dsh-switchman 1.6.0 → 1.7.1

package/README.md

@@ -95,6 +95,7 @@ bash <install-dir>/skills/db-query/scripts/setup.sh
 - After a DSH upgrade, if the factory preset plugin list changed, re-sync `cordis.patch.yml` from the new `presets/*.patch.yml` (keeping the doctrine suffix), then reinstall.
 - Protocol lines (`[SWITCHMAN:LANG|POOLS|WATERMARK|TEAMS]`) are deliberately English and byte-stable — don't localize them.
 - `npm pack --dry-run` should keep the audited 50-file / ~205 kB shape (`docs/` screenshots stay out of the package).
+- **Zero runtime dependencies — a hard invariant.** Host-plane packages (`@deepseek-ai/schemastery`, cordis, …) are declared as `peerDependencies` only, and the running DSH satisfies them from its own copy. An ordinary `dependencies` entry is physically installed into the profile workspace (`<profile>/node_modules/…`), where it shadows the host's copy for every other profile-resident plugin — this is exactly what crashed the dshmarket plugin market while 1.7.0 carried schemastery as a dependency (two schemastery instances in one process; the market's schemas no longer matched the host's). `scripts/validate.mjs` fails if a `dependencies` entry ever reappears, so installing, updating, or removing dsh-switchman can never touch what dshmarket or any other community bundle resolves at runtime.
 
 ## License
 

package/README.zh.md

@@ -95,6 +95,7 @@ bash <安装目录>/skills/db-query/scripts/setup.sh
 - DSH 升级后若出厂预设插件列表有变，从新的 `presets/*.patch.yml` 重新同步 `cordis.patch.yml`（保留 doctrine suffix），然后重装。
 - 协议行（`[SWITCHMAN:LANG|POOLS|WATERMARK|TEAMS]`）刻意保持英文且字节稳定——不要本地化。
 - `npm pack --dry-run` 应保持审计过的 50 文件 / ~205 kB 形态（`docs/` 截图不进包）。
+- **零运行时依赖——硬性不变量。** 宿主侧包（`@deepseek-ai/schemastery`、cordis 等）只允许声明为 `peerDependencies`，由运行中的 DSH 用它自己的副本满足。普通 `dependencies` 会被物理安装进 profile 工作区（`<profile>/node_modules/…`），从而对同处 profile 的其他插件遮蔽宿主副本——1.7.0 把 schemastery 声明为普通依赖时，正是这个机制让 dshmarket 插件市场崩溃（一个进程里出现两份 schemastery，市场产生的 schema 与宿主不再同类）。`scripts/validate.mjs` 会在 `dependencies` 再次出现时直接失败；因此安装、升级、卸载 dsh-switchman 都不会再影响 dshmarket 或任何其他社区包在运行时解析到的模块。
 
 ## License
 

package/cordis.patch.yml

@@ -28,10 +28,15 @@
 #
 # Provenance & maintenance: the four `plugins` lists below are copied verbatim
 # from @deepseek-ai/dsh-web-app presets/{standard,ptc,minimal,cordis}.patch.yml
-# (dsh 0.1.7-rc.2). A Loader override replaces the target row's `config`
-# wholesale, so every field is restated here. After a DSH upgrade that changes
-# the shipped presets, re-sync these lists from the new preset files, then
-# reinstall this bundle.
+# (dsh 0.2.1-rc.1: time-context + tool-schedule + the disabled tool-plugin-market
+# row joined every list, the subagent rows lost `backgroundMode` under the
+# unified activation lifecycles, the ralph/codex/claude-code rows left the
+# presets, minimal gained image-only fs tools and a compaction group, and the
+# plan-mode text was reworded). A Loader override replaces the target row's
+# `config` wholesale, so every field is restated here. After a DSH upgrade that
+# changes the shipped presets, re-sync these lists from the new preset files,
+# then reinstall this bundle; scripts/validate.mjs compares this patch against
+# the installed app's presets field by field.
 #
 # NEVER set `modelSelectionSettings: true` on the tool-subagent-fork row (any
 # preset). Only the spawn row may carry it: while the subagent model-selection
@@ -40,9 +45,19 @@
 # scope, and a second instance makes session composition throw
 # `tool "list_subagent_models" is already registered in this scope` — breaking
 # creation of ALL new sessions and the recomposition of old ones. The shipped
-# presets (verified against installed dsh-web-app 0.2.0-rc.2 and the fork's
-# 0.2.1-alpha.1) enable it on the spawn row only; fork-side explicit routing
-# is not a factory capability.
+# presets (verified against installed dsh-web-app 0.2.0-rc.2, the fork's
+# 0.2.1-alpha.1, and 0.2.1-rc.1) enable it on the spawn row only; fork-side
+# explicit routing is not a factory capability.
+#
+# dsh 0.2.1 REMOVED the `cwd` prompt variable ("centralize working directories
+# and directory context"): registered prompt variables are `provider` and
+# `model` only, and the working directory now reaches the model as REQUIRED
+# runtime context in every preset — it survives even minimal's
+# includeRuntimeContext: false, so no preset needs a directory line here. The
+# suffix therefore carries NO `Your working directory is {{cwd}}.` line any
+# more: an unknown variable in an interpolated section throws at EVERY prompt
+# render and kills every conversation while this bundle is enabled. Do not
+# reintroduce {{cwd}} (or any variable beyond provider/model).
 #
 # Known consequences of overriding:
 # - plugin_manager `set_plugin` on these preset rows reports `overridden`.
@@ -65,8 +80,6 @@
         config:
           prefix: You are a coding agent powered by the {{model}} model.
           suffix: &switchman-team-suffix |
-            Your working directory is {{cwd}}.
-
             # dsh-switchman 自主智能体团队调度规程（本段为当前 profile 的最高行为准则）
 
             总纲：在准确理解需求、严谨思考、最优质量、最低返工率的前提下最小化总 token 消耗。返工是最贵的 token——过程与铺陈可以省，验证与关键推理绝不省。
@@ -99,6 +112,8 @@
         name: '@deepseek-ai/dsh-agent-instructions'
         config:
           maxBytes: 65536
+      - id: time-context
+        name: '@deepseek-ai/dsh-time-context'
       - id: tool-bash
         name: '@deepseek-ai/dsh-tool-bash'
         disabled: !!js process.platform === 'win32'
@@ -113,6 +128,8 @@
           sampleOverCapGlobResults: false
       - id: tool-jobs
         name: '@deepseek-ai/dsh-tool-jobs'
+      - id: tool-schedule
+        name: '@deepseek-ai/dsh-tool-schedule'
       - id: skill-filesystem
         name: '@deepseek-ai/dsh-skill-filesystem'
       - id: tool-skill
@@ -131,7 +148,7 @@
             name: '@deepseek-ai/dsh-plan-mode'
             config:
               section: |
-                You are in plan mode. Stay in plan mode until exit_plan_mode succeeds or the user switches the session mode. Imperative language to implement changes means plan the implementation, not execute it. A user's conversational agreement — including an answer confirming something you asked — approves nothing and does not end plan mode; fold the confirmed decision into the plan and submit it through exit_plan_mode.
+                You are in plan mode. Stay in plan mode until the user approves your plan through exit_plan_mode or switches the session mode. Imperative language to implement changes means plan the implementation, not execute it. A user's conversational agreement — including an answer confirming something you asked — approves nothing and does not end plan mode; fold the confirmed decision into the plan and submit it through exit_plan_mode.
 
                 Explore first. Use non-mutating reads, searches, static analysis, and checks to ground the plan in the actual repository. Do not edit or write files, change configuration, run formatters or code generation that rewrites tracked files, commit, or otherwise carry out the plan. Prefer existing functions and patterns over new machinery.
 
@@ -175,41 +192,17 @@
               provider: spawn
               toolName: subagent
               modelSelectionSettings: true
-              backgroundMode: continuable
           - id: tool-subagent-fork
             name: '@deepseek-ai/dsh-tool-subagent'
             config:
               provider: fork
               toolName: subagent_fork
-              backgroundMode: continuable
-          - id: tool-subagent-codex
-            name: '@deepseek-ai/dsh-tool-subagent'
-            disabled: true
-            config:
-              provider: codex
-              toolName: subagent_codex
-              backgroundMode: one-shot
-              maxDepth: provider-managed
-          - id: tool-subagent-claude-code
-            name: '@deepseek-ai/dsh-tool-subagent'
-            disabled: true
-            config:
-              provider: claude-code
-              toolName: subagent_claude_code
-              backgroundMode: one-shot
-              maxDepth: provider-managed
           - id: workflow-ptc
             name: '@deepseek-ai/dsh-workflow-ptc'
             config:
               provider: spawn
           - id: tool-workflow
             name: '@deepseek-ai/dsh-tool-workflow'
-          - id: tool-ralph
-            name: '@deepseek-ai/dsh-tool-ralph'
-            disabled: true
-            config:
-              subagentProvider: spawn
-              maxRounds: 64
       - id: tool-ask-user
         name: '@deepseek-ai/dsh-tool-ask-user'
       - id: tool-todo
@@ -226,6 +219,9 @@
       - id: tool-plugin-manager
         name: '@deepseek-ai/dsh-plugin-manager/tools'
         disabled: true
+      - id: tool-plugin-market
+        name: '@deepseek-ai/dsh-plugin-market/tools'
+        disabled: true
 
 # -----------------------------------------------------------------------------
 # ptc — shipped list verbatim (workflow tooling disabled, PTC presentation);
@@ -246,6 +242,8 @@
         name: '@deepseek-ai/dsh-agent-instructions'
         config:
           maxBytes: 65536
+      - id: time-context
+        name: '@deepseek-ai/dsh-time-context'
       - id: tool-bash
         name: '@deepseek-ai/dsh-tool-bash'
         disabled: !!js process.platform === 'win32'
@@ -260,6 +258,8 @@
           sampleOverCapGlobResults: false
       - id: tool-jobs
         name: '@deepseek-ai/dsh-tool-jobs'
+      - id: tool-schedule
+        name: '@deepseek-ai/dsh-tool-schedule'
       - id: skill-filesystem
         name: '@deepseek-ai/dsh-skill-filesystem'
       - id: tool-skill
@@ -278,7 +278,7 @@
             name: '@deepseek-ai/dsh-plan-mode'
             config:
               section: |
-                You are in plan mode. Stay in plan mode until exit_plan_mode succeeds or the user switches the session mode. Imperative language to implement changes means plan the implementation, not execute it. A user's conversational agreement — including an answer confirming something you asked — approves nothing and does not end plan mode; fold the confirmed decision into the plan and submit it through exit_plan_mode.
+                You are in plan mode. Stay in plan mode until the user approves your plan through exit_plan_mode or switches the session mode. Imperative language to implement changes means plan the implementation, not execute it. A user's conversational agreement — including an answer confirming something you asked — approves nothing and does not end plan mode; fold the confirmed decision into the plan and submit it through exit_plan_mode.
 
                 Explore first. Use non-mutating reads, searches, static analysis, and checks to ground the plan in the actual repository. Do not edit or write files, change configuration, run formatters or code generation that rewrites tracked files, commit, or otherwise carry out the plan. Prefer existing functions and patterns over new machinery.
 
@@ -322,29 +322,11 @@
               provider: spawn
               toolName: subagent
               modelSelectionSettings: true
-              backgroundMode: continuable
           - id: tool-subagent-fork
             name: '@deepseek-ai/dsh-tool-subagent'
             config:
               provider: fork
               toolName: subagent_fork
-              backgroundMode: continuable
-          - id: tool-subagent-codex
-            name: '@deepseek-ai/dsh-tool-subagent'
-            disabled: true
-            config:
-              provider: codex
-              toolName: subagent_codex
-              backgroundMode: one-shot
-              maxDepth: provider-managed
-          - id: tool-subagent-claude-code
-            name: '@deepseek-ai/dsh-tool-subagent'
-            disabled: true
-            config:
-              provider: claude-code
-              toolName: subagent_claude_code
-              backgroundMode: one-shot
-              maxDepth: provider-managed
           - id: workflow-ptc
             name: '@deepseek-ai/dsh-workflow-ptc'
             disabled: true
@@ -353,12 +335,6 @@
           - id: tool-workflow
             name: '@deepseek-ai/dsh-tool-workflow'
             disabled: true
-          - id: tool-ralph
-            name: '@deepseek-ai/dsh-tool-ralph'
-            disabled: true
-            config:
-              subagentProvider: spawn
-              maxRounds: 64
       - id: tool-ask-user
         name: '@deepseek-ai/dsh-tool-ask-user'
       - id: tool-todo
@@ -379,13 +355,21 @@
       - id: tool-plugin-manager
         name: '@deepseek-ai/dsh-plugin-manager/tools'
         disabled: true
+      - id: tool-plugin-market
+        name: '@deepseek-ai/dsh-plugin-market/tools'
+        disabled: true
 
 # -----------------------------------------------------------------------------
-# minimal — shipped list verbatim (persistent shell); the shipped persona has
-# no suffix, so this override adds the doctrine suffix. The shared suffix also
-# carries the "{{cwd}}" line: minimal's persistent shell tools need the working
-# directory, and the variable is registered host-wide regardless of the
-# persona's includeRuntimeContext: false.
+# minimal — shipped list verbatim (image-only fs tools, compaction, persistent
+# shell). The shipped persona keeps `complete: true` + `includeRuntimeContext:
+# false`, and the shared doctrine suffix stays attached for the four-row shape
+# contract of scripts/validate.mjs — but a complete persona prefix IS the whole
+# system prompt (the registry ignores the suffix and every other section at
+# render), so the doctrine does not reach minimal sessions; only the prefix
+# renders. The working directory does reach them: since dsh 0.2.1 it arrives
+# as REQUIRED runtime context, which includeRuntimeContext: false cannot
+# suppress, so minimal's persistent shell still sees the cwd with no
+# `{{cwd}}` variable and no directory line in the suffix.
 # -----------------------------------------------------------------------------
 - id: preset-minimal
   name: '@deepseek-ai/dsh-agent-preset'
@@ -400,6 +384,23 @@
           complete: true
           includeRuntimeContext: false
           suffix: *switchman-team-suffix
+      - id: tool-fs
+        name: '@deepseek-ai/dsh-tool-fs'
+        config:
+          read: false
+          readImage: true
+          write: false
+          edit: false
+      - id: compaction
+        name: cordis:group
+        group: true
+        isolate:
+          compaction: true
+        config:
+          - id: compaction-basic
+            name: '@deepseek-ai/dsh-compaction-basic'
+          - id: command-compact
+            name: '@deepseek-ai/dsh-command-compact'
       - id: persistent-shell
         name: cordis:group
         group: true
@@ -448,8 +449,8 @@
 
 # -----------------------------------------------------------------------------
 # cordis — shipped list verbatim (cordis tooling, bundled skills directory,
-# plugin-manager tools for profile-context sessions); persona suffix replaced
-# by the shared doctrine suffix.
+# plugin-manager/market tools for profile-context sessions); persona suffix
+# replaced by the shared doctrine suffix.
 # -----------------------------------------------------------------------------
 - id: preset-cordis
   name: '@deepseek-ai/dsh-agent-preset'
@@ -460,12 +461,15 @@
       - id: persona
         name: '@deepseek-ai/dsh-persona'
         config:
-          prefix: You are a coding agent powered by the {{model}} model.
+          prefix: >-
+            You are a coding agent powered by the {{model}} model.
           suffix: *switchman-team-suffix
       - id: agent-instructions
         name: '@deepseek-ai/dsh-agent-instructions'
         config:
           maxBytes: 65536
+      - id: time-context
+        name: '@deepseek-ai/dsh-time-context'
       - id: tool-bash
         name: '@deepseek-ai/dsh-tool-bash'
         disabled: !!js process.platform === 'win32'
@@ -480,6 +484,8 @@
           sampleOverCapGlobResults: false
       - id: tool-jobs
         name: '@deepseek-ai/dsh-tool-jobs'
+      - id: tool-schedule
+        name: '@deepseek-ai/dsh-tool-schedule'
       - id: command-goal
         name: '@deepseek-ai/dsh-command-goal'
       - id: tool-goal
@@ -494,7 +500,7 @@
             name: '@deepseek-ai/dsh-plan-mode'
             config:
               section: |
-                You are in plan mode. Stay in plan mode until exit_plan_mode succeeds or the user switches the session mode. Imperative language to implement changes means plan the implementation, not execute it. A user's conversational agreement — including an answer confirming something you asked — approves nothing and does not end plan mode; fold the confirmed decision into the plan and submit it through exit_plan_mode.
+                You are in plan mode. Stay in plan mode until the user approves your plan through exit_plan_mode or switches the session mode. Imperative language to implement changes means plan the implementation, not execute it. A user's conversational agreement — including an answer confirming something you asked — approves nothing and does not end plan mode; fold the confirmed decision into the plan and submit it through exit_plan_mode.
 
                 Explore first. Use non-mutating reads, searches, static analysis, and checks to ground the plan in the actual repository. Do not edit or write files, change configuration, run formatters or code generation that rewrites tracked files, commit, or otherwise carry out the plan. Prefer existing functions and patterns over new machinery.
 
@@ -538,41 +544,17 @@
               provider: spawn
               toolName: subagent
               modelSelectionSettings: true
-              backgroundMode: continuable
           - id: tool-subagent-fork
             name: '@deepseek-ai/dsh-tool-subagent'
             config:
               provider: fork
               toolName: subagent_fork
-              backgroundMode: continuable
-          - id: tool-subagent-codex
-            name: '@deepseek-ai/dsh-tool-subagent'
-            disabled: true
-            config:
-              provider: codex
-              toolName: subagent_codex
-              backgroundMode: one-shot
-              maxDepth: provider-managed
-          - id: tool-subagent-claude-code
-            name: '@deepseek-ai/dsh-tool-subagent'
-            disabled: true
-            config:
-              provider: claude-code
-              toolName: subagent_claude_code
-              backgroundMode: one-shot
-              maxDepth: provider-managed
           - id: workflow-ptc
             name: '@deepseek-ai/dsh-workflow-ptc'
             config:
               provider: spawn
           - id: tool-workflow
             name: '@deepseek-ai/dsh-tool-workflow'
-          - id: tool-ralph
-            name: '@deepseek-ai/dsh-tool-ralph'
-            disabled: true
-            config:
-              subagentProvider: spawn
-              maxRounds: 64
       - id: tool-ask-user
         name: '@deepseek-ai/dsh-tool-ask-user'
       - id: tool-todo
@@ -598,6 +580,9 @@
       - id: tool-plugin-manager
         name: '@deepseek-ai/dsh-plugin-manager/tools'
         disabled: !!js "!ctx.get('profileContext')"
+      - id: tool-plugin-market
+        name: '@deepseek-ai/dsh-plugin-market/tools'
+        disabled: !!js "!ctx.get('profileContext')"
 
 # -----------------------------------------------------------------------------
 # The bundle's own plugin row. The Host half (index.js) registers the bundled

package/host/subagent-model.js

@@ -9,14 +9,17 @@
  * result content. DSH's own UI shows a child's model nowhere, so this line is
  * the only place the parent learns which model actually runs each child.
  *
- * Shapes verified against dsh 0.2.1-alpha.1 sources:
- * - Tool result values: `{kind:"continuable", subagentId}` (background
- *   continuable — subagentId IS the durable child session id,
- *   subagents.startContinuable's childId), `{kind:"foreground", runId,
- *   output}` (SubagentRun.id equals the published child session id for local
- *   providers; remote providers mint parent-namespaced ids, which simply fail
- *   the cold read and pass through unannotated), and `{kind:"background",
- *   jobId}` (a jobs id, not a session id — skipped).
+ * Shapes verified against dsh 0.2.1-rc.1 sources:
+ * - Tool result values: `{kind:"activation", subagentId}` (dsh 0.2.1's
+ *   unified activation lifecycles — every subagent / subagent_fork call
+ *   returns immediately with the durable child session id) plus the two
+ *   dsh ≤0.2.0 shapes kept as fallbacks so the layer keeps working on older
+ *   installs: `{kind:"continuable", subagentId}` (background continuable —
+ *   subagentId IS the durable child session id) and `{kind:"foreground",
+ *   runId, output}` (SubagentRun.id equals the published child session id
+ *   for local providers; remote providers mint parent-namespaced ids, which
+ *   simply fail the cold read and pass through unannotated). One-shot
+ *   background-job shapes carry no session id and are skipped.
  * - `PostToolDecision` accept: `content` replaces the model-visible content
  *   and cannot coexist with a `value` replacement (the runtime throws on
  *   both), so only value-less accepts gain the line. `next()` is always
@@ -26,10 +29,12 @@
  */
 
 /** Child session id out of one subagent tool result value; null for shapes
- *  without a session id (one-shot background jobs) or malformed values. */
+ *  without a session id (one-shot background jobs) or malformed values.
+ *  "activation" is the dsh ≥0.2.1 unified shape; "continuable" and
+ *  "foreground" are the dsh ≤0.2.0 shapes kept for older installs. */
 function childSessionIdOf(value) {
 	if (value === null || typeof value !== "object") return null;
-	if (value.kind === "continuable" && typeof value.subagentId === "string" && value.subagentId !== "")
+	if ((value.kind === "activation" || value.kind === "continuable") && typeof value.subagentId === "string" && value.subagentId !== "")
 		return value.subagentId;
 	if (value.kind === "foreground" && typeof value.runId === "string" && value.runId !== "")
 		return value.runId;

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "dsh-switchman",
-	"version": "1.6.0",
+	"version": "1.7.1",
 	"description": "Autonomous Agent Teams doctrine for the four shipped DeepSeek Harness presets, plus language preferences, dispatch pools with model ranking, and context watermark control.",
 	"license": "MIT",
 	"keywords": [
@@ -25,7 +25,7 @@
 	"icon": "./icon.svg",
 	"engines": {
 		"node": ">=24",
-		"dsh": ">=0.2.0-rc.0"
+		"dsh": ">=0.2.1-rc.1"
 	},
 	"exports": {
 		".": "./index.js",
@@ -49,8 +49,8 @@
 		"icon.svg",
 		"cordis.patch.yml"
 	],
-	"dependencies": {
-		"@deepseek-ai/schemastery": "3.18.4"
+	"peerDependencies": {
+		"@deepseek-ai/schemastery": "^3.18.1"
 	},
 	"dsh": {
 		"manifestVersion": 1,