speccrew 0.7.26 → 0.7.28

package/.speccrew/agents/speccrew-product-manager.md

@@ -94,7 +94,7 @@ PM continues to next orchestration block
 | Phase 3 | speccrew-pm-requirement-clarify | ✅ dispatch-to-worker |
 | Phase 4a | speccrew-pm-requirement-model | ✅ dispatch-to-worker |
 | Phase 4b | speccrew-pm-requirement-analysis | ✅ dispatch-to-worker |
-| Phase 5 | speccrew-pm-phase5-subprd-dispatch | ✅ dispatch-to-worker |
+| Phase 5 | speccrew-pm-phase5-subprd-dispatch | ⚡ PM direct execution |
 | Phase 6 | speccrew-pm-phase6-verify-confirm | ✅ dispatch-to-worker |
 
 ---
@@ -155,7 +155,7 @@ Phase 0 done. Moving to Phase 1...
 | **Phase 4a** | ISA-95 Modeling (complex) | `speccrew-pm-requirement-model` | Worker dispatch |
 | **Phase 4b** | PRD Generation (complex) | `speccrew-pm-requirement-analysis` | Worker dispatch |
 | **Phase 4** | Simple PRD Generation | `speccrew-pm-requirement-simple` | Worker dispatch |
-| **Phase 5** | Sub-PRD Dispatch (complex) | `speccrew-pm-phase5-subprd-dispatch` | Parallel Worker dispatch |
+| **Phase 5** | Sub-PRD Dispatch (complex) | `speccrew-pm-phase5-subprd-dispatch` | PM Direct Orchestration |
 | **Phase 6** | Verification & Confirmation | `speccrew-pm-phase6-verify-confirm` | Worker dispatch |
 
 ---
@@ -400,21 +400,62 @@ Step 4b: Dispatch Worker with speccrew-pm-requirement-analysis
 
 ---
 
-## Phase 5: Sub-PRD Worker Dispatch
+## Phase 5: Sub-PRD Worker Dispatch (PM Direct Orchestration)
 
-<!-- Phase 5 的详细步骤已提取到 AgentFlow Skill: speccrew-pm-phase5-subprd-dispatch -->
-<!-- 执行时由 orchestration 通过 dispatch-to-worker 调用 -->
+<!-- Phase 5 是 PM 直接执行的编排 Skill，PM 必须读取 workflow.agentflow.xml 并按步骤执行 -->
+<!-- ⚠️ 这里的 dispatch-to-worker 块由 PM Agent 直接执行，不能委派给 Worker -->
+
+**Purpose**: As the orchestration layer, PM Agent directly coordinates batch dispatch of Sub-PRD generation tasks to Worker Agents.
+
+> 🛑 **CRITICAL ARCHITECTURE RULE (Harness Principle 17: Orchestration Layer Separation)**
+>
+> Phase 5 is an **orchestration skill** containing internal `dispatch-to-worker` blocks.
+> - Workers CANNOT dispatch Workers (execution hierarchy)
+> - Therefore, Phase 5 MUST be executed directly by PM Agent
+> - PM Agent reads `workflow.agentflow.xml` and executes each block step-by-step
+
+### 5.1 PM Agent Execution Protocol
+
+**PM Agent MUST:**
+1. Read the skill's `workflow.agentflow.xml` to understand execution steps
+2. Execute each block in order: read plan → init progress → dispatch workers → verify
+3. Use Agent tool to create `speccrew-task-worker` for EACH module
+4. Pass `speccrew-pm-sub-prd-generate` skill name to each Worker
 
-**Purpose**: Dispatch parallel Workers to generate Sub-PRD documents for each module.
+**PM Agent MUST NOT:**
+- Dispatch Phase 5 to a Worker (Worker cannot dispatch sub-Workers)
+- Skip reading the workflow.agentflow.xml
+- Generate Sub-PRD content directly
 
-### 5.1 Prerequisites
+### 5.2 Prerequisites
 
 - Phase 4b completed with valid Dispatch Plan
 - Master PRD exists
 - Dispatch Plan contains module list (count ≥ 2)
+- `.sub-prd-dispatch-plan.json` exists in iteration directory
+
+### 5.3 Workflow Steps (from workflow.agentflow.xml)
 
-### 5.2 Dispatch Strategy
+**Step 5.1: Read Dispatch Plan**
+- Read `.sub-prd-dispatch-plan.json` from iteration directory
+- Parse module list and verify required fields
 
+**Step 5.2: Initialize Progress Tracking**
+
+> 🛑 **MANDATORY: Initialize DISPATCH-PROGRESS.json BEFORE any Worker dispatch**
+>
+> PM Agent MUST:
+> 1. Create temp task file: `.tasks-temp.json`
+> 2. Run: `node {update_progress_script} init --file DISPATCH-PROGRESS.json --stage sub_prd_dispatch --tasks-file .tasks-temp.json`
+> 3. Verify initialization (Total: N | Pending: N | Completed: 0)
+>
+> **FORBIDDEN:**
+> - DO NOT dispatch ANY Worker before DISPATCH-PROGRESS.json is initialized
+> - DO NOT create DISPATCH-PROGRESS.json manually via create_file
+
+**Step 5.3: Batch Dispatch Workers**
+
+Dispatch Strategy:
 | Module Count | Dispatch Strategy |
 |--------------|-------------------|
 | 1-5 modules | Single batch, all parallel |
@@ -424,24 +465,25 @@ Step 4b: Dispatch Worker with speccrew-pm-requirement-analysis
 
 **BATCH SIZE = 5 (maximum parallel Workers per batch)**
 
-### 5.3 Worker Dispatch Rules
-
-> 🛑 **CRITICAL: ONE Worker per Module — NO EXCEPTIONS**
+> 🛑 **MANDATORY: Update Progress After Each Worker Completes**
+>
+> After each Worker returns:
+> 1. Run: `node {update_progress_script} update-task --file DISPATCH-PROGRESS.json --task-id {module_id} --status {completed|failed}`
+> 2. Verify the update succeeded
+>
+> **FORBIDDEN:**
+> - DO NOT skip progress update
+> - DO NOT proceed to next batch until current batch is verified
 
-**PM Agent MUST:**
-1. Read the Dispatch Plan from generate skill output
-2. Initialize DISPATCH-PROGRESS.json via update-progress.js script
-3. For EACH module: invoke ONE `speccrew-task-worker` with `speccrew-pm-sub-prd-generate`
-4. Pass ALL required context parameters to each worker
-5. Dispatch in batches of 5, wait for each batch to complete
-6. After each Worker completes, update DISPATCH-PROGRESS.json via script
+**Step 5.4: Failure Retry**
+- Check for tasks with `failed` status
+- Retry once if failures exist
+- Log persistent failures and continue
 
-**PM Agent MUST NOT:**
-- Generate Sub-PRD files directly (via create_file, write, or any file creation)
-- Invoke speccrew-pm-sub-prd-generate skill directly
-- Dispatch ONE Worker to handle MULTIPLE modules
-- Create or edit any Sub-PRD content as fallback
-- Skip worker dispatch and generate Sub-PRDs inline
+**Step 5.5: Result Verification**
+- Read final DISPATCH-PROGRESS.json
+- Verify all Sub-PRD files exist and size > 3KB
+- Update checkpoint via script
 
 ### 5.4 Worker Context Parameters
 
@@ -458,21 +500,18 @@ Step 4b: Dispatch Worker with speccrew-pm-requirement-analysis
 | `output_path` | Computed | `{output_dir}/{feature_name}-sub-{module_key}.md` |
 | `language` | Detected | User's language |
 
-### 5.5 Progress Tracking
-
-- Track progress in `DISPATCH-PROGRESS.json` via `update-progress.js` script
-- After ALL workers complete, verify Sub-PRD files exist
-- Update checkpoint `sub_prd_dispatch` to `passed`
-
-### 5.6 Completion Verification
+### 5.5 Completion Verification
 
 Before proceeding to Phase 6, verify:
-- [ ] All workers were dispatched via speccrew-task-worker (one Worker per module)
+- [ ] PM Agent executed Phase 5 directly (NOT dispatched to Worker)
+- [ ] DISPATCH-PROGRESS.json was initialized BEFORE any dispatch
+- [ ] All workers were dispatched via Agent tool (one Worker per module)
+- [ ] Progress was updated after each Worker completed
 - [ ] No Sub-PRD was generated by PM Agent directly
 - [ ] All workers completed (DISPATCH-PROGRESS.json counts.pending == 0)
 - [ ] All Sub-PRD files exist and have valid size
 - [ ] `.prd-feature-list.json` contains complete feature data
-- [ ] Checkpoint updated
+- [ ] Checkpoint updated via script
 
 ---
 

package/.speccrew/agents/speccrew-task-worker.md

@@ -112,9 +112,11 @@ When Worker loads a skill (via `skill_name` or `skill_path` parameter):
 **If `skill_path` or `skill_name` is provided:**
 1. If `skill_path` is provided, use it directly; otherwise use Skill Discovery
 2. If Skill file does not exist, immediately report error
-3. If `context` parameters exist, substitute them into placeholders in the Skill
-4. Strictly execute according to the workflow defined in the Skill
-5. Complete the task and output results
+3. **Read SKILL.md** to understand skill overview, constraints, and templates
+4. **Read workflow.agentflow.xml** — this is the AUTHORITATIVE execution definition
+5. If `context` parameters exist, substitute them into placeholders
+6. **Execute blocks defined in workflow.agentflow.xml sequentially** (top-to-bottom, announcing each block)
+7. Complete the task and output results
 
 > 🛑 **CRITICAL — Skill Execution Enforcement**:
 > - If the Skill specifies script execution via `run_in_terminal` or `Bash` → You MUST execute the script via terminal. DO NOT substitute with manual file creation.
@@ -122,6 +124,23 @@ When Worker loads a skill (via `skill_name` or `skill_path` parameter):
 > - If the Skill contains MANDATORY/FORBIDDEN constraints → You MUST follow them strictly.
 > - DO NOT improvise alternative execution paths. If a step fails, report the error — do not attempt workarounds.
 
+### MANDATORY: XML Workflow Loading Protocol
+
+When executing any Skill that contains `workflow.agentflow.xml`:
+
+1. **SKILL.md is METADATA ONLY** — it provides overview, constraints, and template references. It is NOT the execution plan.
+2. **workflow.agentflow.xml is the AUTHORITATIVE execution plan** — Worker MUST read this file and execute its blocks in sequential document order.
+3. **FORBIDDEN**: Starting task execution based solely on SKILL.md without reading workflow.agentflow.xml.
+4. **FORBIDDEN**: Summarizing or paraphrasing the workflow — execute blocks exactly as defined.
+5. **FORBIDDEN**: Skipping, reordering, or merging blocks.
+
+**Execution sequence:**
+```
+Read SKILL.md → Read workflow.agentflow.xml → Execute blocks in XML order → Report results
+```
+
+If workflow.agentflow.xml does not exist in the skill directory, fall back to SKILL.md-based execution.
+
 ### XML Workflow Block Announcement Protocol
 
 When executing a Skill that uses XML workflow format (`<workflow>` root element), you MUST follow the Block Execution Announcement Protocol defined in `docs/rules/agentflow-spec.md`:
@@ -211,10 +230,14 @@ When the task fails or is blocked, output:
 **MUST DO:**
 - If `skill_name` is provided, MUST use Skill Discovery to resolve the full path and strictly follow the Skill definition
 - If `skill_name` is provided but Skill file does not exist, immediately report error
+- If Skill directory contains workflow.agentflow.xml, MUST read it and execute blocks in sequential order — this is the authoritative execution plan
+- MUST announce each XML block before execution using the Block Announcement Protocol
 - Only process the single task assigned to the current Worker
 
 **MUST NOT DO:**
 - Do not skip or ignore a provided Skill file
+- Do not execute tasks based solely on SKILL.md when workflow.agentflow.xml exists — the XML workflow is authoritative
+- Do not skip reading workflow.agentflow.xml even if SKILL.md seems sufficient
 - Do not actively modify code beyond the task scope
 - Do not overstep to handle other tasks
 - Do not assume context information not provided

package/.speccrew/skills/speccrew-pm-phase0-init/SKILL.md

@@ -31,6 +31,12 @@ Applies workspace initialization principles:
 - Progress tracking via WORKFLOW-PROGRESS.json
 - Checkpoint recovery for session continuity
 
+## Templates Used
+
+| Template | Path | Purpose |
+|----------|------|---------|
+| WORKFLOW-PROGRESS | `templates/WORKFLOW-PROGRESS-TEMPLATE.json` | Standard workflow progress tracking with 7 predefined stages |
+
 ## Output Deliverables
 
 | Deliverable | Path | Description |
@@ -68,9 +74,15 @@ Applies workspace initialization principles:
 - Compute all path variables as absolute paths
 - Check for active iteration before creating new one
 - Copy requirement document to iteration's 00.docs directory
+- Create EXACTLY two subdirectories: `00.docs/` (original requirement documents) and `01.product-requirement/` (PRD and clarification outputs)
+- Initialize `WORKFLOW-PROGRESS.json` from `WORKFLOW-PROGRESS-TEMPLATE.json` template — stage names MUST match template exactly (01_prd, 02_feature_design, etc.)
 
 **Must not do:**
 - Manually create WORKFLOW-PROGRESS.json via Write/Edit tools
 - Skip checkpoint recovery check
 - Use relative paths in Worker dispatches
 - Create iteration directory without proper naming convention
+- Pre-create directories for later phases (`02.feature-design/`, `03.system-design/`, `04.development/`, `05.deployment/`, `06.system-test/`, `07.delivery/`) — each phase creates its own directory when needed
+- Invent directory names not defined in `workspace-structure.md`
+- Manually write WORKFLOW-PROGRESS.json content — MUST use template + update-progress.js script
+- Invent custom stage names (e.g., phase0_initialization, phase1_requirement_analysis) — use template-defined names only

package/.speccrew/skills/speccrew-pm-phase0-init/templates/WORKFLOW-PROGRESS-TEMPLATE.json

@@ -0,0 +1,56 @@
+{
+  "created_at": "{created_at}",
+  "updated_at": "{updated_at}",
+  "current_stage": null,
+  "stages": {
+    "01_prd": {
+      "status": "pending",
+      "started_at": null,
+      "completed_at": null,
+      "confirmed_at": null,
+      "output": null
+    },
+    "02_feature_design": {
+      "status": "pending",
+      "started_at": null,
+      "completed_at": null,
+      "confirmed_at": null,
+      "output": null
+    },
+    "03_system_design": {
+      "status": "pending",
+      "started_at": null,
+      "completed_at": null,
+      "confirmed_at": null,
+      "output": null
+    },
+    "04_development": {
+      "status": "pending",
+      "started_at": null,
+      "completed_at": null,
+      "confirmed_at": null,
+      "output": null
+    },
+    "05_deployment": {
+      "status": "pending",
+      "started_at": null,
+      "completed_at": null,
+      "confirmed_at": null,
+      "output": null
+    },
+    "06_system_test": {
+      "status": "pending",
+      "started_at": null,
+      "completed_at": null,
+      "confirmed_at": null,
+      "output": null
+    },
+    "07_delivery": {
+      "status": "pending",
+      "started_at": null,
+      "completed_at": null,
+      "confirmed_at": null,
+      "output": null
+    }
+  }
+}

package/.speccrew/skills/speccrew-pm-phase0-init/workflow.agentflow.xml

@@ -19,6 +19,8 @@
     <field name="text">ALL WORKFLOW-PROGRESS.json operations MUST use update-progress.js script</field>
     <field name="text">ALL path variables MUST be absolute paths</field>
     <field name="text">Check for active iteration before creating new one</field>
+    <field name="text">Phase 0 MUST create EXACTLY TWO subdirectories: 00.docs and 01.product-requirement</field>
+    <field name="text">FORBIDDEN: Pre-creating directories for later phases (02.feature-design, 03.system-design, 04.development, etc.). Each phase creates its own directory when needed.</field>
   </block>
 
   <block type="rule" id="R2" level="forbidden" desc="Phase 0 Forbidden Actions">
@@ -118,7 +120,24 @@
         <block type="task" id="P0-1-B9" action="run-script" status="pending" desc="Copy requirement document to iteration">
           <field name="command">Copy-Item "${user_requirement}" "${iteration_path}/00.docs/" -ErrorAction SilentlyContinue</field>
         </block>
-        
+                
+        <!-- Step 2.7: Load WORKFLOW-PROGRESS template -->
+        <block type="task" id="P0-1-B9A" action="read-file" status="pending" desc="Load WORKFLOW-PROGRESS template">
+          <field name="path" value="${skill_path}/templates/WORKFLOW-PROGRESS-TEMPLATE.json"/>
+          <field name="output" var="workflow_progress_template"/>
+        </block>
+                
+        <!-- Step 2.8: Generate initial WORKFLOW-PROGRESS.json from template -->
+        <block type="task" id="P0-1-B9B" action="generate" status="pending" desc="Create WORKFLOW-PROGRESS.json from template">
+          <field name="template" value="${workflow_progress_template}"/>
+          <field name="output_path" value="${iteration_path}/WORKFLOW-PROGRESS.json"/>
+          <field name="vars">
+            - created_at: ${current_timestamp}
+            - updated_at: ${current_timestamp}
+          </field>
+          <field name="output" var="workflow_progress_path"/>
+        </block>
+                
         <!-- Step 0.1.1: Initialize WORKFLOW-PROGRESS.json -->
         <block type="task" id="P0-1-B10" action="run-script" status="pending" desc="Initialize WORKFLOW-PROGRESS.json via script">
           <!-- MANDATORY: Use update-progress.js script -->

package/.speccrew/skills/speccrew-pm-phase5-subprd-dispatch/SKILL.md

@@ -2,15 +2,32 @@
 
 ## Description
 
-PM Phase 5 Sub-PRD Batch Dispatch Skill. Core orchestration component for Master-Sub PRD workflow, executed directly by PM Agent (NOT dispatched to Worker).
+PM Phase 5 Sub-PRD Batch Dispatch Skill. **Orchestration-layer skill** executed directly by PM Agent (NOT dispatched to Worker).
 
-PM Agent loads and executes this skill directly to coordinate batch dispatch of Sub-PRD generation tasks to Worker Agents, implementing:
-- Reading Dispatch Plan from Master PRD
-- Initializing dispatch progress tracking
-- Dispatching Workers in parallel batches for Sub-PRD generation per module
-- Failure retry and result verification
+This is an **orchestration skill** that PM Agent must execute directly:
+- PM Agent reads `workflow.agentflow.xml` and executes each block step-by-step
+- PM Agent uses Agent tool to dispatch Workers for Sub-PRD generation
+- Workers generate Sub-PRD files; PM Agent only coordinates
 
-**CRITICAL**: This skill contains dispatch-to-worker logic internally. It MUST be executed by PM Agent (who has Agent tool access), NOT by a Worker Agent (who cannot create sub-Workers).
+**CRITICAL ARCHITECTURE RULE (Harness Principle 17: Orchestration Layer Separation)**:
+- This skill contains internal `dispatch-to-worker` blocks
+- Workers CANNOT dispatch Workers (execution hierarchy constraint)
+- Therefore, this skill MUST be executed by PM Agent directly
+- PM Agent MUST NOT dispatch this skill to a Worker
+
+**Execution Method**: PM Agent reads `workflow.agentflow.xml` and follows the workflow steps.
+
+## MANDATORY: PM Must Read workflow.agentflow.xml
+
+> 🛑 **BEFORE executing this skill, PM Agent MUST:**
+> 1. Read the `workflow.agentflow.xml` file in this skill directory
+> 2. Parse each block and execute in order
+> 3. Follow the exact workflow steps defined in the XML
+>
+> **FORBIDDEN:**
+> - DO NOT execute this skill without reading the XML workflow
+> - DO NOT skip blocks defined in the XML
+> - DO NOT dispatch this skill to a Worker Agent
 
 ## Input Parameters
 
@@ -69,27 +86,22 @@ PM Agent loads and executes this skill directly to coordinate batch dispatch of
 ### Step 5.6: User Confirmation
 - [ ] Wait for user confirmation of Sub-PRD generation results
 
-## Key Rules
-
-### MANDATORY - Worker Dispatch Rules
-- **ONE Worker per Module** - Dispatch one independent Worker per Sub-PRD module
-- **PM Must Not Generate Directly** - PM Agent must NOT directly generate Sub-PRD content
-- **Must Use dispatch-to-worker** - All Workers must be executed via `dispatch-to-worker` action
-- **Direct Skill Invocation Forbidden** - Must NOT directly invoke `speccrew-pm-sub-prd-generate` skill
-
-### MANDATORY - Batch Processing Rules
-- **Batch Size = 5** - Maximum 5 parallel Workers per batch
-- **Parallel Dispatch** - Workers in the same batch must be dispatched simultaneously
-- **Sequential Wait** - Wait for current batch to complete before dispatching next batch
-- **Progress Update** - Immediately update DISPATCH-PROGRESS.json after each Worker completes
-
-### MANDATORY - Progress Tracking Rules
-- **Script Initialization** - DISPATCH-PROGRESS.json must be created via update-progress.js
-- **Manual Creation Forbidden** - Must NOT create progress files directly via create_file or PowerShell
-- **Idempotent Update** - Use `update-task` command to update individual task status
-
-### FORBIDDEN - Prohibited Actions
-- PM Agent directly generating Sub-PRD files is forbidden
-- Dispatching one Worker to handle multiple modules is forbidden
-- Skipping Worker dispatch as a fallback after failure is forbidden
-- Marking checkpoint as passed before user confirmation is forbidden
+## Must Do
+
+- **READ workflow.agentflow.xml FIRST** — PM Agent must parse the XML workflow before any execution
+- **Initialize DISPATCH-PROGRESS.json BEFORE any Worker dispatch** — Use update-progress.js init command
+- **Dispatch one Worker per module** — Use Agent tool to create speccrew-task-worker with speccrew-pm-sub-prd-generate skill
+- **Update progress after each Worker completes** — Use update-progress.js update-task command
+- **Verify all Sub-PRD files exist** — Check file existence and size > 3KB
+- **Update checkpoint via script** — Use update-progress.js write-checkpoint command
+
+## Must Not Do
+
+- **DO NOT dispatch this skill to a Worker** — Phase 5 is PM direct execution
+- **DO NOT skip reading workflow.agentflow.xml** — XML workflow is the execution guide
+- **DO NOT dispatch Workers before DISPATCH-PROGRESS.json is initialized** — Progress tracking is mandatory
+- **DO NOT create DISPATCH-PROGRESS.json manually** — Must use update-progress.js script
+- **DO NOT generate Sub-PRD content directly** — Only Workers generate Sub-PRDs
+- **DO NOT skip progress update after Worker completes** — Must call update-task for each completed Worker
+- **DO NOT dispatch one Worker for multiple modules** — One Worker per module
+- **DO NOT proceed to Phase 6 without verifying all Sub-PRDs** — Completion verification is mandatory

package/.speccrew/skills/speccrew-task-worker-execution/workflow.agentflow.xml

@@ -134,7 +134,21 @@
         <field name="output" var="skill_definition"/>
       </block>
       <block type="event" id="E8" action="log" desc="加载技能">📄 Loaded skill: ${resolved_skill_path}</block>
+            
+      <!-- MANDATORY: Load XML Workflow Definition -->
+      <block type="rule" id="R-WORKFLOW-FIRST" level="mandatory" desc="XML workflow is the authoritative execution definition">
+        <field name="text">MANDATORY: workflow.agentflow.xml is the AUTHORITATIVE execution definition. SKILL.md is supplementary context only.</field>
+        <field name="text">Worker MUST read workflow.agentflow.xml BEFORE starting any task execution.</field>
+        <field name="text">Worker MUST execute ONLY the blocks defined in workflow.agentflow.xml, in sequential document order.</field>
+        <field name="text">FORBIDDEN: Executing tasks based solely on SKILL.md description without reading the XML workflow.</field>
+      </block>
       
+      <block type="task" id="T6b" action="read-file" desc="MANDATORY: Read XML workflow definition">
+        <field name="path">${skill_path}/workflow.agentflow.xml</field>
+        <field name="output" var="skill_workflow_xml"/>
+      </block>
+      <block type="event" id="E8b" action="log" desc="Loaded XML workflow">📄 Loaded workflow: ${skill_path}/workflow.agentflow.xml — This is the AUTHORITATIVE execution plan. Execute blocks in document order.</block>
+            
       <!-- 加载 AgentFlow 规范 -->
       <block type="task" id="T7" action="read-file" desc="读取 AgentFlow 规范">
         <field name="path">${workspace_root}/speccrew-workspace/docs/rules/agentflow-spec.md</field>
@@ -143,10 +157,12 @@
       <block type="event" id="E9" action="log" desc="加载规范">📖 Loaded AgentFlow specification</block>
       
       <!-- 执行技能工作流 -->
-      <block type="task" id="T8" action="run-skill" desc="执行技能工作流">
+      <block type="task" id="T8" action="run-skill" desc="Execute skill workflow — follow XML blocks sequentially">
         <field name="skill">${resolved_skill_path}</field>
         <field name="context">${context}</field>
         <field name="skill_path">${skill_path}</field>
+        <field name="workflow_xml">${skill_workflow_xml}</field>
+        <field name="execution_mode">xml-block-sequential</field>
       </block>
       <field name="execution_status" value="success"/>
     </branch>

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "speccrew",
-  "version": "0.7.26",
+  "version": "0.7.28",
   "description": "Spec-Driven Development toolkit for AI-powered IDEs",
   "author": "charlesmu99",
   "repository": {