names).
+ *
+ * The install-location step is what makes a market install work without hand
+ * editing: installing into `desktop` used to leave `profile` unset, so the
+ * managed block went to `web`, the base `agent-loop` kept the factory slot in
+ * `desktop`, and every per-session switch answered `router-unmounted`.
+ * @param ctx - composing context, read for `profileContext`.
+ * @param config - composition entry, read for an explicit `profile`.
+ * @returns the profile name whose patch file this plugin owns.
+ */
+export declare function resolveProfileName(ctx: Context, config: Config): string;
 /**
  * Resolve the summarization model the plugin writes into every hosted preset,
  * or `undefined` when the deployment names none.

package/package.json

@@ -1,152 +1,161 @@
-{
-  "name": "dsh-loop-engine-allinone",
-  "description": "Web-switchable agent loop engine selection for the DeepSeek Harness - out-of-tree plugin with a data-driven engine registry and a pluggable driver layer. Ten engines: Claude Code / Codex / Pi / Kimi Code / Antigravity / ZCode / Qoder CN / Hermes / Grok Build / MiniMax Code. All-in-one fork of kuun993/dsh-loop-engine.",
-  "version": "0.7.3",
-  "publishConfig": {
-    "registry": "https://registry.npmjs.org/"
-  },
-  "repository": {
-    "type": "git",
-    "url": "git+https://github.com/yaliklinty/dsh-loop-engine-allinone.git"
-  },
-  "type": "module",
-  "main": "lib/index.js",
-  "types": "lib/types/index.d.ts",
-  "exports": {
-    ".": {
-      "types": "./lib/types/index.d.ts",
-      "default": "./lib/index.js"
-    },
-    "./invariant": {
-      "types": "./lib/types/invariant.d.ts",
-      "default": "./lib/invariant.js"
-    },
-    "./client": {
-      "types": "./lib/types/client/index.d.ts",
-      "default": "./lib/client.js"
-    },
-    "./src/*": "./src/*",
-    "./package.json": "./package.json"
-  },
-  "files": [
-    "lib/index.js",
-    "lib/invariant.js",
-    "lib/client.js",
-    "lib/types/**/*.d.ts",
-    "cordis.patch.yml",
-    "README.md",
-    "CHANGELOG.md",
-    "CONTRIBUTING.md",
-    "SECURITY.md",
-    "LICENSE",
-    "NOTICE"
-  ],
-  "license": "MIT",
-  "engines": {
-    "node": "^22.19.0 || >=24.0.0"
-  },
-  "packageManager": "pnpm@11.23.0",
-  "scripts": {
-    "build": "node build.mjs",
-    "typecheck": "tsc -p tsconfig.build.json --noEmit",
-    "test": "vitest run",
-    "test:coverage": "vitest run --coverage",
-    "compat:install": "pnpm -C .compat/015 install --frozen-lockfile && pnpm -C .compat/017 install --frozen-lockfile && pnpm -C .compat/020 install --frozen-lockfile",
-    "test:compat": "vitest run --config vitest.config.compat015.ts && vitest run --config vitest.config.compat017.ts && vitest run --config vitest.config.compat020.ts"
-  },
-  "dsh": {
-    "bundle": {
-      "patch": "./cordis.patch.yml"
-    },
-    "client": {
-      "inject": [
-        "@deepseek-ai/dsh-client-locale",
-        "@deepseek-ai/dsh-client-ui-settings",
-        "@deepseek-ai/dsh-api-remotes",
-        "@deepseek-ai/dsh-api-session-controller"
-      ],
-      "platform": "web",
-      "external": [
-        "@deepseek-ai/dsh-client-locale/client",
-        "@deepseek-ai/dsh-client-store",
-        "@deepseek-ai/dsh-client-ui-settings/client",
-        "@deepseek-ai/dsh-api-remotes/client",
-        "@deepseek-ai/dsh-settings/types"
-      ]
-    }
-  },
-  "dependencies": {
-    "@anthropic-ai/claude-agent-sdk": "0.3.220",
-    "@anthropic-ai/sdk": "0.93.0",
-    "@deepseek-ai/schemastery": "3.18.4",
-    "@earendil-works/pi-coding-agent": "0.84.3",
-    "@openai/codex": "0.149.1"
-  },
-  "peerDependencies": {
-    "@deepseek-ai/cordis": "^4.0.1",
-    "@deepseek-ai/dsh-agent": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-agent-loop": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-invariants": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-llm": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-scope": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-session": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-session-persistence": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-settings": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-subprocess": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-timeout": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0",
-    "@deepseek-ai/dsh-typert-protocol": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.1-0"
-  },
-  "devDependencies": {
-    "@deepseek-ai/dsh-agent": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-agent-loop": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-invariants": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-llm": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-scope": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-session": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-session-persistence": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-settings": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-subprocess": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-timeout": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-attachment": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-client-locale": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-client-store": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-client-ui-conversation": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-client-ui-primitives": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-client-ui-renderer": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-client-ui-settings": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-client-ui-slots": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-home-paths": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-session-persistence-jsonl": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-subprocess-local": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-system-prompt": "0.2.0-rc.2",
-    "@deepseek-ai/dsh-typert-protocol": "0.2.0-rc.2",
-    "@types/node": "^22.20.0",
-    "@types/react": "^19.0.0",
-    "@vitest/coverage-v8": "^4.1.8",
-    "esbuild": "^0.28.2",
-    "tsx": "^4.22.4",
-    "typescript": "^6.0.3",
-    "vite": "^7.0.0",
-    "vite-tsconfig-paths": "^6.1.1",
-    "vitest": "^4.1.8"
-  },
-  "bugs": {
-    "url": "https://github.com/yaliklinty/dsh-loop-engine-allinone/issues"
-  },
-  "homepage": "https://github.com/yaliklinty/dsh-loop-engine-allinone#readme",
-  "author": "dsh-loop-engine-allinone contributors",
-  "keywords": [
-    "deepseek-harness",
-    "dsh",
-    "dsh-plugin",
-    "agent-loop",
-    "coding-agent",
-    "claude-code",
-    "codex",
-    "kimi",
-    "antigravity",
-    "grok",
-    "minimax",
-    "acp"
-  ]
-}
+{
+  "name": "dsh-loop-engine-allinone",
+  "description": "Web-switchable agent loop engine selection for the DeepSeek Harness - out-of-tree plugin with a data-driven engine registry and a pluggable driver layer. Ten engines: Claude Code / Codex / Pi / Kimi Code / Antigravity / ZCode / Qoder CN / Hermes / Grok Build / MiniMax Code. All-in-one fork of kuun993/dsh-loop-engine.",
+  "version": "0.7.5",
+  "publishConfig": {
+    "registry": "https://registry.npmjs.org/"
+  },
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/yaliklinty/dsh-loop-engine-allinone.git"
+  },
+  "type": "module",
+  "main": "lib/index.js",
+  "types": "lib/types/index.d.ts",
+  "exports": {
+    ".": {
+      "types": "./lib/types/index.d.ts",
+      "default": "./lib/index.js"
+    },
+    "./invariant": {
+      "types": "./lib/types/invariant.d.ts",
+      "default": "./lib/invariant.js"
+    },
+    "./client": {
+      "types": "./lib/types/client/index.d.ts",
+      "default": "./lib/client.js"
+    },
+    "./src/*": "./src/*",
+    "./package.json": "./package.json"
+  },
+  "files": [
+    "lib/index.js",
+    "lib/invariant.js",
+    "lib/client.js",
+    "lib/types/**/*.d.ts",
+    "cordis.patch.yml",
+    "README.md",
+    "CHANGELOG.md",
+    "CONTRIBUTING.md",
+    "SECURITY.md",
+    "LICENSE",
+    "NOTICE"
+  ],
+  "license": "MIT",
+  "engines": {
+    "node": "^22.19.0 || >=24.0.0"
+  },
+  "packageManager": "pnpm@11.23.0",
+  "scripts": {
+    "build": "node build.mjs",
+    "typecheck": "tsc -p tsconfig.build.json --noEmit",
+    "test": "vitest run",
+    "test:coverage": "vitest run --coverage && node scripts/coverage-gate.mjs",
+    "coverage:baseline": "node scripts/coverage-gate.mjs --update",
+    "compat:install": "pnpm -C .compat/015 install --frozen-lockfile && pnpm -C .compat/017 install --frozen-lockfile && pnpm -C .compat/020 install --frozen-lockfile && pnpm -C .compat/021 install --frozen-lockfile",
+    "test:compat": "vitest run --config vitest.config.compat015.ts && vitest run --config vitest.config.compat017.ts && vitest run --config vitest.config.compat020.ts && vitest run --config vitest.config.compat021.ts",
+    "typecheck:compat": "tsc -p tsconfig.compat021.json"
+  },
+  "dsh": {
+    "bundle": {
+      "patch": "./cordis.patch.yml"
+    },
+    "client": {
+      "inject": [
+        "@deepseek-ai/dsh-client-locale",
+        "@deepseek-ai/dsh-client-ui-settings",
+        "@deepseek-ai/dsh-api-remotes",
+        "@deepseek-ai/dsh-api-session-controller"
+      ],
+      "platform": "web",
+      "external": [
+        "@deepseek-ai/dsh-client-locale/client",
+        "@deepseek-ai/dsh-client-store",
+        "@deepseek-ai/dsh-client-ui-settings/client",
+        "@deepseek-ai/dsh-api-remotes/client",
+        "@deepseek-ai/dsh-settings/types"
+      ]
+    }
+  },
+  "dependencies": {
+    "@anthropic-ai/claude-agent-sdk": "0.3.220",
+    "@anthropic-ai/sdk": "0.93.0",
+    "@deepseek-ai/schemastery": "3.18.4",
+    "@earendil-works/pi-coding-agent": "0.84.3",
+    "@openai/codex": "0.149.1"
+  },
+  "peerDependencies": {
+    "@deepseek-ai/cordis": "^4.0.1",
+    "@deepseek-ai/dsh-agent": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-agent-loop": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-invariants": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-llm": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-scope": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-session": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-session-persistence": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-settings": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-subprocess": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-timeout": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0",
+    "@deepseek-ai/dsh-typert-protocol": ">=0.1.5-rc.1 <0.1.6-0 || >=0.1.7-rc.1 <0.1.8-0 || >=0.2.0-rc.1 <0.2.2-0"
+  },
+  "devDependencies": {
+    "@deepseek-ai/dsh-agent": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-agent-loop": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-invariants": "0.2.0-rc.2",
+    "@deepseek-ai/dsh-llm": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-scope": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-session": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-session-persistence": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-settings": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-subprocess": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-timeout": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-attachment": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-client-locale": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-client-store": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-client-ui-conversation": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-client-ui-primitives": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-client-ui-renderer": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-client-ui-settings": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-client-ui-slots": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-home-paths": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-session-persistence-jsonl": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-subprocess-local": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-system-prompt": "0.2.1-rc.1",
+    "@deepseek-ai/dsh-typert-protocol": "0.2.1-rc.1",
+    "@types/node": "^22.20.0",
+    "@types/react": "^19.0.0",
+    "@vitest/coverage-v8": "^4.1.8",
+    "esbuild": "^0.28.2",
+    "tsx": "^4.22.4",
+    "typescript": "^6.0.3",
+    "vite": "^7.0.0",
+    "vite-tsconfig-paths": "^6.1.1",
+    "vitest": "^4.1.8",
+    "@deepseek-ai/cordis": "4.0.5",
+    "@deepseek-ai/schemastery": "3.18.5"
+  },
+  "bugs": {
+    "url": "https://github.com/yaliklinty/dsh-loop-engine-allinone/issues"
+  },
+  "homepage": "https://github.com/yaliklinty/dsh-loop-engine-allinone#readme",
+  "author": "dsh-loop-engine-allinone contributors",
+  "keywords": [
+    "deepseek-harness",
+    "dsh",
+    "dsh-plugin",
+    "agent-loop",
+    "coding-agent",
+    "claude-code",
+    "codex",
+    "kimi",
+    "antigravity",
+    "grok",
+    "minimax",
+    "acp"
+  ],
+  "peerDependenciesMeta": {
+    "@deepseek-ai/dsh-invariants": {
+      "optional": true
+    }
+  }
+}

package/lib/types/engine-kimi/acp/client.d.ts

@@ -1,123 +0,0 @@
-/**
- * JSON-RPC client for the `kimi acp` subprocess (Agent Client Protocol over
- * stdio).
- *
- * The driver hands the client a process handle carrying its stdin/stdout/stderr
- * (projected from the dsh subprocess seam). The client frames JSON-RPC 2.0
- * records on a bare `\n` (via a byte decoder, tolerating a trailing `\r`),
- * correlates request responses by `id`, dispatches every `session/update`
- * notification to a buffered event stream, and answers the reverse-RPC
- * `session/request_permission` requests the agent publishes. The child is
- * long-lived (one per factory) and stepped over via `newSession` + `prompt`.
- *
- * @module dsh-loop-engine/engine-kimi/acp/client
- */
-import type { KimiProcess, KimiSpawnCapability, KimiSpawnSpec } from '../process.ts';
-import { type AcpFrame, type AcpImageContent, type AcpPermissionOption, type AcpPermissionResponse, type AcpUpdate } from './types.ts';
-/** How the client answers one `session/request_permission`. */
-export type AcpPermissionHandler = (request: AcpFrame) => boolean | Promise<boolean>;
-/**
- * The answerable options of one `session/request_permission` frame. The wire is
- * untrusted, so entries without a usable `optionId` are dropped rather than
- * echoed back.
- * @param frame - the reverse-RPC request frame.
- * @returns the options the client may answer with.
- */
-export declare function permissionOptionsOf(frame: AcpFrame): readonly AcpPermissionOption[];
-/**
- * Encode one approval decision as the ACP `RequestPermissionResponse` the agent
- * correlates against the options it advertised. The agent reads a terminal
- * *option*, never a boolean: an outcome it cannot resolve (or a response that
- * fails to parse at all) is reported to the model as a user rejection, so a
- * decision that no single option expresses becomes `cancelled` — the honest
- * "this client has no answer". Kimi re-uses this RPC for its question and
- * plan-review bridges, which offer one `allow_once` option per choice; picking
- * one of those would answer a question no human was asked, so any ambiguous
- * set (zero or several `allow_once`/`reject_once` candidates) cancels instead.
- * `allow_once` is preferred over `allow_always` because the decision is re-read
- * from the session knobs on every request: letting the agent cache a session
- * grant would outlive a mid-session switch back to `ask`.
- * @param approved - whether the driver approves the request.
- * @param options - the options the agent advertised for this request.
- * @returns the result payload to send back.
- */
-export declare function permissionResponse(approved: boolean, options: readonly AcpPermissionOption[]): AcpPermissionResponse;
-/** Callback receiving every non-response event line. */
-export type AcpUpdateHandler = (update: AcpUpdate) => void;
-/**
- * ACP client over one `kimi acp` child. Created once per driver scope and reused
- * across steps, matching the Pi RPC client's lifecycle.
- */
-export declare class AcpClient {
-    private readonly process;
-    private readonly pending;
-    private readonly updateBuffer;
-    private updateWake;
-    private updateHandler;
-    private permissionHandler;
-    private sealed;
-    private nextId;
-    private readonly decoder;
-    private buffer;
-    /** Whether this client was sealed or its process exited. */
-    get closed(): boolean;
-    /** Mount a client over an already-spawned `kimi acp` process. */
-    constructor(process: KimiProcess);
-    /**
-     * Create a client, spawning the `kimi acp` child through the supplied
-     * capability (or the default node spawn when none is given).
-     * @param spec - the `kimi acp` argv/cwd/env the child should run with.
-     * @param spawn - optional process-spawn capability (the subprocess seam).
-     * @returns the connected client.
-     */
-    static create(spec: KimiSpawnSpec, spawn?: KimiSpawnCapability): AcpClient;
-    /** Register the event dispatch handler. */
-    onUpdate(handler: AcpUpdateHandler): void;
-    /** Register the permission-approval handler (reverse-RPC answers). */
-    onPermission(handler: AcpPermissionHandler): void;
-    /** Send one request and await the correlated response. */
-    request(method: string, params: unknown): Promise<unknown>;
-    /** Send a notification (no correlated response awaited). */
-    notify(method: string, params: unknown): void;
-    /** Open the protocol handshake. */
-    initialize(): Promise<unknown>;
-    /** Start a fresh ACP session and resolve to its session id. */
-    newSession(cwd: string): Promise<string>;
-    /**
-     * Select the model for one ACP session. The agent answers with an error frame
-     * when it cannot serve the id, and {@link request} rejects on that frame — so
-     * an engine that refuses the model fails the step loud rather than silently
-     * running its own default.
-     * @param sessionId - the ACP session the model applies to.
-     * @param modelId - the model id to select.
-     * @returns the agent's response to `session/set_model`.
-     */
-    setModel(sessionId: string, modelId: string): Promise<unknown>;
-    /**
-     * Prompt the agent in a session and resolve when the turn completes.
-     * @param sessionId - the ACP session to prompt.
-     * @param text - the serialized transcript, or a bare slash command.
-     * @param images - image blocks to carry after the text block; omitted or
-     *   empty leaves the request exactly the text-only shape.
-     * @returns the agent's response to `session/prompt`.
-     */
-    prompt(sessionId: string, text: string, images?: readonly AcpImageContent[]): Promise<unknown>;
-    /** Cancel the active turn in a session (fire-and-forget). */
-    cancel(sessionId: string): void;
-    /**
-     * Answer a pending `session/request_permission`.
-     * @param id - the reverse-RPC request id.
-     * @param response - the ACP outcome to answer with.
-     */
-    respondPermission(id: number, response: AcpPermissionResponse): void;
-    /** Consume every buffered update as an async generator. */
-    updates(): AsyncGenerator<AcpUpdate, void, void>;
-    /** Seal the client and request child termination. */
-    dispose(): void;
-    private feed;
-    /** Dispatch one parsed line: a response, an update notification, or a reverse-RPC request. */
-    private dispatch;
-    private settle;
-    private handlePermission;
-}
-//# sourceMappingURL=client.d.ts.map

package/lib/types/engine-kimi/acp/mapping.d.ts

@@ -1,73 +0,0 @@
-/**
- * Maps `kimi acp` `session/update` events to dsh session-log projections.
- *
- * Kimi streams incremental assistant text (`agent_message_chunk`), incremental
- * thinking (`agent_thought_chunk`), a tool-call announcement (`tool_call`) and its
- * progress/result updates (`tool_call_update`). Assistant text and thinking are
- * Deltas; a tool update's content is a whole Snapshot that replaces the previous
- * one. A tool call's identity arrives on the announcement but its input
- * (`rawInput`) does not — only a later update carries it. This module is pure: it
- * classifies an update, extracts chunk deltas, and projects the tool-call
- * identity/input/content/result so the agent can fold them into the durable log.
- * Content blocks use the observed kimi
- * `{ type: 'content', content: { type: 'text', text } }` nesting; unknown block
- * types are ignored.
- *
- * @module dsh-loop-engine/engine-kimi/acp/mapping
- */
-import type { ToolResultMessage } from '@deepseek-ai/dsh-llm';
-import type { AcpContentBlock, AcpToolCallExt, AcpToolCallStreamExt, AcpUpdate } from './types.ts';
-/** Whether the update is an incremental assistant text chunk. */
-export declare function isTextChunk(update: AcpUpdate): update is AcpUpdate & {
-    readonly sessionUpdate: 'agent_message_chunk';
-    readonly content: AcpContentBlock;
-};
-/** Whether the update is an incremental thinking chunk. */
-export declare function isThoughtChunk(update: AcpUpdate): update is AcpUpdate & {
-    readonly sessionUpdate: 'agent_thought_chunk';
-    readonly content: AcpContentBlock;
-};
-/** Whether the update announces a tool call. */
-export declare function isToolCall(update: AcpUpdate): update is AcpToolCallExt;
-/** Whether the update streams a tool call's progress/result. */
-export declare function isToolCallUpdate(update: AcpUpdate): update is AcpToolCallStreamExt;
-/** The delta text of a text/thinking chunk. */
-export declare function chunkDelta(update: AcpUpdate): string;
-/** The raw tool-call id (+ content index) as the wire carries it. */
-export declare function toolCallIdOf(update: AcpUpdate): string;
-/** The tool display name (`title`). */
-export declare function toolCallName(update: AcpUpdate): string;
-/**
- * The tool call's real input as the JSON `arguments` string the durable
- * `tool/call` carries, when the frame supplies it.
- *
- * The `tool_call` announcement never carries `rawInput`; a later
- * `tool_call_update` does (measured against kimi 0.28.x: the execution-start
- * frame, `status: 'in_progress'`). A call's arguments are therefore unknowable
- * at announce time, and the driver logs the call only once an update supplies
- * them — or when the call settles, whichever comes first.
- * @param update - one tool-call announcement or update.
- * @returns the arguments string, or `undefined` when the frame omits the field.
- */
-export declare function toolRawInput(update: AcpToolCallExt | AcpToolCallStreamExt): string | undefined;
-/** Whether a tool stream status is settled (no longer streaming). */
-export declare function isToolSettledStatus(status: string): boolean;
-/** Whether a tool stream status denotes a failure. */
-export declare function isToolErrorStatus(status: string): boolean;
-/**
- * The tool call's content as ONE update carries it: the joined text of its
- * observed `{ type: 'content', content: { type: 'text', text } }` blocks.
- *
- * Kimi re-sends the call's whole content on every `tool_call_update` rather than
- * streaming deltas (measured against 0.28.1: the string grows from the tool
- * input's rendering to the final output), so this is a *snapshot* — callers
- * replace with it, never append. `undefined` means the update carried no
- * content field at all (nothing to replace), which is distinct from a present
- * but empty one.
- * @param update - one tool-call announcement or update.
- * @returns the content snapshot, or `undefined` when the field is absent.
- */
-export declare function toolContentText(update: AcpToolCallExt | AcpToolCallStreamExt): string | undefined;
-/** Project a completed tool call to a durable tool/result message. */
-export declare function toolResult(callId: string, text: string, isError: boolean): ToolResultMessage;
-//# sourceMappingURL=mapping.d.ts.map

package/lib/types/engine-kimi/acp/types.d.ts

@@ -1,143 +0,0 @@
-/**
- * Agent Client Protocol (ACP) wire types for the `kimi acp` driver.
- *
- * Kimi exposes the ACP adapter over stdio (JSON-RPC 2.0). The driver speaks a
- * minimal, stable subset that a normal agent flow needs: initialize, session/new,
- * session/prompt, session/cancel, plus the `session/update` notifications carrying
- * incremental assistant text (`agent_message_chunk`), thinking
- * (`agent_thought_chunk`), and tool calls (`tool_call` / `tool_call_update`).
- * Tool-approval arrives as the reverse-RPC `session/request_permission`, which the
- * client must answer. Shapes below are the live-record deltas observed against the
- * 0.28.1 CLI; content blocks re-use a `{ type, text }` text shape (and tolerate
- * other block types by ignoring them). The image block is modelled for the
- * OUTGOING side — the bytes a dsh step delivers as an ACP image block.
- *
- * @module dsh-loop-engine/engine-kimi/acp/types
- */
-/** A text content block in an ACP update (the observed kimi shape). */
-export interface AcpTextContent {
-    readonly type: 'text';
-    readonly text: string;
-}
-/** One tool-call content block: the observed kimi `{ type: 'content', content }` nesting. */
-export interface AcpToolContentBlock {
-    readonly type: 'content';
-    readonly content: AcpTextContent;
-}
-/** An image content block, as ACP/MCP spells it (base64 bytes + media type). */
-export interface AcpImageContent {
-    readonly type: 'image';
-    /** Base64-encoded image bytes. */
-    readonly data: string;
-    /** The image's own media type. */
-    readonly mimeType: string;
-}
-/** A content block in a chunk/thought update; text is handled, others are ignored. */
-export type AcpContentBlock = AcpTextContent | AcpImageContent | {
-    readonly type: 'resource_link' | 'audio' | string;
-    readonly [key: string]: unknown;
-};
-/** One `session/update` notification (partial; the discriminator is `sessionUpdate`). */
-export interface AcpUpdate {
-    readonly sessionUpdate: string;
-    readonly [key: string]: unknown;
-}
-/**
- * A tool-call announcement (`sessionUpdate: 'tool_call'`).
- *
- * Measured against kimi 0.28.x: an announcement carries the call's identity
- * (`toolCallId`/`title`/`kind`) and an empty content card, but **never** its
- * input — `rawInput` only ever arrives on a later `tool_call_update` (the
- * execution-start frame).
- */
-export interface AcpToolCallExt extends AcpUpdate {
-    readonly sessionUpdate: 'tool_call';
-    readonly toolCallId: string;
-    readonly title: string;
-    readonly kind: string;
-    readonly status: string;
-    readonly content?: readonly AcpToolContentBlock[];
-    /** The call's real input; absent on the announcement (see above). */
-    readonly rawInput?: unknown;
-}
-/** A tool-call progress/result stream (`sessionUpdate: 'tool_call_update'`). */
-export interface AcpToolCallStreamExt extends AcpUpdate {
-    readonly sessionUpdate: 'tool_call_update';
-    readonly toolCallId: string;
-    readonly status: string;
-    readonly content?: readonly AcpToolContentBlock[];
-    /** The call's real input, first present on the execution-start frame. */
-    readonly rawInput?: unknown;
-    /** The tool's raw structured output, present once the call settles. */
-    readonly rawOutput?: unknown;
-}
-/** One option the agent offers for a `session/request_permission`. */
-export interface AcpPermissionOption {
-    /** Opaque id the client sends back in `outcome.optionId` (the only field the agent reads). */
-    readonly optionId: string;
-    /** Human-readable label for the option. */
-    readonly name?: string;
-    /** Option class: `allow_once` / `allow_always` / `reject_once` / `reject_always`. */
-    readonly kind?: string;
-}
-/** The `session/request_permission` reverse-RPC params. */
-export interface AcpPermissionRequest {
-    readonly sessionId?: string;
-    /** Every option the agent will accept an answer for; empty means no answerable choice. */
-    readonly options?: readonly AcpPermissionOption[];
-    /** The tool call the prompt belongs to (presentation/correlation only). */
-    readonly toolCall?: unknown;
-}
-/**
- * Terminal outcome of a permission request: the option the client picked, or
- * `cancelled` when it had no answer to give.
- */
-export type AcpPermissionOutcome = {
-    readonly outcome: 'selected';
-    readonly optionId: string;
-} | {
-    readonly outcome: 'cancelled';
-};
-/** The `session/request_permission` result the agent expects (ACP `RequestPermissionResponse`). */
-export interface AcpPermissionResponse {
-    readonly outcome: AcpPermissionOutcome;
-}
-/** Result of a completed `session/prompt` request (opaque; the turn ended). */
-export interface AcpPromptResult {
-    readonly [key: string]: unknown;
-}
-/** The id-scoped response to a request (result or error). */
-export interface AcpResponse {
-    readonly id: number;
-    readonly result?: unknown;
-    readonly error?: {
-        readonly code: number;
-        readonly message: string;
-    };
-}
-/** A JSON-RPC request/notification frame on the wire. */
-export interface AcpFrame {
-    readonly jsonrpc: '2.0';
-    readonly id?: number;
-    readonly method?: string;
-    readonly params?: unknown;
-    readonly result?: unknown;
-    readonly error?: {
-        readonly code: number;
-        readonly message: string;
-    };
-}
-/** Guards toolkit: is the frame a `session/update` notification carrying an update. */
-export declare function isUpdateFrame(frame: AcpFrame): frame is AcpFrame & {
-    readonly method: 'session/update';
-    readonly params: {
-        readonly sessionId: string;
-        readonly update: AcpUpdate;
-    };
-};
-/** Guards toolkit: is the frame a reverse-RPC request needing an approval answer. */
-export declare function isPermissionRequestFrame(frame: AcpFrame): frame is AcpFrame & {
-    readonly id: number;
-    readonly method: 'session/request_permission';
-};
-//# sourceMappingURL=types.d.ts.map