ccgx-workflow 2.1.2 → 2.3.0

package/templates/commands/spec-plan.md

@@ -24,33 +24,129 @@ description: '多模型分析 → 消除歧义 → 零决策可执行计划'
    - Run `openspec status --change "<change_id>" --json` to review current state.
 
 2. **Multi-Model Implementation Analysis (PARALLEL)**
-   - **CRITICAL**: You MUST launch BOTH {{BACKEND_PRIMARY}} AND {{FRONTEND_PRIMARY}} in a SINGLE message with TWO Bash tool calls.
-   - **DO NOT** call one model first and wait. Launch BOTH simultaneously with `run_in_background: true`.
+
+   **调用通道路由（CCG codeagent 退役，v2.2.0+）**
+
+   双模型并行通道从 `Bash(codeagent-wrapper)` **默认切换**为 plugin spawn：
+
+   1. **优先 plugin spawn**（默认）：装了 `codex@openai-codex` + gemini plugin（推荐 `gemini@gemini-ccgx` fork；或上游 `gemini@google-gemini` 配 repatch 脚本）→ 用 `Agent(subagent_type="codex:codex-rescue")` + `Agent(subagent_type="gemini:gemini-rescue")` 并行，主线接 ≤200 token 摘要。
+   2. **降级 codeagent-wrapper**（BC fallback）：plugin 未装 → fallback Bash 调用，行为与 plugin 路径等价。
+
+   **判定**：preflight `Bash` 跑 `ls ~/.claude/plugins/` 看有无 `codex@*` / `gemini@*` 子目录。
+
+   ⚠️ spec-plan 命令在主线 context 内，**允许** `Agent(...)`——与 subagent "禁止嵌套 spawn" 约束不冲突。
+
+   - **CRITICAL**: You MUST launch BOTH {{BACKEND_PRIMARY}} AND {{FRONTEND_PRIMARY}} in a SINGLE message with TWO parallel tool calls.
+   - **DO NOT** call one model first and wait. Launch BOTH simultaneously.
    - **工作目录**：`{{WORKDIR}}` **必须通过 Bash 执行 `pwd`（Unix）或 `cd`（Windows CMD）获取当前工作目录的绝对路径**，禁止从 `$HOME` 或环境变量推断。如果用户通过 `/add-dir` 添加了多个工作区，先确定任务相关的工作区。
 
-   **Step 2.1**: In ONE message, make TWO parallel Bash calls:
+   **Step 2.1**: In ONE message, spawn TWO models in parallel.
+
+   **⚠ 预备动作（spawn 前必须执行）**：`codex:codex-rescue` / `gemini:gemini-rescue` 是 thin forwarder，不会主动 Read 路径文件。主线必须在 spawn 前先 Read 两个角色提示词文件，把**内容**直接拼入下方 Agent prompt 的 `<role>` 块。
 
-   **FIRST Bash call ({{BACKEND_PRIMARY}})**:
+   - backend role: `Read("~/.claude/.ccg/prompts/{{BACKEND_PRIMARY}}/analyzer.md")` → `${backendRole}`
+   - frontend role: `Read("~/.claude/.ccg/prompts/{{FRONTEND_PRIMARY}}/analyzer.md")` → `${frontendRole}`
+
+   Prompt 结构按 `gpt-5-4-prompting` skill 推荐：`<role>` + `<task>` + `<grounding_rules>` + `<structured_output_contract>`。
+
+   **通道 A — plugin spawn（默认）**：
+
+   **FIRST Agent call ({{BACKEND_PRIMARY}})**:
    ```
-   Bash({
-     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{BACKEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nAnalyze change <change_id> from backend perspective:\n- Implementation approach\n- Technical risks\n- Alternative architectures\n- Edge cases and failure modes\nOUTPUT: JSON with analysis\nEOF",
-     run_in_background: true,
-     timeout: 300000,
-     description: "{{BACKEND_PRIMARY}}: backend analysis"
+   Agent({
+     subagent_type: "codex:codex-rescue",
+     description: "spec-plan: backend analysis",
+     prompt: `<role>
+${backendRole}
+</role>
+
+<workdir>{{WORKDIR}}</workdir>
+
+<task>
+Analyze change <change_id> from backend perspective:
+- Implementation approach
+- Technical risks
+- Alternative architectures
+- Edge cases and failure modes
+</task>
+
+<grounding_rules>
+- Cite file:line for claims about existing code
+- Mark hypotheses; don't state guesses as facts
+- For risks, distinguish "observed" (specific code path) vs "potential" (theoretical)
+</grounding_rules>
+
+<structured_output_contract>
+Return JSON ONLY (no preamble):
+{
+  "approach": "selected implementation strategy with rationale",
+  "alternatives_rejected": [{"alt": "...", "reason": "..."}],
+  "technical_risks": [{"risk": "...", "severity": "high|medium|low", "mitigation": "..."}],
+  "edge_cases": ["..."],
+  "failure_modes": ["..."]
+}
+Return ≤200 token structured summary.
+</structured_output_contract>`
+   })
+   ```
+
+   **SECOND Agent call ({{FRONTEND_PRIMARY}}) - IN THE SAME MESSAGE**:
+   ```
+   Agent({
+     subagent_type: "gemini:gemini-rescue",
+     description: "spec-plan: frontend analysis",
+     prompt: `<role>
+${frontendRole}
+</role>
+
+<workdir>{{WORKDIR}}</workdir>
+
+<task>
+Analyze change <change_id> from frontend/integration perspective:
+- Maintainability assessment
+- Scalability considerations
+- Integration conflicts
+</task>
+
+<grounding_rules>
+- Cite file:line for claims about existing code
+- Mark hypotheses; don't state guesses as facts
+</grounding_rules>
+
+<structured_output_contract>
+Return JSON ONLY (no preamble):
+{
+  "maintainability": {"score": "good|moderate|concerning", "factors": ["..."]},
+  "scalability": [{"dimension": "...", "consideration": "..."}],
+  "integration_conflicts": [{"target": "...", "conflict": "...", "resolution": "..."}]
+}
+Return ≤200 token structured summary.
+</structured_output_contract>`
    })
    ```
 
-   **SECOND Bash call ({{FRONTEND_PRIMARY}}) - IN THE SAME MESSAGE**:
+   **通道 B — codeagent-wrapper fallback**（plugin 未装时降级，并行用 `run_in_background: true`）：
+
    ```
    Bash({
-     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{FRONTEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nAnalyze change <change_id> from frontend/integration perspective:\n- Maintainability assessment\n- Scalability considerations\n- Integration conflicts\nOUTPUT: JSON with analysis\nEOF",
+     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{BACKEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nROLE_FILE: ~/.claude/.ccg/prompts/{{BACKEND_PRIMARY}}/analyzer.md\nAnalyze change <change_id> from backend perspective:\n- Implementation approach\n- Technical risks\n- Alternative architectures\n- Edge cases and failure modes\nOUTPUT: JSON with analysis\nEOF",
+     run_in_background: true,
+     timeout: 300000,
+     description: "{{BACKEND_PRIMARY}}: backend analysis (BC)"
+   })
+   Bash({
+     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{FRONTEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nROLE_FILE: ~/.claude/.ccg/prompts/{{FRONTEND_PRIMARY}}/analyzer.md\nAnalyze change <change_id> from frontend/integration perspective:\n- Maintainability assessment\n- Scalability considerations\n- Integration conflicts\nOUTPUT: JSON with analysis\nEOF",
      run_in_background: true,
      timeout: 300000,
-     description: "{{FRONTEND_PRIMARY}}: frontend analysis"
+     description: "{{FRONTEND_PRIMARY}}: frontend analysis (BC)"
    })
    ```
 
-   **Step 2.2 (事件驱动)**: spawn 两个 Bash bg 后说明 task-id 然后 **turn end**。引擎在每个 task 完成时自动发 `<task-notification>`，主线在通知触发的新 turn 处理结果。**不调 TaskOutput**。两个 task 都收到通知后才进 step 2.3。
+   > ⚠️ 通道 B `codeagent-wrapper` 已标 **deprecated**，仅为不能升级 plugin 的环境保留。
+
+   **Step 2.2 (事件驱动)**：
+   - **通道 A（plugin）**：两个 Agent 同 message 内 spawn，主线接 ≤200 token 摘要。Agent 完成后自动返回结果。
+   - **通道 B（BC wrapper）**：spawn 两个 Bash bg 后说明 task-id 然后 **turn end**。引擎在每个 task 完成时自动发 `<task-notification>`，主线在通知触发的新 turn 处理结果。**不调 TaskOutput**。两个 task 都收到通知后才进 step 2.3。
 
    ⛔ **禁止**：调 `TaskOutput({block: true, timeout: 600000})` (旧 freeze poll 模式) / Kill task。
    ⚠️ **失败处理**：notification status=failed / exit ≠ 0 / parse 失败 → v1.7.87 标准 2-retry / 5s / 3-attempts；3 次全失败才降级单模型。

package/templates/commands/spec-research.md

@@ -49,8 +49,20 @@ description: '需求 → 约束集（并行探索 + OPSX 提案）'
    - Each boundary should be self-contained: no cross-communication needed.
 
 4. **Parallel Multi-Model Exploration**
-   - **CRITICAL**: You MUST launch BOTH {{BACKEND_PRIMARY}} AND {{FRONTEND_PRIMARY}} in a SINGLE message with TWO Bash tool calls.
-   - **DO NOT** call one model first and wait. Launch BOTH simultaneously with `run_in_background: true`.
+
+   **调用通道路由（CCG codeagent 退役，v2.2.0+）**
+
+   双模型并行通道从 `Bash(codeagent-wrapper)` **默认切换**为 plugin spawn：
+
+   1. **优先 plugin spawn**（默认）：装了 `codex@openai-codex` + gemini plugin（推荐 `gemini@gemini-ccgx` fork，含 P-1..P-21 + W1/W2/I1 patch；或上游 `gemini@google-gemini` 配 repatch 脚本）→ 用 `Agent(subagent_type="codex:codex-rescue")` + `Agent(subagent_type="gemini:gemini-rescue")` 并行，主线接 ≤200 token 摘要。
+   2. **降级 codeagent-wrapper**（BC fallback）：plugin 未装 → fallback Bash 调用，行为与 plugin 路径等价。
+
+   **判定**：preflight `Bash` 跑 `ls ~/.claude/plugins/` 看有无 `codex@*` / `gemini@*` 子目录。
+
+   ⚠️ spec-research 命令在主线 context 内，**允许** `Agent(...)`——与 subagent "禁止嵌套 spawn" 约束不冲突。
+
+   - **CRITICAL**: You MUST launch BOTH {{BACKEND_PRIMARY}} AND {{FRONTEND_PRIMARY}} in a SINGLE message with TWO parallel tool calls.
+   - **DO NOT** call one model first and wait. Launch BOTH simultaneously.
    - **工作目录**：`{{WORKDIR}}` **必须通过 Bash 执行 `pwd`（Unix）或 `cd`（Windows CMD）获取当前工作目录的绝对路径**，禁止从 `$HOME` 或环境变量推断。如果用户通过 `/add-dir` 添加了多个工作区，先确定任务相关的工作区。
 
    **Output Template** (instruct both models to use this format):
@@ -67,29 +79,123 @@ description: '需求 → 约束集（并行探索 + OPSX 提案）'
    }
    ```
 
-   **Step 4.1**: In ONE message, make TWO parallel Bash calls:
+   **Step 4.1**: In ONE message, spawn TWO models in parallel.
+
+   **⚠ 预备动作（spawn 前必须执行）**：`codex:codex-rescue` / `gemini:gemini-rescue` 是 thin forwarder（一行 `Bash node companion.mjs task <prompt>` 转发），**不会**主动 Read 路径文件。主线必须在 spawn 前先 Read 两个角色提示词文件，把**内容**直接拼入下方 Agent prompt 的 `<role>` 块——不是写路径。
+
+   - backend role: `Read("~/.claude/.ccg/prompts/{{BACKEND_PRIMARY}}/analyzer.md")` → `${backendRole}`
+   - frontend role: `Read("~/.claude/.ccg/prompts/{{FRONTEND_PRIMARY}}/analyzer.md")` → `${frontendRole}`
 
-   **FIRST Bash call ({{BACKEND_PRIMARY}} — backend boundaries)**:
+   Prompt 结构按 `gpt-5-4-prompting` skill 推荐（XML 块、紧凑、操作性）：`<role>` + `<task>` + `<grounding_rules>` + `<structured_output_contract>`。
+
+   **通道 A — plugin spawn（默认）**：
+
+   **FIRST Agent call ({{BACKEND_PRIMARY}} — backend boundaries)**:
    ```
-   Bash({
-     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{BACKEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nExplore backend context boundaries for <change description>:\n- Existing structures and patterns\n- Conventions in use\n- Hard constraints limiting solution space\n- Dependencies and risks\nOUTPUT: JSON using the output template above\nEOF",
-     run_in_background: true,
-     timeout: 300000,
-     description: "{{BACKEND_PRIMARY}}: backend boundary exploration"
+   Agent({
+     subagent_type: "codex:codex-rescue",
+     description: "spec-research: backend boundary exploration",
+     prompt: `<role>
+${backendRole}
+</role>
+
+<workdir>{{WORKDIR}}</workdir>
+
+<task>
+Explore backend context boundaries for <change description>:
+- Existing structures and patterns
+- Conventions in use
+- Hard constraints limiting solution space
+- Dependencies and risks
+</task>
+
+<grounding_rules>
+- Cite file:line for every claim about existing code
+- Mark hypotheses explicitly; don't state guesses as facts
+- If a question can't be answered from the repo, list it in open_questions
+</grounding_rules>
+
+<structured_output_contract>
+Return JSON ONLY (no preamble, no commentary):
+{
+  "module_name": "context boundary explored",
+  "existing_structures": ["key patterns found"],
+  "existing_conventions": ["standards in use"],
+  "constraints_discovered": ["hard constraints limiting solution space"],
+  "open_questions": ["ambiguities requiring user input"],
+  "dependencies": ["cross-module dependencies"],
+  "risks": ["potential blockers"],
+  "success_criteria_hints": ["observable success behaviors"]
+}
+Return ≤200 token structured summary.
+</structured_output_contract>`
    })
    ```
 
-   **SECOND Bash call ({{FRONTEND_PRIMARY}} — frontend boundaries) - IN THE SAME MESSAGE**:
+   **SECOND Agent call ({{FRONTEND_PRIMARY}} — frontend boundaries) - IN THE SAME MESSAGE**:
+   ```
+   Agent({
+     subagent_type: "gemini:gemini-rescue",
+     description: "spec-research: frontend boundary exploration",
+     prompt: `<role>
+${frontendRole}
+</role>
+
+<workdir>{{WORKDIR}}</workdir>
+
+<task>
+Explore frontend context boundaries for <change description>:
+- Existing structures and patterns
+- Conventions in use
+- Hard constraints limiting solution space
+- Dependencies and risks
+</task>
+
+<grounding_rules>
+- Cite file:line for every claim about existing code
+- Mark hypotheses explicitly; don't state guesses as facts
+- If a question can't be answered from the repo, list it in open_questions
+</grounding_rules>
+
+<structured_output_contract>
+Return JSON ONLY (no preamble, no commentary):
+{
+  "module_name": "context boundary explored",
+  "existing_structures": ["key patterns found"],
+  "existing_conventions": ["standards in use"],
+  "constraints_discovered": ["hard constraints limiting solution space"],
+  "open_questions": ["ambiguities requiring user input"],
+  "dependencies": ["cross-module dependencies"],
+  "risks": ["potential blockers"],
+  "success_criteria_hints": ["observable success behaviors"]
+}
+Return ≤200 token structured summary.
+</structured_output_contract>`
+   })
+   ```
+
+   **通道 B — codeagent-wrapper fallback**（plugin 未装时降级，并行用 `run_in_background: true`）：
+
    ```
    Bash({
-     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{FRONTEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nExplore frontend context boundaries for <change description>:\n- Existing structures and patterns\n- Conventions in use\n- Hard constraints limiting solution space\n- Dependencies and risks\nOUTPUT: JSON using the output template above\nEOF",
+     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{BACKEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nROLE_FILE: ~/.claude/.ccg/prompts/{{BACKEND_PRIMARY}}/analyzer.md\nExplore backend context boundaries for <change description>:\n- Existing structures and patterns\n- Conventions in use\n- Hard constraints limiting solution space\n- Dependencies and risks\nOUTPUT: JSON using the output template above\nEOF",
      run_in_background: true,
      timeout: 300000,
-     description: "{{FRONTEND_PRIMARY}}: frontend boundary exploration"
+     description: "{{BACKEND_PRIMARY}}: backend boundary exploration (BC)"
+   })
+   Bash({
+     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{FRONTEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nROLE_FILE: ~/.claude/.ccg/prompts/{{FRONTEND_PRIMARY}}/analyzer.md\nExplore frontend context boundaries for <change description>:\n- Existing structures and patterns\n- Conventions in use\n- Hard constraints limiting solution space\n- Dependencies and risks\nOUTPUT: JSON using the output template above\nEOF",
+     run_in_background: true,
+     timeout: 300000,
+     description: "{{FRONTEND_PRIMARY}}: frontend boundary exploration (BC)"
    })
    ```
 
-   **Step 4.2 (事件驱动)**: spawn 两个 Bash bg 后说明 task-id 然后 **turn end**。引擎在每个 task 完成时自动发 `<task-notification>`，主线在通知触发的新 turn 处理结果。**不调 TaskOutput**。两个 task 都收到通知后才进 step 4.3。
+   > ⚠️ 通道 B `codeagent-wrapper` 已标 **deprecated**，仅为不能升级 plugin 的环境保留。
+
+   **Step 4.2 (事件驱动)**：
+   - **通道 A（plugin）**：两个 Agent 同 message 内 spawn，主线接 ≤200 token 摘要。Agent 完成后自动返回结果到调用上下文，主线直接处理。
+   - **通道 B（BC wrapper）**：spawn 两个 Bash bg 后说明 task-id 然后 **turn end**。引擎在每个 task 完成时自动发 `<task-notification>`，主线在通知触发的新 turn 处理结果。**不调 TaskOutput**。两个 task 都收到通知后才进 step 4.3。
 
    ⛔ **禁止**：调 `TaskOutput({block: true, timeout: 600000})` (旧 freeze poll 模式) / Kill task。
    ⚠️ **失败处理**：notification status=failed / exit ≠ 0 / parse 失败 → v1.7.87 标准 2-retry / 5s / 3-attempts；3 次全失败才降级单模型。

package/templates/commands/spec-review.md

@@ -25,33 +25,166 @@ description: '双模型交叉审查（独立工具，随时可用）'
    - Load relevant spec constraints and PBT properties from `openspec/changes/<id>/specs/`.
 
 3. **Multi-Model Review (PARALLEL)**
-   - **CRITICAL**: You MUST launch BOTH {{BACKEND_PRIMARY}} AND {{FRONTEND_PRIMARY}} in a SINGLE message with TWO Bash tool calls.
-   - **DO NOT** call one model first and wait. Launch BOTH simultaneously with `run_in_background: true`.
+
+   **调用通道路由（CCG codeagent 退役，v2.2.0+）**
+
+   双模型并行通道从 `Bash(codeagent-wrapper)` **默认切换**为 plugin spawn：
+
+   1. **优先 plugin spawn**（默认）：装了 `codex@openai-codex` + gemini plugin（推荐 `gemini@gemini-ccgx` fork；或上游 `gemini@google-gemini` 配 repatch 脚本）→ 用 `Agent(subagent_type="codex:codex-rescue")` + `Agent(subagent_type="gemini:gemini-rescue")` 并行，主线接 ≤200 token 摘要。
+   2. **降级 codeagent-wrapper**（BC fallback）：plugin 未装 → fallback Bash 调用，行为与 plugin 路径等价。
+
+   **判定**：preflight `Bash` 跑 `ls ~/.claude/plugins/` 看有无 `codex@*` / `gemini@*` 子目录。
+
+   ⚠️ spec-review 命令在主线 context 内，**允许** `Agent(...)`——与 subagent "禁止嵌套 spawn" 约束不冲突。
+
+   - **CRITICAL**: You MUST launch BOTH {{BACKEND_PRIMARY}} AND {{FRONTEND_PRIMARY}} in a SINGLE message with TWO parallel tool calls.
+   - **DO NOT** call one model first and wait. Launch BOTH simultaneously.
    - **工作目录**：`{{WORKDIR}}` **必须通过 Bash 执行 `pwd`（Unix）或 `cd`（Windows CMD）获取当前工作目录的绝对路径**，禁止从 `$HOME` 或环境变量推断。如果用户通过 `/add-dir` 添加了多个工作区，先确定任务相关的工作区。
 
-   **Step 3.1**: In ONE message, make TWO parallel Bash calls:
+   **Step 3.1**: In ONE message, spawn TWO models in parallel.
 
-   **FIRST Bash call ({{BACKEND_PRIMARY}})**:
+   **⚠ 预备动作（spawn 前必须执行）**：`codex:codex-rescue` / `gemini:gemini-rescue` 是 thin forwarder，不会主动 Read 路径文件。主线必须在 spawn 前先 Read 两个角色提示词文件，把**内容**直接拼入下方 Agent prompt 的 `<role>` 块。
+
+   - backend role: `Read("~/.claude/.ccg/prompts/{{BACKEND_PRIMARY}}/reviewer.md")` → `${backendRole}`
+   - frontend role: `Read("~/.claude/.ccg/prompts/{{FRONTEND_PRIMARY}}/reviewer.md")` → `${frontendRole}`
+
+   Prompt 结构按 `gpt-5-4-prompting` skill 推荐（review 类任务额外加 `<grounding_rules>` + `<dig_deeper_nudge>`）。
+
+   **通道 A — plugin spawn（默认）**：
+
+   **FIRST Agent call ({{BACKEND_PRIMARY}})**:
    ```
-   Bash({
-     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{BACKEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nReview proposal <proposal_id> implementation:\n\n## {{BACKEND_PRIMARY}} Review Dimensions\n1. **Spec Compliance**: Verify ALL constraints from spec are satisfied\n2. **PBT Properties**: Check invariants, idempotency, bounds are correctly implemented\n3. **Logic Correctness**: Edge cases, error handling, algorithm correctness\n4. **Backend Security**: Injection vulnerabilities, auth checks, input validation\n5. **Regression Risk**: Interface compatibility, type safety, breaking changes\n\n## Output Format (JSON)\n{\n  \"findings\": [\n    {\n      \"severity\": \"Critical|Warning|Info\",\n      \"dimension\": \"spec_compliance|pbt|logic|security|regression\",\n      \"file\": \"path/to/file.ts\",\n      \"line\": 42,\n      \"description\": \"What is wrong\",\n      \"constraint_violated\": \"Constraint ID from spec (if applicable)\",\n      \"fix_suggestion\": \"How to fix\"\n    }\n  ],\n  \"passed_checks\": [\"List of verified constraints/properties\"],\n  \"summary\": \"Overall assessment\"\n}\nEOF",
-     run_in_background: true,
-     timeout: 300000,
-     description: "{{BACKEND_PRIMARY}}: backend/logic review"
+   Agent({
+     subagent_type: "codex:codex-rescue",
+     description: "spec-review: backend/logic review",
+     prompt: `<role>
+${backendRole}
+</role>
+
+<workdir>{{WORKDIR}}</workdir>
+
+<task>
+Review proposal <proposal_id> implementation along backend dimensions:
+1. Spec Compliance — Verify ALL constraints from spec are satisfied
+2. PBT Properties — Check invariants, idempotency, bounds correctly implemented
+3. Logic Correctness — Edge cases, error handling, algorithm correctness
+4. Backend Security — Injection vulnerabilities, auth checks, input validation
+5. Regression Risk — Interface compatibility, type safety, breaking changes
+
+Read-only review. Do NOT modify any file.
+</task>
+
+<grounding_rules>
+- Every finding must cite file:line
+- If a constraint was satisfied, list it in passed_checks (don't omit positives)
+- "I couldn't verify X" is acceptable; fabricating X is not
+- Distinguish bug (observable failure path) from concern (style/maintainability preference)
+</grounding_rules>
+
+<dig_deeper_nudge>
+- For each Critical finding, check whether the same root cause appears elsewhere in the diff
+- Prefer one well-cited Critical over five vague Warnings
+</dig_deeper_nudge>
+
+<structured_output_contract>
+Return JSON ONLY (no preamble):
+{
+  "findings": [
+    {
+      "severity": "Critical|Warning|Info",
+      "dimension": "spec_compliance|pbt|logic|security|regression",
+      "file": "path/to/file.ts",
+      "line": 42,
+      "description": "What is wrong",
+      "constraint_violated": "Constraint ID from spec (if applicable)",
+      "fix_suggestion": "How to fix"
+    }
+  ],
+  "passed_checks": ["List of verified constraints/properties"],
+  "summary": "Overall assessment"
+}
+Return ≤200 token structured summary.
+</structured_output_contract>`
+   })
+   ```
+
+   **SECOND Agent call ({{FRONTEND_PRIMARY}}) - IN THE SAME MESSAGE**:
+   ```
+   Agent({
+     subagent_type: "gemini:gemini-rescue",
+     description: "spec-review: patterns/integration review",
+     prompt: `<role>
+${frontendRole}
+</role>
+
+<workdir>{{WORKDIR}}</workdir>
+
+<task>
+Review proposal <proposal_id> implementation along frontend/integration dimensions:
+1. Pattern Consistency — Naming conventions, code style, project patterns
+2. Maintainability — Readability, complexity, documentation adequacy
+3. Integration Risk — Dependency changes, cross-module impacts
+4. Frontend Security — XSS, CSRF, sensitive data exposure
+5. Spec Alignment — Implementation matches spec intent (not just letter)
+
+Read-only review. Do NOT modify any file.
+</task>
+
+<grounding_rules>
+- Every finding must cite file:line
+- If an aspect was clean, list it in passed_checks
+- "I couldn't verify X" is acceptable; fabricating X is not
+</grounding_rules>
+
+<dig_deeper_nudge>
+- For each Critical finding, check whether the same pattern issue appears in adjacent files
+- Prefer one well-cited Critical over five vague Warnings
+</dig_deeper_nudge>
+
+<structured_output_contract>
+Return JSON ONLY (no preamble):
+{
+  "findings": [
+    {
+      "severity": "Critical|Warning|Info",
+      "dimension": "patterns|maintainability|integration|security|alignment",
+      "file": "path/to/file.ts",
+      "line": 42,
+      "description": "What is wrong",
+      "spec_reference": "Spec section (if applicable)",
+      "fix_suggestion": "How to fix"
+    }
+  ],
+  "passed_checks": ["List of verified aspects"],
+  "summary": "Overall assessment"
+}
+Return ≤200 token structured summary.
+</structured_output_contract>`
    })
    ```
 
-   **SECOND Bash call ({{FRONTEND_PRIMARY}}) - IN THE SAME MESSAGE**:
+   **通道 B — codeagent-wrapper fallback**（plugin 未装时降级，并行用 `run_in_background: true`）：
+
    ```
    Bash({
-     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{FRONTEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nReview proposal <proposal_id> implementation:\n\n## {{FRONTEND_PRIMARY}} Review Dimensions\n1. **Pattern Consistency**: Naming conventions, code style, project patterns\n2. **Maintainability**: Readability, complexity, documentation adequacy\n3. **Integration Risk**: Dependency changes, cross-module impacts\n4. **Frontend Security**: XSS, CSRF, sensitive data exposure\n5. **Spec Alignment**: Implementation matches spec intent (not just letter)\n\n## Output Format (JSON)\n{\n  \"findings\": [\n    {\n      \"severity\": \"Critical|Warning|Info\",\n      \"dimension\": \"patterns|maintainability|integration|security|alignment\",\n      \"file\": \"path/to/file.ts\",\n      \"line\": 42,\n      \"description\": \"What is wrong\",\n      \"spec_reference\": \"Spec section (if applicable)\",\n      \"fix_suggestion\": \"How to fix\"\n    }\n  ],\n  \"passed_checks\": [\"List of verified aspects\"],\n  \"summary\": \"Overall assessment\"\n}\nEOF",
+     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{BACKEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nROLE_FILE: ~/.claude/.ccg/prompts/{{BACKEND_PRIMARY}}/reviewer.md\nReview proposal <proposal_id> implementation:\n[Backend review dimensions - same as 通道 A]\nOUTPUT (JSON): [same schema as 通道 A]\nEOF",
      run_in_background: true,
      timeout: 300000,
-     description: "{{FRONTEND_PRIMARY}}: patterns/integration review"
+     description: "{{BACKEND_PRIMARY}}: backend/logic review (BC)"
+   })
+   Bash({
+     command: "~/.claude/bin/codeagent-wrapper --progress --backend {{FRONTEND_PRIMARY}} {{GEMINI_MODEL_FLAG}}- \"{{WORKDIR}}\" <<'EOF'\nROLE_FILE: ~/.claude/.ccg/prompts/{{FRONTEND_PRIMARY}}/reviewer.md\nReview proposal <proposal_id> implementation:\n[Frontend review dimensions - same as 通道 A]\nOUTPUT (JSON): [same schema as 通道 A]\nEOF",
+     run_in_background: true,
+     timeout: 300000,
+     description: "{{FRONTEND_PRIMARY}}: patterns/integration review (BC)"
    })
    ```
 
-   **Step 3.2 (事件驱动)**: spawn 两个 Bash bg 后说明 task-id 然后 **turn end**。引擎在每个 task 完成时自动发 `<task-notification>`，主线在通知触发的新 turn 处理结果。**不调 TaskOutput**。两个 task 都收到通知后才进 step 3.3。
+   > ⚠️ 通道 B `codeagent-wrapper` 已标 **deprecated**，仅为不能升级 plugin 的环境保留。
+
+   **Step 3.2 (事件驱动)**：
+   - **通道 A（plugin）**：两个 Agent 同 message 内 spawn，主线接 ≤200 token 摘要。Agent 完成后自动返回结果。
+   - **通道 B（BC wrapper）**：spawn 两个 Bash bg 后说明 task-id 然后 **turn end**。引擎在每个 task 完成时自动发 `<task-notification>`，主线在通知触发的新 turn 处理结果。**不调 TaskOutput**。两个 task 都收到通知后才进 step 3.3。
 
    ⛔ **禁止**：调 `TaskOutput({block: true, timeout: 600000})` (旧 freeze poll 模式) / Kill task。
    ⚠️ **失败处理**：notification status=failed / exit ≠ 0 / parse 失败 → v1.7.87 标准 2-retry / 5s / 3-attempts；3 次全失败才降级单模型。