weir-harness 0.0.0-stage → 1.1.0

package/README.md

@@ -1,3 +1,40 @@
-# Temporary Holding Version
+# weir-harness
 
-This version is a temporary placeholder for this package. An operational version to replace this has been submitted for review and is awaiting a staged release.
+> 文档导航：产品门面见仓库根 [README.md](../../README.md)；各特性的行为/配置/设计/失败语义见 [docs/features/](../../docs/features/README.md)；版本记录见 [CHANGELOG.md](../../CHANGELOG.md)；开发纪律见 [AGENTS.md](../../AGENTS.md)。
+
+Weir — 以 Orchestrator 为中心的 DSH agent 预设 bundle。声明预设 `weir`（GUI 显示名 "Weir"），落地七项设计：
+
+| 模块 | 功能 |
+|---|---|
+| `weir-harness/core` | Orchestrator 协作纪律提示词 section（`orchestrator:doctrine`） |
+| `weir-harness/intent-gate` | 意图门：关键词命中 → 上下文注入 / 推理提档（`agent/pre-step` + `agent/request` 瀑布）；可选语义分类器（regex 默认 / llm sidecar / jev 实验，fail-open + 审计） |
+| `weir-harness/delegate` | `delegate` 工具：类别路由（模型链解析）+ 精选只读代理（finder/scholar/advisor，bash 受只读白名单守卫）+ 批量 + 后台 job 道 + ESCALATE 契约 + 受监督分组（`group` 二元终态 + 异常续推 + 合并报告 + `resume_agent`/`terminate_agent`） |
+| `weir-harness/todo-driver` | todo 空转续推：`turn/end` reason 区分用户打断/供应商错误，`stop_continuation` 逃生舱 |
+| `weir-harness/context-guard` | 上下文压力阈值：软阈值提示 + `compact_context` 择时压缩 + 硬阈值边界强制 + 压缩后续推 |
+| `weir-harness/hashline-edit` | read 锚点增强（`N#XX|`）+ fail-closed `hash_edit` 工具（预设默认唯一编辑工具，`hideStockEdit: false` 可回归共存） |
+| `weir-harness/lsp` | 自研零依赖 LSP 客户端：四个只读语义工具（diagnostics/definition/references/symbols）；`lsp.enabled` 能力总闸（默认关），会话面板开关 + `/lsp` 命令 + `lsp` 工具按会话启用（`weirLsp` 投影持久化）；13 族社区配方注册表 + `lspServers` 用户自定义服务器 + 扩展目录可执行解析 |
+| （内建采用） | 拉取式后台通知由 DSH 内建 job 道承担；本预设配置 `maxConsecutiveWakes: 8` |
+
+技能（`skills/`，经 `customSkillDirs` 挂载）：`deep-work`、`research`、`review-work`、`debugging`、`git-master`、`refactor`、`programming`（含 references/）、`remove-ai-slops`、`work-with-pr`、`remove-deadcode`（英文正文；模板层英文、实例内容跟随会话语言）。
+
+另含 profile 层 `weir-harness/settings` 行：Weir 设置页（Settings 应用自动生成）+ host 层 `weirSettings` 服务（生效配置分层：模块默认 ← 行 config ← 设置值）。schema 依赖 vendored 于 `src/vendor/`（见该目录 THIRD-PARTY.md）。
+
+## 安装
+
+要求 DeepSeek Harness 版本 `>=0.2.0-rc.2 <0.3.0`（不兼容的运行时会在安装前被明确拒绝）。
+
+CLI 一条命令：
+
+```sh
+dsh plugin --profile web add weir-harness
+```
+
+或 Harness 会话内：`plugin_manager` 的 `install_bundle`，target 填 `weir-harness`（可带 `@版本`）。
+
+安装后在新建会话的预设选择器中选择 **Weir**；GUI 设置页出现「Weir」专页。升级到最新版：同一安装命令改填 `weir-harness@latest`。
+
+## 开发
+
+纯 ESM JavaScript，无构建步骤（`src/**/*.js` 即运行时输入）。测试：`node --test "test/**/*.test.js"`；静态检查：`pnpm run check`（tsc checkJs 覆盖强类型核心；适配层经单测覆盖）。
+
+开发者 link 安装：Harness 会话内 `plugin_manager install_bundle` 以本目录绝对路径安装；重复安装用 `set_bundle` 禁用/启用循环重应用。

package/cordis.patch.yml

@@ -0,0 +1,496 @@
+# Agent preset weir: one `@deepseek-ai/dsh-agent-preset` declaration inserted
+# by this bundle's patch. The preset is edited through bundle updates; the Web
+# editor is read-only for preset rows.
+# Presets own discovery and the skill tool in every host composition.
+- id: skill-filesystem
+  disabled: true
+- id: tool-skill
+  disabled: true
+- insert:
+    # Profile-level settings row: one-stop Weir configuration page in the
+    # Settings app (auto-generated from the declared schema) plus the
+    # host-plane weirSettings service the preset modules overlay with.
+    - id: weir-settings
+      name: 'weir-harness/settings'
+      config:
+        # Product defaults, mirrored 1:1 from the preset modules' code
+        # defaults so the settings page shows EFFECTIVE values (the row is the
+        # composition base; user edits override it). Keep in sync with the
+        # module defaults on every change.
+        intentGateClassifier: regex
+        intentGateTimeoutMs: 1500
+        supervisionMaxRetries: 5
+        supervisionInitialBackoffMs: 30000
+        supervisionMaxBackoffMs: 300000
+        todoEnabled: true
+        todoMaxConsecutive: 8
+        todoErrorRetryMax: 5
+        todoErrorBackoffBaseMs: 30000
+        todoErrorBackoffCapMs: 300000
+        guardEnabled: true
+        guardSoftThreshold: 0.72
+        guardHardThreshold: 0.88
+        hashlineHideStockEdit: true
+        # Worktree lanes (git-worktree-lanes): capability switch, lane root,
+        # active-lane cap, automatic dependency setup, lane-watch timeout, and
+        # the global auto-approve default (manual / auto-keep / auto-clean).
+        worktreeEnabled: true
+        worktreeRoot: .weir/worktrees
+        worktreeMaxActive: 4
+        worktreeAutoSetup: true
+        worktreeWatchTimeoutMinutes: 360
+        worktreeAutoApprove: auto-clean
+        editLockEnabled: false
+        # Message-driven auto-resume (edit-lock-auto-resume-on-message): a genuine
+        # user message resumes a stopped session and confirms its retained files.
+        # Read live per message, so toggling applies without a restart.
+        editLockAutoResume: true
+        editLockStaleSweep: true
+        robashEnabled: true
+        # The five whitelist tables are NOT declared here on purpose. A patch
+        # row's `config` is replaced WHOLE rather than deep-merged, so a default
+        # that rides this row is discarded the moment a profile declares its own
+        # settings row — which made every later whitelist addition unreachable
+        # for any profile that had ever edited a list. The product defaults now
+        # live in the bundle's `whitelist-defaults.json`, which the plugin reads
+        # itself at runtime; the settings keys of the same names ADD to them.
+        lspEnabled: false
+        notifyEnabled: true
+        notifyOnComplete: true
+        notifyOnAttention: true
+        notifyMinTurnSeconds: 15
+        notifySound: true
+        notifyForeground: skip
+        # Session blackboard: write-token TTL (minutes); expiry releases the
+        # key and notifies its subscribers. Volatile — edits apply live.
+        blackboardWriteTokenTtlMinutes: 60
+    # System notifications (profile level, so every session is covered, not only
+    # Weir-preset ones): approval / question / plan-review / failure / long-turn
+    # completion. Pure observer of session events; settings overlay is read live.
+    - id: weir-notify
+      name: 'weir-harness/notify'
+    # Capability read channel (silent-capability-reads): the plugin-owned
+    # typert remote serving the Badge/manager READ verbs (receipt / list /
+    # conditions) so reads never touch the session log. Host layer by
+    # necessity: the typert gateway resolves the service from the host root
+    # (S27); the preset-layer skill-selection feeds its faces through the
+    # module-level bridge. Inert with a warning when typert is absent.
+    - id: weir-capability-remote
+      name: 'weir-harness/capability-remote'
+    # Session blackboard panel data channel (design D8): plugin-owned typert
+    # remote serving list/read/apply/write/remove so panel reads never touch
+    # the session log. Host layer for the same S27 reason as the capability
+    # remote above (gateway resolves from the host root); the preset-layer
+    # blackboard plugin feeds the kernel/TTL faces through the module-level
+    # bridge. Inert with a warning when typert is absent.
+    - id: weir-blackboard-remote
+      name: 'weir-harness/blackboard-remote'
+    # Browser half of the same package: the Weir settings page registered
+    # as a top-level section of the global Settings panel (./client export,
+    # dsh.client manifest).
+    - id: ui-weir-settings
+      name: 'weir-harness'
+    # Isolated reusable model picker client module (own package + own client
+    # export): the settings page requires it and renders it inside an error
+    # boundary, so a picker failure can never blank the page.
+    - id: ui-weir-model-picker
+      name: 'weir-model-picker'
+    - id: preset-weir
+      name: '@deepseek-ai/dsh-agent-preset'
+      config:
+        id: weir
+        name: Weir
+        description: Orchestrator-centric harness — intent gate, category delegation, todo driver, pull-based background jobs, auto-compaction, hashline edit.
+        order: 5
+        plugins:
+          - id: persona
+            name: '@deepseek-ai/dsh-persona'
+            config:
+              prefix: >-
+                You are the Orchestrator, the user's single point of collaboration, powered by the {{model}} model.
+                You understand intent, decompose work, delegate to specialist sub-agents, choose parallel fan-out
+                or sequenced waiting, verify evidence, and report results.
+              suffix: Your working directory is {{cwd}}.
+          - id: weir-core
+            name: 'weir-harness/core'
+          - id: agent-instructions
+            name: '@deepseek-ai/dsh-agent-instructions'
+            config:
+              maxBytes: 65536
+          - id: tool-bash
+            name: '@deepseek-ai/dsh-tool-bash'
+            disabled: !!js process.platform === 'win32'
+          - id: tool-pwsh
+            name: '@deepseek-ai/dsh-tool-pwsh'
+            disabled: !!js process.platform !== 'win32'
+          - id: tool-fs
+            name: '@deepseek-ai/dsh-tool-fs'
+          - id: tool-fs-search
+            name: '@deepseek-ai/dsh-tool-fs-search'
+            config:
+              sampleOverCapGlobResults: false
+          - id: tool-jobs
+            name: '@deepseek-ai/dsh-tool-jobs'
+            config:
+              # F5 拉取式后台：结算通知已内建为紧凑 + 边界注入 + 拉取式读取；
+              # 这里仅限制连续唤醒次数，防止通知唤醒风暴。
+              maxConsecutiveWakes: 8
+          # Weir capabilities: skill selection + managed MCP. The manager
+          # publishes weirMcpGate / weirMcpManager via reflect; preset
+          # services must sit in an isolate realm or the registry rejects the
+          # whole preset ("Preset services require isolate realms"). The
+          # consumer (skill-selection reads weirMcpManager for /capabilities
+          # list) shares the group (S11).
+          - id: capabilities
+            name: cordis:group
+            group: true
+            isolate:
+              weirMcpGate: true
+              weirMcpManager: true
+            config:
+              - id: weir-skill-selection
+                name: 'weir-harness/skill-selection'
+                config:
+                  # preset-skill-applicability (D2): profile-supplemental roots
+                  # used ONLY to classify apply-missing names as known-but-
+                  # inapplicable (skip with report) vs genuinely unknown — the
+                  # same creative root the weir-creative row enumerates.
+                  supplementalSkillDirs:
+                    - !!js (() => { try { const p = process.getBuiltinModule('node:path'); return p.join(p.dirname(process.getBuiltinModule('node:module').createRequire(baseUrl).resolve('@deepseek-ai/dsh-agent-preset/package.json')), 'skills') } catch { return process.getBuiltinModule('node:path').join(process.getBuiltinModule('node:os').tmpdir(), 'dsh-agent-preset-skills-unavailable') } })()
+              # Weir-managed MCP: runtime per-server isolated mounts with the
+              # proxy facade, admission gate and adopt flow (task 8.2+).
+              - id: weir-mcp-manager
+                name: 'weir-harness/mcp-manager'
+          - id: tool-skill
+            name: '@deepseek-ai/dsh-tool-skill'
+          - id: planning
+            name: cordis:group
+            group: true
+            isolate:
+              planMode: true
+            config:
+              - id: plan-mode
+                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.
+
+                    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.
+
+                    The tool catalog stays the same across modes for request-cache stability. These plan-mode rules override any later tool description or guidance that suggests using mutation tools; those tools remain listed to keep the tool catalog unchanged. Do not use todo_write to track this planning phase: it tracks implementation after an approved plan, while the plan itself belongs in exit_plan_mode.
+
+                    Resolve discoverable facts by inspection. Use ask_user_question only for user-owned choices or material ambiguity that inspection cannot answer. Do not ask the user where code lives or how current behavior works when you can find out.
+
+                    Make the plan decision-complete: state the goal and success criteria; group implementation changes by subsystem; identify public API, schema, and data-flow changes; cover edge cases, failure modes, tests, acceptance criteria, and explicit assumptions. Keep it concise enough to review but detailed enough that another engineer can implement it without making design decisions.
+
+                    When ready, call exit_plan_mode with the complete plan markdown, starting with a # title. Make exit_plan_mode the only and final tool call in that assistant response: it presents the plan for approval, and implementation begins only in a later step after approval. Do not paste the final plan as a plain reply or ask "should I proceed?" through prose or ask_user_question. If review rejects it, incorporate the feedback and present again. If the review channel is unavailable or aborted, stay in plan mode and ask the user to switch modes manually; do not proceed with implementation.
+          - id: compaction
+            name: cordis:group
+            group: true
+            isolate:
+              compaction: true
+              toolResultPruner: true
+            config:
+              - id: compaction-basic
+                name: '@deepseek-ai/dsh-compaction-basic'
+              - id: command-compact
+                name: '@deepseek-ai/dsh-command-compact'
+              - id: tool-result-pruner
+                name: '@deepseek-ai/dsh-compaction-tool-result-pruner'
+                config:
+                  thresholdChars: 8192
+                  headChars: 4096
+                  tailChars: 1024
+              # F6 自动压缩：上下文压力阈值 + 模型择时压缩 + 边界强制 + 压缩后续推。
+              # `compaction` 服务被本组 isolate：消费者必须与提供者同 realm（否则
+              # inject 永久等待，预设加载失败）。
+              - id: context-guard
+                name: 'weir-harness/context-guard'
+          - id: delegation
+            name: cordis:group
+            group: true
+            isolate:
+              workflowEngine: true
+              # 编辑锁服务：提供者 edit-lock 与消费者 todo-driver / hashline-edit /
+              # lsp 同在本组。预设服务必须 isolate，否则预设加载以
+              # "Preset services require isolate realms" 失败。
+              weirEditLock: true
+              # Worktree lanes service: provider (worktree) and consumer
+              # (delegate) share this realm (S24 / S26 S-A).
+              weirWorktreeLanes: true
+              # Session blackboard: provider + tool consumers (slice 1) and the
+              # delegate write-contract consumer (slice 2) share this realm.
+              weirBlackboard: true
+            config:
+              - id: tool-subagent-control
+                name: '@deepseek-ai/dsh-tool-subagent-control'
+              - id: tool-subagent-list-agents
+                name: '@deepseek-ai/dsh-tool-subagent-control/list-agents'
+              - id: tool-subagent
+                name: '@deepseek-ai/dsh-tool-subagent'
+                config:
+                  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: 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
+              # F1 意图门：关键词意图 → 上下文注入 / 推理提档
+              - id: intent-gate
+                name: 'weir-harness/intent-gate'
+              # Worktree 车道：宿主状态机 + 车道工具 + /worktree 命令 + 面板端点；
+              # 提供 weirWorktreeLanes 服务（delegate 消费：绑定、车道守卫、结算）
+              - id: worktree
+                name: 'weir-harness/worktree'
+              # F3 分类委派：类别路由 + 精选只读代理 + 后台 job 道
+              - id: delegate
+                name: 'weir-harness/delegate'
+              # 会话黑板：会话级 KV + 写入权仲裁 + 五个模型工具（提供 weirBlackboard 服务）
+              - id: blackboard
+                name: 'weir-harness/blackboard'
+              # F4 todo 续推：空转续推 + 用户打断/供应商错误语义 + 逃生舱
+              - id: todo-driver
+                name: 'weir-harness/todo-driver'
+              # 编辑锁（实验，默认关）：设置 editLockEnabled 开启并重启后生效。
+              # 必须排在 hashline-edit / lsp 之前，二者在挂载时捕获锁服务。
+              - id: edit-lock
+                name: 'weir-harness/edit-lock'
+              # F7 锚点编辑：read 锚点增强 + fail-closed hash_edit 工具
+              - id: hashline-edit
+                name: 'weir-harness/hashline-edit'
+                config:
+                  # hash_edit 为预设默认唯一编辑工具；配置 false 回归与 stock edit 共存
+                  hideStockEdit: true
+              # LSP 语义工具：能力总闸（默认关）。设置 lspEnabled 开启后，会话内
+              # 由面板开关或 `lsp` 工具按会话启用/禁用（新会话默认关）。
+              - id: lsp
+                name: 'weir-harness/lsp'
+          - id: tool-ask-user
+            name: '@deepseek-ai/dsh-tool-ask-user'
+          - id: tool-todo
+            name: '@deepseek-ai/dsh-tool-todo'
+            config:
+              allowParallelInProgress: true
+          - id: tool-web
+            name: '@deepseek-ai/dsh-tool-web'
+            config:
+              fetch: true
+              searchTimeoutMs: 60000
+          - id: present
+            name: '@deepseek-ai/dsh-tool-present'
+          - id: tool-plugin-manager
+            name: '@deepseek-ai/dsh-plugin-manager/tools'
+            disabled: true
+    # Weir 创造模式 (weir-creative): the full `weir` composition above
+    # fused with the DSH creative mode (cordis preset) deltas — runtime
+    # inspection tools, persistent plugin management, and the Cordis
+    # development skills. MAINTENANCE RULE: any change to the `weir`
+    # preset's plugin list MUST be mirrored here (guarded by
+    # test/preset-creative.test.js); only the three documented deltas may
+    # diverge.
+    - id: preset-weir-creative
+      name: '@deepseek-ai/dsh-agent-preset'
+      config:
+        id: weir-creative
+        name: Weir 创造模式
+        description: Weir creative mode — the full Weir harness fused with the DSH creative mode for debugging and experimenting with DSH itself — runtime inspection (cordis_inspect_*), persistent plugin management, and the Cordis plugin/preset development skills.
+        order: 6
+        plugins:
+          - id: persona
+            name: '@deepseek-ai/dsh-persona'
+            config:
+              prefix: >-
+                You are the Orchestrator, the user's single point of collaboration, powered by the {{model}} model.
+                You understand intent, decompose work, delegate to specialist sub-agents, choose parallel fan-out
+                or sequenced waiting, verify evidence, and report results.
+              suffix: Your working directory is {{cwd}}.
+          - id: weir-core
+            name: 'weir-harness/core'
+          # Creative-only: how to use the fused tooling (inspection order,
+          # plugin-management discipline, development skills, composition
+          # rules). The `weir` preset does not mount this section.
+          - id: creative-guide
+            name: 'weir-harness/creative-guide'
+          - id: agent-instructions
+            name: '@deepseek-ai/dsh-agent-instructions'
+            config:
+              maxBytes: 65536
+          - id: tool-bash
+            name: '@deepseek-ai/dsh-tool-bash'
+            disabled: !!js process.platform === 'win32'
+          - id: tool-pwsh
+            name: '@deepseek-ai/dsh-tool-pwsh'
+            disabled: !!js process.platform !== 'win32'
+          - id: tool-fs
+            name: '@deepseek-ai/dsh-tool-fs'
+          - id: tool-fs-search
+            name: '@deepseek-ai/dsh-tool-fs-search'
+            config:
+              sampleOverCapGlobResults: false
+          - id: tool-jobs
+            name: '@deepseek-ai/dsh-tool-jobs'
+            config:
+              maxConsecutiveWakes: 8
+          # Same capabilities group as `weir` (S11 realm structure), with
+          # the creative delta: customSkillDirs resolves the Cordis
+          # development skills shipped inside @deepseek-ai/dsh-agent-preset,
+          # and baselineScopes admits them into the first-session baseline.
+          - id: capabilities
+            name: cordis:group
+            group: true
+            isolate:
+              weirMcpGate: true
+              weirMcpManager: true
+            config:
+              - id: weir-skill-selection
+                name: 'weir-harness/skill-selection'
+                config:
+                  supplementalSkillDirs:
+                    - !!js (() => { try { const p = process.getBuiltinModule('node:path'); return p.join(p.dirname(process.getBuiltinModule('node:module').createRequire(baseUrl).resolve('@deepseek-ai/dsh-agent-preset/package.json')), 'skills') } catch { return process.getBuiltinModule('node:path').join(process.getBuiltinModule('node:os').tmpdir(), 'dsh-agent-preset-skills-unavailable') } })()
+                  customSkillDirs:
+                    # Same resolution as the cordis preset's skill-filesystem
+                    # row, wrapped fail-safe: an unresolvable package degrades
+                    # to a nonexistent path (enumerated as an empty root),
+                    # never to a profile-load failure.
+                    - !!js (() => { try { const p = process.getBuiltinModule('node:path'); return p.join(p.dirname(process.getBuiltinModule('node:module').createRequire(baseUrl).resolve('@deepseek-ai/dsh-agent-preset/package.json')), 'skills') } catch { return process.getBuiltinModule('node:path').join(process.getBuiltinModule('node:os').tmpdir(), 'dsh-agent-preset-skills-unavailable') } })()
+                  baselineScopes:
+                    - weir-builtin
+                    - custom
+              - id: weir-mcp-manager
+                name: 'weir-harness/mcp-manager'
+          - id: tool-skill
+            name: '@deepseek-ai/dsh-tool-skill'
+          - id: planning
+            name: cordis:group
+            group: true
+            isolate:
+              planMode: true
+            config:
+              - id: plan-mode
+                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.
+
+                    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.
+
+                    The tool catalog stays the same across modes for request-cache stability. These plan-mode rules override any later tool description or guidance that suggests using mutation tools; those tools remain listed to keep the tool catalog unchanged. Do not use todo_write to track this planning phase: it tracks implementation after an approved plan, while the plan itself belongs in exit_plan_mode.
+
+                    Resolve discoverable facts by inspection. Use ask_user_question only for user-owned choices or material ambiguity that inspection cannot answer. Do not ask the user where code lives or how current behavior works when you can find out.
+
+                    Make the plan decision-complete: state the goal and success criteria; group implementation changes by subsystem; identify public API, schema, and data-flow changes; cover edge cases, failure modes, tests, acceptance criteria, and explicit assumptions. Keep it concise enough to review but detailed enough that another engineer can implement it without making design decisions.
+
+                    When ready, call exit_plan_mode with the complete plan markdown, starting with a # title. Make exit_plan_mode the only and final tool call in that assistant response: it presents the plan for approval, and implementation begins only in a later step after approval. Do not paste the final plan as a plain reply or ask "should I proceed?" through prose or ask_user_question. If review rejects it, incorporate the feedback and present again. If the review channel is unavailable or aborted, stay in plan mode and ask the user to switch modes manually; do not proceed with implementation.
+          - id: compaction
+            name: cordis:group
+            group: true
+            isolate:
+              compaction: true
+              toolResultPruner: true
+            config:
+              - id: compaction-basic
+                name: '@deepseek-ai/dsh-compaction-basic'
+              - id: command-compact
+                name: '@deepseek-ai/dsh-command-compact'
+              - id: tool-result-pruner
+                name: '@deepseek-ai/dsh-compaction-tool-result-pruner'
+                config:
+                  thresholdChars: 8192
+                  headChars: 4096
+                  tailChars: 1024
+              - id: context-guard
+                name: 'weir-harness/context-guard'
+          - id: delegation
+            name: cordis:group
+            group: true
+            isolate:
+              workflowEngine: true
+              weirEditLock: true
+              weirWorktreeLanes: true
+              weirBlackboard: true
+            config:
+              - id: tool-subagent-control
+                name: '@deepseek-ai/dsh-tool-subagent-control'
+              - id: tool-subagent-list-agents
+                name: '@deepseek-ai/dsh-tool-subagent-control/list-agents'
+              - id: tool-subagent
+                name: '@deepseek-ai/dsh-tool-subagent'
+                config:
+                  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: 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: intent-gate
+                name: 'weir-harness/intent-gate'
+              - id: worktree
+                name: 'weir-harness/worktree'
+              - id: delegate
+                name: 'weir-harness/delegate'
+              - id: blackboard
+                name: 'weir-harness/blackboard'
+              - id: todo-driver
+                name: 'weir-harness/todo-driver'
+              # Must precede hashline-edit / lsp (mount-time capture).
+              - id: edit-lock
+                name: 'weir-harness/edit-lock'
+              - id: hashline-edit
+                name: 'weir-harness/hashline-edit'
+                config:
+                  hideStockEdit: true
+              - id: lsp
+                name: 'weir-harness/lsp'
+          - id: tool-ask-user
+            name: '@deepseek-ai/dsh-tool-ask-user'
+          - id: tool-todo
+            name: '@deepseek-ai/dsh-tool-todo'
+            config:
+              allowParallelInProgress: true
+          - id: tool-web
+            name: '@deepseek-ai/dsh-tool-web'
+            config:
+              fetch: true
+              searchTimeoutMs: 60000
+          # Creative delta 1: read-only Cordis runtime inspection tools.
+          - id: tool-cordis
+            name: '@deepseek-ai/dsh-tool-cordis'
+          - id: present
+            name: '@deepseek-ai/dsh-tool-present'
+          # Creative delta 2: persistent plugin management, enabled under a
+          # profile context exactly like the shipped creative mode.
+          - id: tool-plugin-manager
+            name: '@deepseek-ai/dsh-plugin-manager/tools'
+            disabled: !!js "!ctx.get('profileContext')"