speccrew 0.7.36 → 0.7.38

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

@@ -175,6 +175,12 @@ Phase 0 done. Moving to Phase 1...
 - Iteration directory naming: `{number}-{type}-{name}` (e.g., `001-feature-litemes`)
 - DO NOT manually create WORKFLOW-PROGRESS.json — MUST use `update-progress.js` script
 
+**Progress Sync Recovery**: If WORKFLOW-PROGRESS.json or DISPATCH-PROGRESS.json exists but appears stale or inconsistent with actual file state, run:
+```
+node "speccrew-workspace/scripts/update-progress.js" sync --phase {current_phase}
+```
+This rebuilds progress from actual file system state, preventing phantom task tracking.
+
 ---
 
 ## Phase 1: Knowledge Base Availability Check
@@ -645,6 +651,7 @@ This agent MAY directly create/modify ONLY:
 3. ❌ DO NOT create DISPATCH-PROGRESS.json manually (use init script)
 4. ❌ DO NOT create any Sub-PRD content as fallback if worker fails
 5. ❌ DO NOT dispatch Sub-PRDs sequentially — use parallel batch (5/batch)
+6. ❌ DO NOT create temporary helper scripts (bash/powershell/node) for one-off operations — use existing workspace scripts or direct tool calls
 
 ## MANDATORY: Worker Dispatch Prompt Format (Harness Principle 22)
 
@@ -697,6 +704,19 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 3. Unrecoverable errors that prevent further progress
 4. Security-sensitive operations (e.g., deleting existing files)
 
+### FORBIDDEN ON SCRIPT FAILURE
+- When a script execution fails, MUST STOP immediately
+- NEVER provide A/B/C recovery options to the user
+- NEVER ask "should I try alternative approach?"
+- The ONLY permitted action: report the exact error and STOP
+
+### OUTPUT EFFICIENCY
+- Worker MUST write design/code content directly to files using tools
+- NEVER display file content in conversation messages
+- NEVER echo back what was written to a file
+- Response after file write: only confirm filename + status (e.g., "Created PRD.md ✓")
+- This reduces token waste and prevents context window overflow
+
 ---
 
 # Deliverables

package/.speccrew/agents/speccrew-system-deployer.md

@@ -56,6 +56,7 @@ Your core task is: execute build, database migration, service startup, and smoke
 | Phase 3 | HARD STOP | User must confirm deployment results before proceeding to testing |
 | ALL | ABORT ON FAILURE | If any skill invocation fails → STOP and report. Do NOT attempt manual recovery |
 | ALL | SCRIPT ENFORCEMENT | All .checkpoints.json and WORKFLOW-PROGRESS.json updates via update-progress.js |
+| ALL | ANTI-SCRIPT | Orchestrator/Worker MUST NOT create temporary helper scripts; all operations use existing workspace scripts or direct tool calls |
 
 ## MANDATORY SKILL ENFORCEMENT
 
@@ -105,6 +106,13 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 3. Unrecoverable errors that prevent further progress
 4. Skill invocation failure — report and wait for user decision
 
+### OUTPUT EFFICIENCY
+- Worker MUST write design/code content directly to files using tools
+- NEVER display file content in conversation messages
+- NEVER echo back what was written to a file
+- Response after file write: only confirm filename + status (e.g., "Created deployment-report.md ✓")
+- This reduces token waste and prevents context window overflow
+
 ## ABORT CONDITIONS
 
 > **If ANY of the following conditions occur, the System Deployer Agent MUST immediately STOP the workflow and report to user.**
@@ -119,6 +127,12 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 8. **Script Execution Failure**: `node ... update-progress.js` fails → STOP. Do NOT manually create/edit JSON files.
 9. **Techs Knowledge Missing**: Required deployment configuration not found in techs knowledge → STOP. Report missing configuration.
 
+### FORBIDDEN ON SCRIPT FAILURE
+- When a script execution fails, Worker MUST STOP immediately
+- NEVER provide A/B/C recovery options to the user
+- NEVER ask "should I try alternative approach?"
+- The ONLY permitted action: report the exact error and STOP
+
 ## TIMESTAMP INTEGRITY
 
 > **All timestamps in progress files (.checkpoints.json, WORKFLOW-PROGRESS.json) are generated exclusively by `update-progress.js` script.**
@@ -186,6 +200,12 @@ If WORKFLOW-PROGRESS.json does not exist:
 - Do not block execution due to missing progress files
 - Log informational message: "Progress tracking not available (WORKFLOW-PROGRESS.json not found). Running in compatibility mode."
 
+**Progress Sync Recovery**: If DISPATCH-PROGRESS.json exists but appears stale or inconsistent with actual file state, run:
+```
+node "speccrew-workspace/scripts/update-progress.js" sync --phase {current_phase}
+```
+This rebuilds progress from actual file system state, preventing phantom task tracking.
+
 ## Phase 0.5: IDE Directory Detection
 
 Before dispatching skills, detect the IDE directory for skill path resolution:

package/.speccrew/agents/speccrew-system-designer.md

@@ -129,6 +129,7 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 | Phase 6 | HARD STOP | User must confirm all designs before finalizing |
 | ALL | ABORT ON FAILURE | If any skill invocation fails → STOP and report. Do NOT generate content manually as fallback |
 | ALL | SCRIPT ENFORCEMENT | All .checkpoints.json and WORKFLOW-PROGRESS.json updates via update-progress.js script. Manual JSON creation FORBIDDEN |
+| ALL | ANTI-SCRIPT | Agent MUST NOT create custom scripts (.sh, .ps1, .js). Use only update-progress.js provided. Temporary PowerShell/Bash commands for JSON manipulation are FORBIDDEN |
 
 ## ABORT CONDITIONS
 
@@ -136,6 +137,12 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 
 1. **Skill Invocation Failure**: Framework evaluation skill or any platform design skill call returns error → STOP. Do NOT generate content manually.
 2. **Script Execution Failure**: `node ... update-progress.js` fails → STOP. Do NOT manually create/edit JSON files.
+
+   **FORBIDDEN ON SCRIPT FAILURE**:
+   - DO NOT provide A/B/C alternative options
+   - DO NOT suggest "skip to next phase"
+   - DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ONLY correct response: "STOP: update-progress.js failed with [error]. Task: [id]. Command: [cmd]."
 3. **Missing Intermediate Artifacts**: Feature Spec not found, API Contract missing, or framework-evaluation.md not generated → STOP.
 4. **User Rejection**: User rejects framework evaluation, DESIGN-OVERVIEW, or Joint Confirmation → STOP, ask for specific revision requirements.
 5. **Worker Batch Failure**: If >50% workers in a batch fail → STOP entire batch, report to user with failure details.
@@ -656,6 +663,11 @@ Before dispatching, create or update dispatch tracking:
 
 2. **Check existing progress** (from Step 0.3) — skip `completed` tasks
 3. **Update status** to `in_progress` for tasks being dispatched:
+4. **If progress file appears out-of-sync** (many tasks show "pending" but output files already exist):
+   ```bash
+   node {update_progress_script} sync --file {iterations_dir}/{current}/03.system-design/DISPATCH-PROGRESS.json --dir {iterations_dir}/{current}/03.system-design --suffix "-design.md"
+   ```
+   This recovers progress from actual files on disk.
    ```bash
    node {update_progress_script} update-task --file {iterations_dir}/{current}/03.system-design/DISPATCH-PROGRESS.json --task-id {task_id} --status in_progress
    ```

package/.speccrew/agents/speccrew-system-developer.md

@@ -65,6 +65,7 @@ Your core task is: based on the System Design (HOW to build), execute and coordi
 | Phase 5 | INTEGRATION-CHECK | Cross-platform API & data consistency MUST be verified before delivery |
 | ALL | ABORT ON FAILURE | If any worker fails → STOP and report. Do NOT generate code manually as fallback |
 | ALL | SCRIPT ENFORCEMENT | All progress file updates via update-progress.js script. Manual JSON creation FORBIDDEN |
+| ALL | ANTI-SCRIPT | Agent and Workers MUST NOT create custom automation scripts (.sh, .ps1, .js). Use only update-progress.js provided. Temporary PowerShell/Bash commands for JSON manipulation are FORBIDDEN |
 
 ## MANDATORY WORKER ENFORCEMENT
 
@@ -158,6 +159,11 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 3. **Worker Invocation Failure**: speccrew-task-worker call fails or returns error → STOP. Do NOT attempt to write code as fallback.
 4. **Review Worker Failure**: Platform-specific review skill (speccrew-dev-review-*) fails after maximum re-dispatch attempts (3) → STOP. Report review blocker.
 5. **Script Execution Failure**: `node ... update-progress.js` fails → STOP. Do NOT manually create/edit JSON files.
+   **FORBIDDEN ON SCRIPT FAILURE**:
+   - DO NOT provide A/B/C alternative options
+   - DO NOT suggest "skip to next phase"
+   - DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ONLY correct response: "STOP: update-progress.js failed with [error]. Task: [id]. Command: [cmd]."
 6. **Batch Failure Threshold**: If >50% workers in a batch fail → STOP entire batch, report to user with failure details.
 7. **Code Quality Deadlock**: If review identifies unfixable issues after 3 re-dispatch attempts → STOP and report as technical debt.
 8. **Cross-platform Integration Failure**: Critical API/data inconsistencies detected in Phase 5 that block downstream testing → STOP and report integration risks.
@@ -244,6 +250,11 @@ Check for existing dispatch progress to support module-level retry:
    ```
 
 4. **If DISPATCH-PROGRESS.json does not exist**: Will create fresh dispatch progress
+5. **If progress file appears out-of-sync** (many tasks show "pending" but output files already exist):
+   ```bash
+   node {update_progress_script} sync --file {iterations_dir}/{current}/04.development/DISPATCH-PROGRESS.json --dir {iterations_dir}/{current}/04.development --suffix "-task.md"
+   ```
+   This recovers progress from actual files on disk. Use when progress tracking was interrupted.
 
 ### Phase 0.4: Backward Compatibility
 

package/.speccrew/agents/speccrew-team-leader.md

@@ -233,6 +233,7 @@ Turn 3: Create Task(C) → wait for completion
 - Do NOT manually create JSON files when a Skill should generate them
 - Do NOT execute analysis work that should be dispatched to a Worker via Task tool
 - Do NOT stop between stages to ask the user for direction
+- Do NOT create temporary helper scripts (bash/powershell/node scripts) for one-off operations — use existing workspace scripts or direct tool calls
 
 ## CONTINUOUS EXECUTION RULES
 
@@ -254,6 +255,21 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 3. Unrecoverable errors that prevent further progress
 4. Security-sensitive operations (e.g., deleting existing files)
 
+### FORBIDDEN ON SCRIPT FAILURE
+
+- When a script execution fails, MUST STOP immediately
+- NEVER provide A/B/C recovery options to the user
+- NEVER ask "should I try alternative approach?"
+- The ONLY permitted action: report the exact error and STOP
+
+### OUTPUT EFFICIENCY
+
+- Worker MUST write design/code content directly to files using tools
+- NEVER display file content in conversation messages
+- NEVER echo back what was written to a file
+- Response after file write: only confirm filename + status (e.g., "Created config.json ✓")
+- This reduces token waste and prevents context window overflow
+
 ### Batch Execution Behavior
 
 - When multiple items need processing, process ALL of them sequentially without asking

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

@@ -68,6 +68,7 @@ Phase 6: Delivery Summary
 | Phase 6 | JOINT-CONFIRMATION | All test reports must be confirmed by user before delivery |
 | ALL | ABORT ON FAILURE | If any worker fails → STOP and report. Do NOT generate artifacts manually as fallback |
 | ALL | SCRIPT ENFORCEMENT | All progress file updates via update-progress.js script. Manual JSON creation FORBIDDEN |
+| ALL | ANTI-SCRIPT | Orchestrator/Worker MUST NOT create temporary helper scripts; all operations use existing workspace scripts or direct tool calls |
 
 ## TIMESTAMP INTEGRITY
 
@@ -147,6 +148,12 @@ If ANY of these occur, workflow is INVALID:
 7. **Critical Bugs Found**: Unresolved critical/high-severity bugs block delivery → STOP before delivery phase.
 8. **Cross-platform Inconsistency**: Test results inconsistent across platforms indicating environment issues → STOP and diagnose.
 
+### FORBIDDEN ON SCRIPT FAILURE
+- When a script execution fails, Worker MUST STOP immediately
+- NEVER provide A/B/C recovery options to the user
+- NEVER ask "should I try alternative approach?"
+- The ONLY permitted action: report the exact error and STOP
+
 ---
 
 ## CONTINUOUS EXECUTION RULES
@@ -176,6 +183,13 @@ This agent MUST execute tasks continuously without unnecessary interruptions.
 - If context window is approaching limit, save progress to checkpoint and inform user how to resume
 - NEVER voluntarily stop mid-batch to ask if user wants to continue
 
+### OUTPUT EFFICIENCY
+- Worker MUST write design/code content directly to files using tools
+- NEVER display file content in conversation messages
+- NEVER echo back what was written to a file
+- Response after file write: only confirm filename + status (e.g., "Created src/auth.ts ✓")
+- This reduces token waste and prevents context window overflow
+
 # Workflow
 
 ## Phase 0: Workflow Progress Management
@@ -244,6 +258,12 @@ Dispatch Resume Status:
 └── test_execution: {completed}/{total} completed, {failed} failed, {pending} pending
 ```
 
+**Progress Sync Recovery**: If DISPATCH-PROGRESS.json exists but appears stale or inconsistent with actual file state, run:
+```
+node "{workspace_path}/scripts/update-progress.js" sync --phase {current_phase}
+```
+This rebuilds progress from actual file system state, preventing phantom task tracking.
+
 ### Step 0.4: Backward Compatibility
 
 If `WORKFLOW-PROGRESS.json` does not exist:

package/.speccrew/skills/speccrew-dev-backend/SKILL.md

@@ -295,6 +295,64 @@ If the skill fails at any step:
 - `RUNTIME_ERROR`: Service startup failure, runtime exception
 - `BLOCKED`: Blocked by external dependency or unresolved design issue
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All implementation code, task records, and helper scripts MUST be written directly to output files
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Implementing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full source code blocks or file contents
+   - ❌ Complete implementation listings
+   - ❌ Large configuration file dumps
+   - ❌ Architecture diagrams displayed in chat
+   - ❌ API endpoint listings longer than 3 lines
+4. **Rationale**: Workers run in batch mode (up to 6 concurrent). Displaying code content in conversation wastes context window and provides no value since content goes to files anyway.
+
+## ABORT CONDITIONS
+
+When script execution or build/compile fails:
+
+1. **STOP immediately** — Report: Task ID, error message, failed command
+2. **FORBIDDEN responses on failure**:
+   - ❌ DO NOT provide A/B/C alternative options
+   - ❌ DO NOT suggest "skip this step and continue"
+   - ❌ DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ❌ DO NOT create temporary scripts to work around the issue
+3. **ONLY correct response**: Report the failure in Task Completion Report with status FAILED and error details
+
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All implementation code, task records, and helper scripts MUST be written directly to output files
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Implementing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full source code blocks or file contents
+   - ❌ Complete implementation listings
+   - ❌ Large configuration file dumps
+   - ❌ Architecture diagrams displayed in chat
+   - ❌ API endpoint listings longer than 3 lines
+4. **Rationale**: Workers run in batch mode (up to 6 concurrent). Displaying code content in conversation wastes context window and provides no value since content goes to files anyway.
+
+## ABORT CONDITIONS
+
+When script execution or build/compile fails:
+
+1. **STOP immediately** — Report: Task ID, error message, failed command
+2. **FORBIDDEN responses on failure**:
+   - ❌ DO NOT provide A/B/C alternative options
+   - ❌ DO NOT suggest "skip this step and continue"
+   - ❌ DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ❌ DO NOT create temporary scripts to work around the issue
+3. **ONLY correct response**: Report the failure in Task Completion Report with status FAILED and error details
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-dev-backend/workflow.agentflow.xml

@@ -81,6 +81,7 @@
     <block type="task" id="B10" action="create-file" desc="Copy template to Task Record path">
       <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/04.development/${platform_id}/[module]-task.md</field>
       <field name="content">${task_template} with top-level placeholders replaced</field>
+      <field name="encoding">utf8</field>
       <field name="output" var="task_record_path"/>
     </block>
 
@@ -195,6 +196,7 @@
         <block type="task" id="B21" action="create-file" desc="Write technical debt document">
           <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/tech-debt/[module]-tech-debt.md</field>
           <field name="content">${tech_debt} using unified template</field>
+          <field name="encoding">utf8</field>
           <field name="output" var="tech_debt_path"/>
         </block>
       </branch>

package/.speccrew/skills/speccrew-dev-desktop-electron/SKILL.md

@@ -321,6 +321,35 @@ If the skill fails at any step:
 
 ---
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All implementation code, task records, and helper scripts MUST be written directly to output files
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Implementing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full source code blocks or file contents
+   - ❌ Complete implementation listings
+   - ❌ Large configuration file dumps
+   - ❌ Architecture diagrams displayed in chat
+   - ❌ API endpoint listings longer than 3 lines
+4. **Rationale**: Workers run in batch mode (up to 6 concurrent). Displaying code content in conversation wastes context window and provides no value since content goes to files anyway.
+
+## ABORT CONDITIONS
+
+When script execution or build/compile fails:
+
+1. **STOP immediately** — Report: Task ID, error message, failed command
+2. **FORBIDDEN responses on failure**:
+   - ❌ DO NOT provide A/B/C alternative options
+   - ❌ DO NOT suggest "skip this step and continue"
+   - ❌ DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ❌ DO NOT create temporary scripts to work around the issue
+3. **ONLY correct response**: Report the failure in Task Completion Report with status FAILED and error details
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-dev-desktop-electron/workflow.agentflow.xml

@@ -91,6 +91,7 @@
     <block type="task" id="B9" action="create-file" desc="Copy template to Task Record path">
       <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/04.development/${platform_id}/[feature-name]-task.md</field>
       <field name="content">${task_template} with top-level placeholders replaced</field>
+      <field name="encoding">utf8</field>
       <field name="output" var="task_record_path"/>
     </block>
 
@@ -211,6 +212,7 @@
         <block type="task" id="B22" action="create-file" desc="Write technical debt document">
           <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/tech-debt/[feature-name]-tech-debt.md</field>
           <field name="content">${tech_debt} with categories: Security, Performance, Refactoring, Dependencies</field>
+          <field name="encoding">utf8</field>
           <field name="output" var="tech_debt_path"/>
         </block>
       </branch>

package/.speccrew/skills/speccrew-dev-desktop-tauri/SKILL.md

@@ -314,6 +314,35 @@ If the skill fails at any step:
 
 ---
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All implementation code, task records, and helper scripts MUST be written directly to output files
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Implementing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full source code blocks or file contents
+   - ❌ Complete implementation listings
+   - ❌ Large configuration file dumps
+   - ❌ Architecture diagrams displayed in chat
+   - ❌ API endpoint listings longer than 3 lines
+4. **Rationale**: Workers run in batch mode (up to 6 concurrent). Displaying code content in conversation wastes context window and provides no value since content goes to files anyway.
+
+## ABORT CONDITIONS
+
+When script execution or build/compile fails:
+
+1. **STOP immediately** — Report: Task ID, error message, failed command
+2. **FORBIDDEN responses on failure**:
+   - ❌ DO NOT provide A/B/C alternative options
+   - ❌ DO NOT suggest "skip this step and continue"
+   - ❌ DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ❌ DO NOT create temporary scripts to work around the issue
+3. **ONLY correct response**: Report the failure in Task Completion Report with status FAILED and error details
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-dev-desktop-tauri/workflow.agentflow.xml

@@ -89,6 +89,7 @@
     <block type="task" id="B9" action="create-file" desc="Copy template to Task Record path">
       <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/04.development/${platform_id}/[feature-name]-task.md</field>
       <field name="content">${task_template} with top-level placeholders replaced</field>
+      <field name="encoding">utf8</field>
       <field name="output" var="task_record_path"/>
     </block>
 
@@ -198,6 +199,7 @@
         <block type="task" id="B21" action="create-file" desc="Write technical debt document">
           <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/tech-debt/[feature-name]-tech-debt.md</field>
           <field name="content">${tech_debt} with categories: Security, Performance, Refactoring, Dependencies</field>
+          <field name="encoding">utf8</field>
           <field name="output" var="tech_debt_path"/>
         </block>
       </branch>

package/.speccrew/skills/speccrew-dev-frontend/SKILL.md

@@ -269,6 +269,35 @@ If the skill fails at any step:
 - `RUNTIME_ERROR`: Runtime exception in browser/dev server
 - `BLOCKED`: Blocked by external dependency or unresolved design issue
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All implementation code, task records, and helper scripts MUST be written directly to output files
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Implementing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full source code blocks or file contents
+   - ❌ Complete implementation listings
+   - ❌ Large configuration file dumps
+   - ❌ Architecture diagrams displayed in chat
+   - ❌ API endpoint listings longer than 3 lines
+4. **Rationale**: Workers run in batch mode (up to 6 concurrent). Displaying code content in conversation wastes context window and provides no value since content goes to files anyway.
+
+## ABORT CONDITIONS
+
+When script execution or build/compile fails:
+
+1. **STOP immediately** — Report: Task ID, error message, failed command
+2. **FORBIDDEN responses on failure**:
+   - ❌ DO NOT provide A/B/C alternative options
+   - ❌ DO NOT suggest "skip this step and continue"
+   - ❌ DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ❌ DO NOT create temporary scripts to work around the issue
+3. **ONLY correct response**: Report the failure in Task Completion Report with status FAILED and error details
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-dev-frontend/workflow.agentflow.xml

@@ -104,6 +104,7 @@
     <block type="task" id="B12" action="create-file" desc="Copy template to Task Record path">
       <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/04.development/${platform_id}/{feature-name}-tasks.md</field>
       <field name="content">${task_template} with top-level placeholders replaced</field>
+      <field name="encoding">utf8</field>
       <field name="output" var="task_record_path"/>
     </block>
 
@@ -221,6 +222,7 @@
         <block type="task" id="B24" action="create-file" desc="Write tech debt document">
           <field name="path">speccrew-workspace/iterations/{number}-{type}-{name}/tech-debt/{debt-id}.md</field>
           <field name="content">${tech_debt}</field>
+          <field name="encoding">utf8</field>
           <field name="output" var="tech_debt_path"/>
         </block>
       </branch>

package/.speccrew/skills/speccrew-dev-mobile/SKILL.md

@@ -252,6 +252,35 @@ If the skill fails at any step:
 - `RUNTIME_ERROR`: App crash on simulator/emulator, runtime exception
 - `BLOCKED`: Blocked by external dependency, native module issue, or unresolved design issue
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All implementation code, task records, and helper scripts MUST be written directly to output files
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Implementing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full source code blocks or file contents
+   - ❌ Complete implementation listings
+   - ❌ Large configuration file dumps
+   - ❌ Architecture diagrams displayed in chat
+   - ❌ API endpoint listings longer than 3 lines
+4. **Rationale**: Workers run in batch mode (up to 6 concurrent). Displaying code content in conversation wastes context window and provides no value since content goes to files anyway.
+
+## ABORT CONDITIONS
+
+When script execution or build/compile fails:
+
+1. **STOP immediately** — Report: Task ID, error message, failed command
+2. **FORBIDDEN responses on failure**:
+   - ❌ DO NOT provide A/B/C alternative options
+   - ❌ DO NOT suggest "skip this step and continue"
+   - ❌ DO NOT run ad-hoc PowerShell/Bash commands as workaround
+   - ❌ DO NOT create temporary scripts to work around the issue
+3. **ONLY correct response**: Report the failure in Task Completion Report with status FAILED and error details
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-dev-mobile/workflow.agentflow.xml

@@ -92,6 +92,7 @@
     <block type="task" id="B9" action="create-file" desc="Copy template to Task Record path">
       <field name="path">speccrew-workspace/iterations/{iteration}/04.development/${platform_id}/{module}-task.md</field>
       <field name="content">${task_template} with top-level placeholders replaced</field>
+      <field name="encoding">utf8</field>
       <field name="output" var="task_record_path"/>
     </block>
 
@@ -199,6 +200,7 @@
         <block type="task" id="B22" action="create-file" desc="Write technical debt document">
           <field name="path">speccrew-workspace/iterations/{iteration}/tech-debt/{feature-name}-tech-debt.md</field>
           <field name="content">${tech_debt} with categories: Platform-specific workarounds, Offline sync, Error handling, Accessibility, Test coverage</field>
+          <field name="encoding">utf8</field>
           <field name="output" var="tech_debt_path"/>
         </block>
       </branch>

package/.speccrew/skills/speccrew-fd-api-contract/SKILL.md

@@ -31,7 +31,8 @@ tools: Read, Write, Glob, Grep
    - Feature Spec document (`feature_spec_path`) does not exist or is empty → STOP
    - API Contract template file does not exist → STOP
    - `node ... update-progress.js` script execution fails → **HARD STOP**: Do NOT manually create or edit JSON progress files. Report the script error and wait for user resolution.
-   - User rejects Joint Confirmation → STOP, ask user for specific revision requirements
+
+> **NOTE**: This skill does NOT include user confirmation. Confirmation is handled at the orchestrator/dispatcher level after all features are processed. This enables continuous batch execution.
 
 ## Step 1: Read Input
 
@@ -105,68 +106,11 @@ Fill each section with API contract details from Step 2 and Step 3.
 
 For each API, locate its section anchor in the template and use `search_replace` to fill request parameters, response structure, success example, and error codes.
 
-## Step 5: Joint Confirmation
-
-After both documents (Feature Spec + API Contract) are ready, request user confirmation:
-
-**Conditional Confirmation Logic:**
-```
-IF feature_id is provided THEN
-  → Execute Feature-granular confirmation (情况 A)
-ELSE
-  → Execute Module-level confirmation (情况 B)
-```
-
-**情况 A：提供了 feature_id（Feature 粒度确认）**
-```
-Feature 设计阶段交付物已准备就绪：
-
-📋 Feature 信息：
-   - Feature ID: {feature_id}
-   - Feature 名称: {feature_name}
-
-📄 文档清单：
-   - Feature Spec: speccrew-workspace/iterations/{number}-{type}-{name}/02.feature-design/{feature-id}-{feature-name}-feature-spec.md
-   - API Contract: speccrew-workspace/iterations/{number}-{type}-{name}/03.api-contract/{feature-id}-{feature-name}-api-contract.md
-
-请确认以下关键点：
-1. 当前 Feature 的技术方案是否可行？
-2. API 定义是否满足前端需求？
-3. 数据模型设计是否合理？
-
-⚠️ 确认后，API Contract 将成为前后端协作的唯一基准。
-   在设计/开发阶段只读引用，不允许修改。
-   如需变更，必须返回此阶段重新确认。
-
-✅ 当前仅确认单个 Feature 的交付物。
-   全局阶段状态（02_feature_design）将在所有 Feature 完成后由 Feature Designer Agent 统一更新。
-```
-
-**情况 B：未提供 feature_id（向后兼容，模块级确认）**
-```
-Feature design phase deliverables are ready:
-- Feature Spec: speccrew-workspace/iterations/{number}-{type}-{name}/02.feature-design/[feature-name]-feature-spec.md
-- API Contract: speccrew-workspace/iterations/{number}-{type}-{name}/03.api-contract/[feature-name]-api-contract.md
-
-Please confirm the following key points:
-1. Is the overall technical solution feasible?
-2. Does the API definition meet frontend requirements?
-3. Is the data model reasonable?
-
-⚠️ After confirmation, the API contract will be the sole baseline for frontend-backend collaboration.
-   Read-only reference in design/development phase, no modifications allowed.
-   If changes are needed, must return to this phase for re-confirmation.
-
-After confirmation, you can start frontend and backend Designer Agents separately.
-```
-
-## Step 6: Update Progress Files
-
-After user confirms Joint Confirmation:
+## Step 5: Update Progress Files
 
 > **SCRIPT ENFORCEMENT RULE**: All `.checkpoints.json` and `WORKFLOW-PROGRESS.json` updates MUST be performed via `node speccrew-workspace/scripts/update-progress.js` commands. Manually creating or editing these JSON files is FORBIDDEN. If the script fails, STOP and report the error — do NOT attempt manual JSON construction.
 
-### 6a: Update Checkpoints File
+### 5a: Update Checkpoints File
 
 Update the `.checkpoints.json` file to record confirmation status.
 
@@ -229,7 +173,7 @@ Write/Update `.checkpoints.json`：
 - Preserve existing checkpoint statuses when updating
 - Log: "✅ Checkpoint (api_contract_joint) passed and recorded"
 
-### 6b: Update Workflow Progress
+### 5b: Update Workflow Progress
 
 Update `WORKFLOW-PROGRESS.json` to reflect current feature/module status.
 
@@ -292,7 +236,7 @@ Update `WORKFLOW-PROGRESS.json` to reflect current feature/module status.
 > Feature 粒度的 API Contract 完成后，全局阶段状态（`02_feature_design.status` 和 `current_stage`）**不由本 Skill 更新**。
 > 全局状态由 **Feature Designer Agent** 统一管理，当检测到所有 Feature 都完成时，统一更新为 `confirmed` 并推进到下一阶段。
 
-### 6.3 Backward Compatibility
+### 5.3 Backward Compatibility
 
 If `WORKFLOW-PROGRESS.json` does not exist:
 - Log: "⚠️ No workflow progress file found. Skipping workflow update."
@@ -307,4 +251,4 @@ If `WORKFLOW-PROGRESS.json` does not exist:
 - [ ] URL naming conforms to backend architecture specifications
 - [ ] Error code list is complete
 - [ ] File has been written to correct path (with proper naming convention based on feature_id)
-- [ ] Summary of both documents has been shown to user and confirmation requested (including Feature ID if applicable)
+- [ ] API Contract document has been generated at the correct path

package/.speccrew/skills/speccrew-fd-api-contract/workflow.agentflow.xml

@@ -150,50 +150,13 @@ ${api_contract_details.content}</field>
 ${api_list.error_codes}</field>
     </block>
 
-    <!-- Step 7: Joint Confirmation -->
-    <block type="gateway" id="G3" mode="exclusive" desc="Select confirmation mode">
-      <branch test="${feature_id} != null" name="Feature-granular confirmation">
-        <block type="event" id="E1" action="confirm" title="Feature Design Confirmation" type="yesno" desc="Request user confirmation for feature">
-          <field name="preview">Feature design phase deliverables ready:
-
-Feature ID: ${feature_id}
-Feature Name: ${feature_name}
-
-Documents:
-- Feature Spec: ${iteration_path}/02.feature-design/${feature_id}-${feature_name}-feature-spec.md
-- API Contract: ${iteration_path}/03.api-contract/${feature_id}-${feature_name}-api-contract.md
-
-Please confirm:
-1. Is the technical solution feasible?
-2. Does the API definition meet frontend requirements?
-3. Is the data model reasonable?
-
-After confirmation, API Contract becomes the sole baseline for frontend-backend collaboration.</field>
-        </block>
-      </branch>
-      <branch default="true" name="Module-level confirmation">
-        <block type="event" id="E2" action="confirm" title="Feature Design Confirmation" type="yesno" desc="Request user confirmation">
-          <field name="preview">Feature design phase deliverables ready:
-- Feature Spec: ${iteration_path}/02.feature-design/${feature_name}-feature-spec.md
-- API Contract: ${iteration_path}/03.api-contract/${feature_name}-api-contract.md
-
-Please confirm:
-1. Is the overall technical solution feasible?
-2. Does the API definition meet frontend requirements?
-3. Is the data model reasonable?
-
-After confirmation, API Contract becomes the sole baseline for frontend-backend collaboration.</field>
-        </block>
-      </branch>
-    </block>
-
-    <!-- Step 8a: Update Checkpoints File -->
+    <!-- Step 7a: Update Checkpoints File -->
     <block type="task" id="B9" action="run-script" desc="Update checkpoints via script">
       <field name="command">node "${workspace_path}/scripts/update-progress.js" write-checkpoint --file "${iteration_path}/02.feature-design/.checkpoints.json" --stage 02_feature_design --checkpoint api_contract_joint --passed true</field>
       <field name="output" var="checkpoint_result"/>
     </block>
 
-    <!-- Step 8b: Update Workflow Progress -->
+    <!-- Step 7b: Update Workflow Progress -->
     <block type="gateway" id="G4" mode="exclusive" desc="Update progress based on mode">
       <branch test="${feature_id} != null" name="Feature mode">
         <block type="task" id="B10" action="run-script" desc="Update workflow progress for feature">

package/.speccrew/skills/speccrew-sd-backend/SKILL.md

@@ -262,6 +262,23 @@ After completing all steps, output a structured completion report for the System
 - `VALIDATION_ERROR`: Input validation failed (e.g., invalid Feature Spec format)
 - `BLOCKED`: Blocked by external dependency or prerequisite not met
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All design content (architecture diagrams, API mappings, component specifications, data models) MUST be written directly to the output file
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Designing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full document sections or drafts
+   - ❌ Mermaid diagrams displayed in chat
+   - ❌ API endpoint listings
+   - ❌ Data model tables
+   - ❌ Architecture descriptions longer than 2 lines
+4. **Rationale**: Workers run in batch mode. Displaying design content in conversation wastes context window and provides no value since content goes to file anyway.
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-sd-desktop/SKILL.md

@@ -252,6 +252,23 @@ After completing all steps, output a structured completion report for the System
 - `VALIDATION_ERROR`: Input validation failed (e.g., invalid Feature Spec format)
 - `BLOCKED`: Blocked by external dependency or prerequisite not met
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All design content (architecture diagrams, API mappings, component specifications, data models) MUST be written directly to the output file
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Designing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full document sections or drafts
+   - ❌ Mermaid diagrams displayed in chat
+   - ❌ API endpoint listings
+   - ❌ Data model tables
+   - ❌ Architecture descriptions longer than 2 lines
+4. **Rationale**: Workers run in batch mode. Displaying design content in conversation wastes context window and provides no value since content goes to file anyway.
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-sd-framework-evaluate/SKILL.md

@@ -221,6 +221,23 @@ Failed At: Step {N}
 --- END REPORT ---
 ```
 
+## OUTPUT EFFICIENCY RULES
+
+When executing this skill:
+
+1. **Direct-to-File Output**: All design content (architecture diagrams, API mappings, component specifications, data models) MUST be written directly to the output file
+2. **Minimal Conversation Output**: Only output:
+   - Block execution announcements (1 line each): `"[Block XX] Designing..."`
+   - Error messages requiring attention
+   - Task Completion Report (final summary)
+3. **FORBIDDEN in conversation**:
+   - ❌ Full document sections or drafts
+   - ❌ Mermaid diagrams displayed in chat
+   - ❌ API endpoint listings
+   - ❌ Data model tables
+   - ❌ Architecture descriptions longer than 2 lines
+4. **Rationale**: Workers run in batch mode. Displaying design content in conversation wastes context window and provides no value since content goes to file anyway.
+
 # Key Rules
 
 | Rule | Description |

package/.speccrew/skills/speccrew-system-designer-orchestration/workflow.agentflow.xml

@@ -46,6 +46,15 @@
       <field name="command">node ${update_progress_script} read --file ${iterations_dir}/${current_iteration}/03.system-design/DISPATCH-PROGRESS.json --summary</field>
       <field name="output" var="dispatch_progress"/>
     </block>
+
+    <!-- Step 0.4: Sync Recovery — Fix out-of-sync dispatch progress -->
+    <!-- When dispatch progress shows many "pending" tasks but output files already exist on disk,
+         use sync subcommand to recover progress from actual files instead of re-dispatching. -->
+    <block type="rule" id="P0-R1" level="mandatory" desc="Sync recovery rule for out-of-sync dispatch progress">
+      <field name="text">If DISPATCH-PROGRESS.json shows many tasks as "pending" but corresponding output files already exist on disk, use sync subcommand to recover progress</field>
+      <field name="text">DO NOT re-dispatch tasks that already have output files — sync first to update progress from disk</field>
+      <field name="text">Sync command: node ${update_progress_script} sync --file ${iterations_dir}/${current_iteration}/03.system-design/DISPATCH-PROGRESS.json --dir ${iterations_dir}/${current_iteration}/03.system-design --suffix "-design.md"</field>
+    </block>
   </sequence>
 
   <!-- ========== Phase 0.5: IDE Directory Detection ========== -->

package/.speccrew/skills/speccrew-system-developer-orchestration/workflow.agentflow.xml

@@ -46,6 +46,13 @@
       <field name="command">node ${update_progress_script} read --file ${iterations_dir}/${current_iteration}/04.development/DISPATCH-PROGRESS.json --summary</field>
       <field name="output" var="dispatch_progress"/>
     </block>
+
+    <!-- Progress Sync Recovery -->
+    <block type="rule" id="P0-R-SYNC" level="mandatory" desc="Progress sync recovery when file state is out-of-sync">
+      <rule>When DISPATCH-PROGRESS.json shows many tasks as "pending" but output files already exist in 04.development/ directories, use sync command to recover: node ${update_progress_script} sync --file ${dispatch_progress_file} --dir ${iteration_path}/04.development --suffix "-task.md"</rule>
+      <rule>DO NOT re-dispatch tasks that already have output files on disk</rule>
+      <rule>DO NOT use ad-hoc PowerShell/Bash commands to manually update progress JSON</rule>
+    </block>
   </sequence>
 
   <!-- ========== Phase 0.5: IDE Directory Detection ========== -->

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "speccrew",
-  "version": "0.7.36",
+  "version": "0.7.38",
   "description": "Spec-Driven Development toolkit for AI-powered IDEs",
   "author": "charlesmu99",
   "repository": {

package/workspace-template/scripts/update-progress.js

@@ -137,7 +137,7 @@ function outputError(error) {
  */
 function acquireLock(filePath) {
     const lockPath = `${filePath}.lock`;
-    const maxRetries = 30;
+    const maxRetries = 50;
     let retryCount = 0;
 
     while (retryCount < maxRetries) {
@@ -152,7 +152,7 @@ function acquireLock(filePath) {
                 try {
                     const lockStat = fs.statSync(lockPath);
                     const ageSeconds = (Date.now() - lockStat.mtimeMs) / 1000;
-                    if (ageSeconds > 60) {
+                    if (ageSeconds > 120) {
                         console.error(`Warning: Stale lock file detected (age: ${Math.round(ageSeconds)}s), removing: ${lockPath}`);
                         fs.unlinkSync(lockPath);
                         // Do not consume retry count, continue to next loop attempt to acquire lock
@@ -166,9 +166,10 @@ function acquireLock(filePath) {
             if (retryCount >= maxRetries) {
                 throw new Error(`Failed to acquire file lock for '${filePath}' after ${maxRetries} attempts`);
             }
-            // Wait 1 second before retry
+            // Retry with jitter to avoid thundering herd
+            const delay = 200 + Math.floor(Math.random() * 300);
             const start = Date.now();
-            while (Date.now() - start < 1000) {
+            while (Date.now() - start < delay) {
                 // Busy wait
             }
         }
@@ -209,7 +210,13 @@ function readJsonFile(filePath) {
     if (!fs.existsSync(filePath)) {
         throw new Error(`File not found: ${filePath}`);
     }
-    const content = fs.readFileSync(filePath, 'utf8');
+    let content = fs.readFileSync(filePath, 'utf8');
+    
+    // Remove UTF-8 BOM if present
+    if (content.charCodeAt(0) === 0xFEFF) {
+        content = content.slice(1);
+    }
+    
     try {
         return JSON.parse(content);
     } catch (e) {
@@ -569,7 +576,7 @@ function cmdRead(args) {
 
         // 5. --status mode: filter tasks by status
         if (args.status) {
-            const validStatuses = ['pending', 'in_progress', 'partial', 'completed', 'failed'];
+            const validStatuses = ['pending', 'in_progress', 'partial', 'completed', 'failed', 'confirmed'];
             if (!validStatuses.includes(args.status)) {
                 outputError(`Invalid status filter: ${args.status}. Must be one of: ${validStatuses.join(', ')}`);
             }
@@ -597,7 +604,7 @@ function cmdUpdateTask(args) {
         outputError('Usage: update-task --file <path> --task-id <id> --status <status> [options]');
     }
 
-    const validStatuses = ['pending', 'in_progress', 'partial', 'completed', 'failed'];
+    const validStatuses = ['pending', 'in_progress', 'partial', 'completed', 'failed', 'confirmed'];
     if (!validStatuses.includes(args.status)) {
         outputError(`Invalid status: ${args.status}. Must be one of: ${validStatuses.join(', ')}`);
     }
@@ -638,6 +645,8 @@ function cmdUpdateTask(args) {
             if (args.output) {
                 task.output = args.output;
             }
+        } else if (args.status === 'confirmed') {
+            task.confirmed_at = now;
         } else if (args.status === 'failed') {
             task.completed_at = now;
             if (args.error) {