harnessed 4.6.0 → 4.8.0

package/workflows/disciplines/operational.yaml

@@ -12,7 +12,7 @@ auto_enforce: true
 rules:
   - id: biome-preempt
     description: |
-      TS/JS commit 前必跑 `pnpm exec biome check --write` (3 CI-red recurrences Phase 2.1.1/2.2/2.3 project memory)
+      Before any TS/JS commit you must run `pnpm exec biome check --write` (3 CI-red recurrences Phase 2.1.1/2.2/2.3 project memory)
     enforcement: auto-fix
     trigger: |
       phase.type == 'commit' AND
@@ -23,8 +23,8 @@ rules:
 
   - id: a7-adr-conservation
     description: |
-      新 ADR 不动旧 ADR main body,baseline tag iterate,CI A7 step 守恒
-      (sister Phase 1.3-2.6 ADR errata 路径)
+      A new ADR does not touch the main body of old ADRs; iterate via baseline tag, conserved by the CI A7 step
+      (sister Phase 1.3-2.6 ADR errata path)
     enforcement: warn
     trigger: phase.type == 'commit' AND changed_files contains 'docs/adr/'
     check_method: external-cmd
@@ -40,14 +40,14 @@ rules:
 
   - id: no-skip-hooks
     description: |
-      不允许 --no-verify / --no-gpg-sign 等 skip hook flag,除非用户明确要求
+      Skip-hook flags such as --no-verify / --no-gpg-sign are not allowed, unless the user explicitly requests them
     enforcement: halt
     trigger: cmd.type == 'git-commit' AND args contains '--no-verify'
     check_method: regex
 
   - id: destructive-ops-explicit
     description: |
-      Destructive ops (git reset --hard / push --force / rm -rf / drop table) 必须用户 explicit 确认
+      Destructive ops (git reset --hard / push --force / rm -rf / drop table) require explicit user confirmation
     enforcement: halt
     trigger: cmd.is_destructive == true
     check_method: heuristic
@@ -55,7 +55,7 @@ rules:
 
   - id: authorization-not-transitive
     description: |
-      用户 approve 一次 push ≠ 全程 approve;每个 destructive / shared-state 操作单独确认
+      A user approving one push ≠ approval for the whole session; each destructive / shared-state operation needs separate confirmation
     enforcement: warn
     trigger: cmd.requires_approval == true AND session.has_prior_approval == true
     check_method: heuristic

package/workflows/disciplines/operational.zh-Hans.yaml

@@ -0,0 +1,79 @@
+# workflows/disciplines/operational.yaml
+# 操作纪律 — bundled discipline (biome preempt + commit safety +
+# A7 ADR + no-skip-hooks + authorization-not-transitive + destructive-ops-explicit)
+# Phase v3.0-3.3 W0.4 ship per D-09
+# Snapshot of CLAUDE.md as of v3.0 ship date (K7 mitigation, verbatim)
+
+schema_version: harnessed.discipline.v1
+discipline: operational
+enforcement_layer: commit
+auto_enforce: true
+
+rules:
+  - id: biome-preempt
+    description: |
+      TS/JS commit 前必跑 `pnpm exec biome check --write` (3 CI-red recurrences Phase 2.1.1/2.2/2.3 project memory)
+    enforcement: auto-fix
+    trigger: |
+      phase.type == 'commit' AND
+      changed_files matches '\\.(ts|tsx|js|mjs)$'
+    check_method: external-cmd
+    auto_fix_cmd: 'corepack pnpm exec biome check --write'
+    # NEW src/discipline/enforcement/before-commit.ts hook; exit 0 才允许 commit
+
+  - id: a7-adr-conservation
+    description: |
+      新 ADR 不动旧 ADR main body,baseline tag iterate,CI A7 step 守恒
+      (sister Phase 1.3-2.6 ADR errata 路径)
+    enforcement: warn
+    trigger: phase.type == 'commit' AND changed_files contains 'docs/adr/'
+    check_method: external-cmd
+    auto_fix_cmd: 'scripts/check-adr-conservation.sh'
+
+  - id: no-push-without-approval
+    description: |
+      NEVER push to remote without user explicit approval (project CLAUDE.md commit safety)
+    enforcement: halt
+    trigger: cmd.type == 'git-push'
+    check_method: external-cmd
+    # ralph-loop / subagent 自动调 git push → halt;user explicit `git push` 在主 session 允许
+
+  - id: no-skip-hooks
+    description: |
+      不允许 --no-verify / --no-gpg-sign 等 skip hook flag,除非用户明确要求
+    enforcement: halt
+    trigger: cmd.type == 'git-commit' AND args contains '--no-verify'
+    check_method: regex
+
+  - id: destructive-ops-explicit
+    description: |
+      Destructive ops (git reset --hard / push --force / rm -rf / drop table) 必须用户 explicit 确认
+    enforcement: halt
+    trigger: cmd.is_destructive == true
+    check_method: heuristic
+    # heuristic — 命令含 'rm -rf' / 'reset --hard' / '--force' / 'DROP TABLE' / 'delete from'
+
+  - id: authorization-not-transitive
+    description: |
+      用户 approve 一次 push ≠ 全程 approve;每个 destructive / shared-state 操作单独确认
+    enforcement: warn
+    trigger: cmd.requires_approval == true AND session.has_prior_approval == true
+    check_method: heuristic
+
+  # v3.6.0 Phase 3 — transparent-skip discipline (P0b 下半,Audit § fallback
+  # 三条铁律 "拿不准 → 倾向跳过 + 透明声明"). prompt-inject method delivered
+  # via src/workflow/run.ts TRANSPARENT_SKIP_RULES const appended to
+  # criticalSystemReminder_EXPERIMENTAL alongside ESCALATION_RULES (v3.5.0
+  # Phase 2 sister). Spawned subagent reads the rule + emits the verbatim
+  # skip message when gate context is ambiguous instead of silent execution.
+  - id: transparent-skip-on-low-confidence
+    description: |
+      When a phase gate condition cannot be confidently evaluated (e.g.
+      expected context fields missing, default-valued, or contradictory),
+      prefer skip + transparent explanation over silent execution.
+      Verbatim format: "这次跳过了 <phase>, 因为 <reason>. 如果你认为
+      需要请明说" / English: "Skipped <phase> because <reason>. Tell me
+      if you actually need it."
+    enforcement: warn
+    trigger: gate_evaluation_low_confidence
+    check_method: prompt-inject

package/workflows/disciplines/output-style.yaml

@@ -10,7 +10,7 @@ auto_enforce: true
 
 rules:
   - id: bluf-conclusion-first
-    description: 结论先行 (BLUF),每段先一句给答案/判断/推荐,再展开理由
+    description: Conclusion first (BLUF); each paragraph leads with one sentence giving the answer/judgment/recommendation, then expands the reasoning
     enforcement: warn
     trigger: always-on
     check_method: heuristic
@@ -18,7 +18,7 @@ rules:
     # 失败 → warn 'BLUF missing: first sentence is not a conclusion'
 
   - id: no-sycophantic-open-close
-    description: 删除"好问题/太棒了/完美/希望对你有帮助/还需要别的吗"等开场闭合套话
+    description: Remove opening/closing filler phrases like "good question / awesome / perfect / hope this helps / anything else"
     enforcement: auto-fix
     trigger: always-on
     check_method: regex
@@ -26,14 +26,14 @@ rules:
     # NEW src/discipline/enforcement/strip-sycophantic.ts — regex strip 13 known phrase
 
   - id: no-emoji-unless-requested
-    description: 禁用 emoji,除非用户明确要求 (项目 markdown 文件如 README/PR/CHANGELOG 不约束)
+    description: No emoji unless the user explicitly requests it (project markdown files such as README/PR/CHANGELOG are not constrained)
     enforcement: warn
     trigger: response.target == 'chat' AND user.requested_emoji == false
     check_method: regex
     # regex /\p{Emoji_Presentation}/u; 命中 + trigger fires → warn
 
   - id: no-em-dash
-    description: 禁用 em-dash (—— / —);替代:补充用括号/冒号,并列用顿号,转折单独成句
+    description: No em-dash (—— / —); instead use parentheses or colons for asides, commas for lists, and a separate sentence for contrast
     enforcement: auto-fix
     trigger: response.target == 'chat'
     check_method: regex
@@ -41,13 +41,13 @@ rules:
     # auto-replace '——' → ',  ' (单段内逗号);'—' → ':' OR ','
 
   - id: precise-quantifier
-    description: 量词精确,能给具体数字/文件名/行号/commit hash 就不用"一些/几个/多个"
+    description: Precise quantifiers; when a concrete number/filename/line number/commit hash is available, do not say "some/a few/several"
     enforcement: info
     trigger: always-on
     check_method: llm-judge
 
   - id: no-end-recap
-    description: 不做结尾总结,除非用户明确要求"汇总/recap"
+    description: No closing summary unless the user explicitly asks to "summarize/recap"
     enforcement: warn
     trigger: response.target == 'chat'
     check_method: heuristic
@@ -55,7 +55,7 @@ rules:
     # → warn 'redundant end recap'
 
   - id: no-empty-continuation-question
-    description: 禁空洞续作询问 (要不要我帮你 X?/还需要别的吗?);保留工作流必要下一步指引
+    description: No empty continuation questions ("want me to help you X?" / "anything else?"); keep workflow-necessary next-step guidance
     enforcement: warn
     trigger: response.target == 'chat'
     check_method: regex

package/workflows/disciplines/output-style.zh-Hans.yaml

@@ -0,0 +1,62 @@
+# workflows/disciplines/output-style.yaml
+# 对话回答风格 — bundled discipline (BLUF + no-sycophantic + no-emoji + no-em-dash + 量词精确 + no-end-recap + no-empty-continuation-question)
+# Phase v3.0-3.3 W0.4 ship per D-09
+# Snapshot of CLAUDE.md as of v3.0 ship date (K7 mitigation, verbatim)
+
+schema_version: harnessed.discipline.v1
+discipline: output-style
+enforcement_layer: output
+auto_enforce: true
+
+rules:
+  - id: bluf-conclusion-first
+    description: 结论先行 (BLUF),每段先一句给答案/判断/推荐,再展开理由
+    enforcement: warn
+    trigger: always-on
+    check_method: heuristic
+    # heuristic — 第一句长度 ≤ 100 char + 含答案动词 ('是' / '不是' / '推荐' / '应该' / 'use X')
+    # 失败 → warn 'BLUF missing: first sentence is not a conclusion'
+
+  - id: no-sycophantic-open-close
+    description: 删除"好问题/太棒了/完美/希望对你有帮助/还需要别的吗"等开场闭合套话
+    enforcement: auto-fix
+    trigger: always-on
+    check_method: regex
+    auto_fix_cmd: 'strip-sycophantic'
+    # NEW src/discipline/enforcement/strip-sycophantic.ts — regex strip 13 known phrase
+
+  - id: no-emoji-unless-requested
+    description: 禁用 emoji,除非用户明确要求 (项目 markdown 文件如 README/PR/CHANGELOG 不约束)
+    enforcement: warn
+    trigger: response.target == 'chat' AND user.requested_emoji == false
+    check_method: regex
+    # regex /\p{Emoji_Presentation}/u; 命中 + trigger fires → warn
+
+  - id: no-em-dash
+    description: 禁用 em-dash (—— / —);替代:补充用括号/冒号,并列用顿号,转折单独成句
+    enforcement: auto-fix
+    trigger: response.target == 'chat'
+    check_method: regex
+    auto_fix_cmd: 'replace-em-dash'
+    # auto-replace '——' → ',  ' (单段内逗号);'—' → ':' OR ','
+
+  - id: precise-quantifier
+    description: 量词精确,能给具体数字/文件名/行号/commit hash 就不用"一些/几个/多个"
+    enforcement: info
+    trigger: always-on
+    check_method: llm-judge
+
+  - id: no-end-recap
+    description: 不做结尾总结,除非用户明确要求"汇总/recap"
+    enforcement: warn
+    trigger: response.target == 'chat'
+    check_method: heuristic
+    # heuristic — 末尾 200 char 含 '## 总结' / '## Summary' / 'In summary' / '综上所述'
+    # → warn 'redundant end recap'
+
+  - id: no-empty-continuation-question
+    description: 禁空洞续作询问 (要不要我帮你 X?/还需要别的吗?);保留工作流必要下一步指引
+    enforcement: warn
+    trigger: response.target == 'chat'
+    check_method: regex
+    # regex /要不要|还需要别的|希望对你有帮助/  + trigger fires → warn

package/workflows/disciplines/priority.yaml

@@ -20,8 +20,8 @@ priority_hierarchy:
 rules:
   - id: multi-capability-arbitration
     description: |
-      多 capability fires_when 同时 true → 按 priority_hierarchy order 选最高
-      e.g., gstack /office-hours + GSD /gsd-discuss-phase 同 fire → gstack 先跑
+      Multiple capabilities fires_when true at once → pick the highest by priority_hierarchy order
+      e.g., gstack /office-hours + GSD /gsd-discuss-phase fire together → gstack runs first
     enforcement: warn
     trigger: capabilities.fired_count > 1
     check_method: heuristic

package/workflows/disciplines/priority.zh-Hans.yaml

@@ -0,0 +1,28 @@
+# workflows/disciplines/priority.yaml
+# 优先级仲裁 — bundled discipline (7-tier hierarchy: gstack > gsd > superpowers > planning-with-files > karpathy > mattpocock > parallel)
+# Phase v3.0-3.3 W0.4 ship per D-09
+# Snapshot of CLAUDE.md as of v3.0 ship date (K7 mitigation, verbatim)
+
+schema_version: harnessed.discipline.v1
+discipline: priority
+enforcement_layer: workflow
+auto_enforce: true
+
+priority_hierarchy:
+  - gstack                       # 决策层 最高优先
+  - gsd                          # 整体 orchestration
+  - superpowers                  # 子任务执行质量
+  - planning-with-files          # 计划持久化
+  - karpathy                     # 编码行为
+  - mattpocock                   # 招式 by-demand
+  - parallel                     # subagent / Agent Teams / ralph-loop 执行机制
+
+rules:
+  - id: multi-capability-arbitration
+    description: |
+      多 capability fires_when 同时 true → 按 priority_hierarchy order 选最高
+      e.g., gstack /office-hours + GSD /gsd-discuss-phase 同 fire → gstack 先跑
+    enforcement: warn
+    trigger: capabilities.fired_count > 1
+    check_method: heuristic
+    # NEW src/discipline/enforcement/before-spawn.ts hook — sort fired capabilities by tier rank

package/workflows/discuss/auto/SKILL.zh-Hans.md

@@ -0,0 +1,75 @@
+---
+name: discuss
+description: |
+  Stage ① Discuss 主控编排器 — 3 个子工作流并行门控评估（chain-isolation 铁律）。
+  战略层 / Phase 层 / 子任务层独立判断 gate, 可能 3 个全跑 / 1-2 个 / 全跳 + 透明声明。
+  schema_version: harnessed.workflow.v3 with delegates_to (3 sub: strategic + phase + subtask, mode parallel)
+  + disciplines_applied (6 default) + tools_available (planning-with-files)。
+  Triggered by slash command `/discuss`
+  (bare per ADR 0030 namespace policy D-02 LOCK) after `harnessed setup`.
+trigger_phrases:
+  - "discuss"
+  - "讨论澄清"
+  - "新功能澄清"
+  - "stage 1 discuss"
+  - "三层澄清"
+---
+
+# discuss 主控编排器 (v3)
+
+## 概述
+
+4-stage cadence Stage ① 主控编排器，委托给 3 个独立子工作流
+（打包三层独立澄清：战略层 / phase 层 / 子任务层）：
+
+| 子工作流 | gate ref | mode | 触发时机 |
+| --- | -------- | ---- | ---------- |
+| `strategic` | `judgments.stage-routing.discuss-strategic-delegate.fires` | parallel | new_project / new_milestone / new_feature / is_major_release |
+| `phase` | `judgments.stage-routing.discuss-phase-delegate.fires` | parallel | open_decisions ≥ 2 / cross_phase_data_flow / scope_days > 1 |
+| `subtask` | `judgments.stage-routing.discuss-subtask-delegate.fires` | parallel | approaches ≥ 2 / core_algorithm / has_api_contract / error_cost=high |
+
+引擎运行时通过 `runMasterOrchestrator` 以
+T3.5.W0.1（sister sub-workflow `runWorkflow` SDK pattern，并行 fan-out 通过
+subagent 默认方式 — Path A `query()` 并行 OR Path B sub-shell 兜底，per
+T3.5.W2.1 dogfood cycle decision LOCK）生成 3 个子工作流阶段。
+
+## Chain-isolation 铁律（sister CLAUDE.md "Fallback 三条铁律"）
+
+- 拿不准 → 倾向跳过, 但在响应里**透明声明**: "这次跳过了 X, 因为 Y。如果你认为需要请明说"
+- 用户明示 → 覆盖判据 (用户说 "先 brainstorm" / "跑 office-hours" / "讨论一下" 时无条件激活)
+- **链式互不前置**: 跳过战略层 ≠ 必须跳过 phase 层; 每层独立判断 (防止 "上层没跑下层不敢跑" 的死板)
+
+## 能力引用
+
+Sister `workflows/capabilities.yaml`:
+- `planning-with-files` — Bucket 4 核心能力（impl: claude-code-plugin, cmd: /plan，discuss 汇入 findings.md）
+- sub `strategic` upstream → `gstack-office-hours` + `gstack-plan-ceo-review`
+- sub `phase` upstream → `gsd-discuss-phase`
+- sub `subtask` upstream → `superpowers-brainstorming` + `grill-with-docs` + `grill-me`
+
+## 调用方式
+
+- Slash command: `/discuss <text>`（bare per ADR 0030 namespace policy D-02 LOCK，`harnessed setup` 后可用）
+
+## 如何调用
+
+使用 Bash 工具运行：
+
+```bash
+echo "$ARGUMENTS" | harnessed run discuss --task-stdin
+```
+
+若 `$ARGUMENTS` 为空，运行 `harnessed run discuss`（不带 stdin pipe）。
+
+执行完成后，Bash 输出会在 stderr 打印 `Next:` 提示，建议下一个阶段。请根据对话上下文决定是否调用——该提示仅供参考，非强制指令。
+
+<!-- harnessed-generated:v3.4.4 -->
+
+## 参考资料
+
+- D-01 主控编排器委托模式
+- D-02 bare slash cmd 约定（ADR 0030 namespace policy LOCK）
+- D-04 Stage ① Discuss 三层独立判
+- workflows/judgments/stage-routing.yaml — discuss-{strategic,phase,subtask}-delegate triggers
+- workflows/discuss/{strategic,phase,subtask}/workflow.yaml — 3 个子工作流 Phase 3.4 已发布
+- workflows/judgments/fallback.yaml — 链式互不前置 chain-isolation 铁律

package/workflows/discuss/phase/SKILL.zh-Hans.md

@@ -0,0 +1,73 @@
+---
+name: discuss-phase
+description: |
+  Stage ①.b Phase 层 discuss 子工作流 — GSD /gsd-discuss-phase 灰色澄清（≥2 个开放
+  实现决策 / 跨 phase API contract 不清 / phase scope > 1 天 / 灰色地带）。
+  schema_version: harnessed.workflow.v3 with disciplines_applied (6 default) + tools_available
+  (gsd-discuss-phase + planning-with-files) + 2 phases（01-gsd-discuss + 02-persist findings.md
+  + knowledge.md）。Triggered by harnessed CLI `harnessed discuss-phase --phase <num>` or slash
+  command `/discuss-phase` after `harnessed setup`.
+trigger_phrases:
+  - "discuss phase"
+  - "phase 层澄清"
+  - "gsd discuss phase"
+  - "灰色地带澄清"
+  - "跑 discuss-phase"
+---
+
+# discuss-phase 工作流 (v3)
+
+## 概述
+
+2 阶段子工作流，将 CLAUDE.md "Stage ①.b Phase 层 — GSD /gsd-discuss-phase 灰色澄清"
+映射到 harnessed 运行时（Phase v3.0-3.4 W0.2 — D-04 Stage ① Discuss 三层 + Pattern A
+子工作流发布）。
+
+| 阶段 | id | upstream | model | 能力 / 调用 | 门控 / 产物 |
+| ----- | -- | -------- | ----- | -------------------- | ---------------- |
+| 1 | `01-gsd-discuss` | gsd | sonnet | `{{ capabilities.gsd-discuss-phase.cmd }}` | `gate: judgments.phase-gate.gsd-discuss-phase.fires` |
+| 2 | `02-persist` | planning-with-files | haiku | `{{ capabilities.planning-with-files.cmd }}` + `invokes: /plan` | `artifacts_expected: [findings.md, knowledge.md]` |
+
+## 能力引用
+
+Sister `workflows/capabilities.yaml` 条目：
+- `gsd-discuss-phase` — Bucket 2（impl: gsd, cmd: /gsd-discuss-phase）
+- `planning-with-files` — Bucket 4（impl: claude-code-plugin, cmd: /plan）
+
+## 门控引用
+
+Sister `workflows/judgments/phase-gate.yaml`:
+- `gsd-discuss-phase.fires` — `phase.open_decisions >= 2 or phase.has_cross_phase_data_flow == true or phase.scope_days > 1`
+
+## 调用方式
+
+- Slash command: `/discuss-phase <num>`（`harnessed setup` 后可用）
+
+## 路由规则
+
+跳过条件（sister CLAUDE.md "Phase 层 ❌ 跳过"）：
+- 单一明确子任务
+- 跟前 phase 同 module 同 pattern
+- < 1 天工作量
+- bug 修复且已有最小复现
+
+## 如何调用
+
+使用 Bash 工具运行：
+
+```bash
+echo "$ARGUMENTS" | harnessed run discuss-phase --task-stdin
+```
+
+若 `$ARGUMENTS` 为空，运行 `harnessed run discuss-phase`（不带 stdin pipe）。
+
+执行完成后，Bash 输出会在 stderr 打印 `Next:` 提示，建议下一个阶段。请根据对话上下文决定是否调用——该提示仅供参考，非强制指令。
+
+<!-- harnessed-generated:v3.4.4 -->
+
+## 参考资料
+
+- D-04 Stage ① Discuss 三层（战略层 / phase 层 / 子任务层）
+- workflows/capabilities.yaml — gsd-discuss-phase / planning-with-files
+- workflows/judgments/phase-gate.yaml — gsd-discuss-phase trigger
+- workflows/defaults.yaml — ralph_max_iterations.discuss-phase.* values (W2.2 backfill)

package/workflows/discuss/strategic/SKILL.zh-Hans.md

@@ -0,0 +1,78 @@
+---
+name: discuss-strategic
+description: |
+  Stage ①.a 战略层 discuss 子工作流 — gstack /office-hours + /plan-ceo-review 新功能 / 新
+  milestone / 新产品方向启动前强制治理关卡（打包 gstack governance gate，新功能前必跑）。
+  schema_version: harnessed.workflow.v3 with disciplines_applied (6 default) + tools_available
+  (gstack-office-hours + gstack-plan-ceo-review + planning-with-files) + 3 phases（gate
+  judgments.strategic-gate.* fires + planning-with-files /plan findings.md 持久化）。
+  Triggered by slash command
+  `/discuss-strategic` after `harnessed setup`.
+trigger_phrases:
+  - "discuss strategic"
+  - "战略层澄清"
+  - "office hours"
+  - "新功能启动"
+  - "跑 discuss-strategic"
+---
+
+# discuss-strategic 工作流 (v3)
+
+## 概述
+
+3 阶段子工作流，将 CLAUDE.md "Stage ①.a 战略层 — 新功能启动前强制 🔒"
+映射到 harnessed 运行时（Phase v3.0-3.4 W0.1 — D-04 Stage ① Discuss 三层 + D-12 gstack
+治理关卡 + Pattern A 子工作流发布）。
+
+| 阶段 | id | upstream | model | 能力 / 调用 | 门控 / 产物 |
+| ----- | -- | -------- | ----- | -------------------- | ---------------- |
+| 1 | `01-office-hours` | gstack | opus | `{{ capabilities.gstack-office-hours.cmd }}` | `gate: judgments.strategic-gate.office-hours.fires` |
+| 2 | `02-plan-ceo-review` | gstack | opus | `{{ capabilities.gstack-plan-ceo-review.cmd }}` | `gate: judgments.strategic-gate.plan-ceo-review.fires` |
+| 3 | `03-persist` | planning-with-files | haiku | `{{ capabilities.planning-with-files.cmd }}` + `invokes: /plan` | `artifacts_expected: [findings.md]` |
+
+## 能力引用
+
+Sister `workflows/capabilities.yaml` 条目：
+- `gstack-office-hours` — Bucket 3 治理关卡（impl: gstack, cmd: /office-hours）
+- `gstack-plan-ceo-review` — Bucket 3 治理关卡（impl: gstack, cmd: /plan-ceo-review）
+- `planning-with-files` — Bucket 4 核心能力（impl: claude-code-plugin, cmd: /plan）
+
+## 门控引用
+
+Sister `workflows/judgments/strategic-gate.yaml`:
+- `office-hours.fires` — `phase.type in ['new_project', 'new_milestone', 'new_feature']`
+- `plan-ceo-review.fires` — `phase.type == 'new_project' or phase.is_major_release == true`
+
+## 调用方式
+
+- Slash command: `/discuss-strategic <text>`（`harnessed setup` 后可用）
+
+## 路由规则
+
+跳过条件（sister CLAUDE.md "战略层 ❌ 跳过"）：
+- bug 修复 / 技术债 / 实现细节
+- scope 已在历史 `.planning/` 或 design doc 定义
+- continuing 已有 phase 的执行
+- 用户已给明确 ticket / spec
+
+## 如何调用
+
+使用 Bash 工具运行：
+
+```bash
+echo "$ARGUMENTS" | harnessed run discuss-strategic --task-stdin
+```
+
+若 `$ARGUMENTS` 为空，运行 `harnessed run discuss-strategic`（不带 stdin pipe）。
+
+执行完成后，Bash 输出会在 stderr 打印 `Next:` 提示，建议下一个阶段。请根据对话上下文决定是否调用——该提示仅供参考，非强制指令。
+
+<!-- harnessed-generated:v3.4.4 -->
+
+## 参考资料
+
+- D-04 Stage ① Discuss 三层（战略层 / phase 层 / 子任务层）
+- D-12 gstack 治理关卡强制
+- workflows/capabilities.yaml — gstack-office-hours / gstack-plan-ceo-review / planning-with-files
+- workflows/judgments/strategic-gate.yaml — office-hours / plan-ceo-review triggers
+- workflows/defaults.yaml — ralph_max_iterations.discuss-strategic.* values (W2.2 backfill)

package/workflows/discuss/subtask/SKILL.zh-Hans.md

@@ -0,0 +1,79 @@
+---
+name: discuss-subtask
+description: |
+  Stage ①.c 子任务层 discuss 子工作流 — superpowers brainstorming（≥2 种方案 / 核心算法 /
+  API contract / 错误成本高）。schema_version: harnessed.workflow.v3 with disciplines_applied (6
+  default) + tools_available (superpowers-brainstorming + grill-with-docs + grill-me) + 1 phase
+  （01-brainstorm，含 spec_ambiguous + no_docs 条件调用 invokes_tools）。Triggered by
+  harnessed CLI `harnessed discuss-subtask --task <text>` or slash command `/discuss-subtask`
+  after `harnessed setup`.
+trigger_phrases:
+  - "discuss subtask"
+  - "子任务澄清"
+  - "brainstorm"
+  - "多 approach 探索"
+  - "跑 discuss-subtask"
+---
+
+# discuss-subtask 工作流 (v3)
+
+## 概述
+
+1 阶段子工作流，将 CLAUDE.md "Stage ①.c 子任务层 — superpowers brainstorming"
+映射到 harnessed 运行时（Phase v3.0-3.4 W0.3 — D-04 Stage ① Discuss 三层 + D-05
+phase.invokes_tools 条件触发 + Pattern A 子工作流发布）。
+
+| 阶段 | id | upstream | model | 能力 / invokes_tools | 门控 |
+| ----- | -- | -------- | ----- | -------------------------- | ---- |
+| 1 | `01-brainstorm` | superpowers | opus | `{{ capabilities.superpowers-brainstorming.cmd }}` + conditional `invokes_tools[grill-with-docs, grill-me]` | `gate: judgments.subtask-gate.brainstorming.fires` |
+
+## 能力引用
+
+Sister `workflows/capabilities.yaml` 条目：
+- `superpowers-brainstorming` — Bucket 4（impl: superpowers, cmd: superpowers:brainstorming）
+- `grill-with-docs` — Bucket 1 mattpocock（impl: mattpocock-skills, cmd: /grill-with-docs）
+- `grill-me` — Bucket 1 mattpocock（impl: mattpocock-skills, cmd: /grill-me）
+
+## 门控引用
+
+Sister `workflows/judgments/subtask-gate.yaml`:
+- `brainstorming.fires` — `subtask.approaches >= 2 or subtask.core_algorithm == true or subtask.has_api_contract == true or subtask.error_cost == 'high'`
+
+## 条件工具调用（D-05 invokes_tools）
+
+- `phase.spec_ambiguous == true` → 触发 `grill-with-docs`
+- `phase.spec_ambiguous == true AND phase.no_docs == true` → 触发 `grill-me`
+
+## 调用方式
+
+- Slash command: `/discuss-subtask <text>`（`harnessed setup` 后可用）
+
+## 路由规则
+
+跳过条件（sister CLAUDE.md "子任务层 ❌ 跳过"）：
+- 常规 CRUD / 跟现有模式一致
+- 单一明显实现（< 20 行 / 一个文件）
+- 标准库直接调用
+- bug 修复且已知根因
+
+## 如何调用
+
+使用 Bash 工具运行：
+
+```bash
+echo "$ARGUMENTS" | harnessed run discuss-subtask --task-stdin
+```
+
+若 `$ARGUMENTS` 为空，运行 `harnessed run discuss-subtask`（不带 stdin pipe）。
+
+执行完成后，Bash 输出会在 stderr 打印 `Next:` 提示，建议下一个阶段。请根据对话上下文决定是否调用——该提示仅供参考，非强制指令。
+
+<!-- harnessed-generated:v3.4.4 -->
+
+## 参考资料
+
+- D-04 Stage ① Discuss 三层（战略层 / phase 层 / 子任务层）
+- D-05 phase.invokes_tools 条件触发（NEW v3）
+- workflows/capabilities.yaml — superpowers-brainstorming / grill-with-docs / grill-me
+- workflows/judgments/subtask-gate.yaml — brainstorming trigger
+- workflows/defaults.yaml — ralph_max_iterations.discuss-subtask.* values (W2.2 backfill)

package/workflows/plan/architecture/SKILL.zh-Hans.md

@@ -0,0 +1,74 @@
+---
+name: plan-architecture
+description: |
+  Stage ②.a 架构层 plan 子工作流 — gstack /plan-eng-review（复杂架构强制治理关卡；
+  打包的 plan-stage 节奏：复杂架构阶段前必须执行）。schema_version:
+  harnessed.workflow.v3 with disciplines_applied (6 default) + tools_available
+  [plan-eng-review] + 1 phase (01-plan-eng-review) + gate literal expr
+  `phase.is_complex_architecture == true`。Triggered by harnessed CLI `harnessed
+  plan-architecture --module <name>` or slash command `/plan-architecture` after
+  `harnessed setup`.
+trigger_phrases:
+  - "plan architecture"
+  - "架构审查"
+  - "plan eng review"
+  - "复杂架构"
+  - "跑 plan-architecture"
+---
+
+# plan-architecture 工作流 (v3)
+
+## 概述
+
+1 阶段子工作流，将 CLAUDE.md "⚠️ 复杂架构前 — plan-eng-review" 映射到 harnessed
+运行时（Phase v3.0-3.4 W0.4 — D-04 Stage ② Plan 二层 + D-12 gstack 治理关卡 + Pattern A
+sub-workflow ship）。
+
+| phase | id | upstream | model | capability | gate |
+| ----- | -- | -------- | ----- | ---------- | ---- |
+| 1 | `01-plan-eng-review` | gstack | opus | `{{ capabilities.plan-eng-review.cmd }}` | `gate: phase.is_complex_architecture == true` (literal expr) |
+
+## 能力引用
+
+Sister `workflows/capabilities.yaml` 条目：
+- `plan-eng-review` — Bucket 7 gstack 33 optional (impl: gstack, cmd: /plan-eng-review)
+  - Pattern A E.2 LOCK：无 `gstack-` 前缀，遵循 CLAUDE.md gstack skill 路由约定
+
+## 门控引用
+
+内联 literal expr `phase.is_complex_architecture == true` — sister capability
+`fires_when` 子句原文。W2.2 backfill 可加 `judgments/phase-gate.yaml`
+trigger `is-complex-architecture` 重构 ref。
+
+## 调用方式
+
+- Slash command：`/plan-architecture <name>`（`harnessed setup` 后生效）
+
+## 路由规则
+
+触发条件（sister CLAUDE.md "⚠️ 复杂架构前"）：
+- 涉及多模块跨界设计（≥3 模块协同）
+- 新引入核心抽象（新 framework / 新数据模型 / 新协议）
+- 性能 / scaling 关键路径
+- 引入显著技术债 / migration 风险
+
+## 如何调用
+
+使用 Bash 工具运行：
+
+```bash
+echo "$ARGUMENTS" | harnessed run plan-architecture --task-stdin
+```
+
+若 `$ARGUMENTS` 为空，运行 `harnessed run plan-architecture`（不带 stdin pipe）。
+
+完成后，Bash 输出会在 stderr 打印 `Next:` 提示，建议下一阶段。是否调用由对话上下文决定——该提示仅供参考，非强制指令。
+
+<!-- harnessed-generated:v3.4.4 -->
+
+## 参考文档
+
+- D-04 Stage ② Plan 二层（架构 / 计划）
+- D-12 gstack 治理关卡（复杂架构强制）
+- workflows/capabilities.yaml — plan-eng-review (Bucket 7)
+- workflows/defaults.yaml — ralph_max_iterations.plan-architecture.* values (W2.2 backfill)