5-phase-workflow 1.1.1 → 1.1.2

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "5-phase-workflow",
-  "version": "1.1.1",
+  "version": "1.1.2",
   "description": "A 5-phase feature development workflow for Claude Code",
   "bin": {
     "5-phase-workflow": "bin/install.js"

package/src/commands/5/configure.md

@@ -19,6 +19,37 @@ After running this command, proceed through the standard phases:
 4. `/5:verify-implementation` - Verify configuration
 5. `/5:review-code` - Review generated files
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND ONLY CREATES THE CONFIGURATION FEATURE SPEC. IT DOES NOT CONFIGURE.**
+
+Your job in this command:
+✅ Analyze project (detect type, build commands, etc.)
+✅ Gather user preferences via questions
+✅ Create feature spec at `.5/CONFIGURE/feature.md`
+✅ Tell user to run /5:plan-implementation CONFIGURE
+
+Your job is NOT:
+❌ Write config.json directly (Phase 3 does this)
+❌ Create CLAUDE.md directly (Phase 3 does this)
+❌ Generate skills directly (Phase 3 does this)
+❌ Skip user interaction
+❌ Assume project structure
+
+**After creating the feature spec and informing the user, YOUR JOB IS COMPLETE. EXIT IMMEDIATELY.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Write config.json** - That's Phase 3's job (via configure-project skill)
+- ❌ **Create CLAUDE.md** - That's Phase 3's job
+- ❌ **Generate skills** - That's Phase 3's job
+- ❌ **Skip questions** - Always confirm detected values with user
+- ❌ **Assume structure** - Detect or ask, don't guess
+
+**If you find yourself creating config files or CLAUDE.md, STOP IMMEDIATELY. You should only be creating the feature spec.**
+
 ## Configuration Process
 
 ### Step 1: Analyze Project (auto-detect, no user interaction)

package/src/commands/5/discuss-feature.md

@@ -19,6 +19,41 @@ This skill is part of **Phase 1** (Feature Planning) of the 5-phase workflow:
 
 This skill enables **optional iterative refinement** of feature specs after initial planning through discussion, clarification, and requirement changes. Use it when the initial spec needs adjustments before proceeding to implementation planning.
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND ONLY UPDATES FEATURE SPECIFICATIONS. IT DOES NOT PLAN OR IMPLEMENT.**
+
+Your job in this command:
+✅ Read existing feature specification
+✅ Ask what user wants to discuss
+✅ Explore codebase if needed for context
+✅ Ask clarifying questions
+✅ Update feature specification
+✅ Tell user to run /5:plan-implementation
+
+Your job is NOT:
+❌ Create new feature specs (use /5:plan-feature)
+❌ Create implementation plans
+❌ Map to technical components
+❌ Start implementation
+❌ Write any code
+❌ Rewrite entire feature spec (only update changed sections)
+
+**After updating the feature spec and informing the user, YOUR JOB IS COMPLETE. EXIT IMMEDIATELY.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Create new feature specs** - That's /5:plan-feature
+- ❌ **Create implementation plans** - That's Phase 2 (/5:plan-implementation)
+- ❌ **Start implementation** - That's Phase 3 (/5:implement-feature)
+- ❌ **Write any code** - This is planning only
+- ❌ **Rewrite entire spec** - Only update sections that changed
+- ❌ **Delete previous Q&A** - Append new discussions, keep history
+
+**If you find yourself creating plans or writing code, STOP IMMEDIATELY. You are exceeding this command's scope.**
+
 ## Use Cases
 
 Use this skill when:
@@ -196,15 +231,19 @@ After updating the spec, tell the developer:
 
 ## Instructions Summary
 
+Follow these steps **IN ORDER** and **STOP after step 9**:
+
 1. **Extract feature name** - From user input or by matching ticket ID
 2. **Read feature spec** - Load current state from `.5/{feature-name}/feature.md`
 3. **Ask initial question** - What do they want to discuss?
 4. **Explore if needed** - Understand codebase context
 5. **Interactive Q&A** - Multiple rounds of clarifying questions
-6. **Iterate** - Allow continued discussion
-7. **Update feature spec** - Modify relevant sections
+6. **Iterate** - Allow continued discussion until user is satisfied
+7. **Update feature spec** - Modify only the relevant sections that changed
 8. **Track changes** - Document discussion history
-9. **Inform developer** - Summarize changes and next steps
+9. **Inform developer** - Summarize changes and ask: "Discuss more or run `/clear` followed by `/5:plan-implementation {feature-name}`?"
+
+**🛑 STOP HERE. YOUR JOB IS COMPLETE. DO NOT CREATE IMPLEMENTATION PLANS.**
 
 ## Key Principles
 
@@ -216,14 +255,6 @@ After updating the spec, tell the developer:
 6. **Preserve history** - Keep previous Q&A, add new discussions
 7. **Clear next steps** - Guide user on what to do next
 
-## DO NOT in This Skill
-
-- DO NOT rewrite the entire feature spec (only update changed sections)
-- DO NOT skip asking what the user wants to discuss
-- DO NOT create new feature specs (use /plan-feature for that)
-- DO NOT proceed to technical planning (that's Phase 2's job)
-- DO NOT start implementation
-- DO NOT delete previous Q&A sections (append to them)
 
 ## Discussion Patterns
 

package/src/commands/5/implement-feature.md

@@ -31,6 +31,43 @@ This command is a **thin orchestrator** that:
 - The plan contains self-contained prompts - agents execute without codebase exploration
 - Skills are called by agents when specified in the plan
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND ORCHESTRATES IMPLEMENTATION. IT DELEGATES TO AGENTS.**
+
+Your job in this phase:
+✅ Load implementation plan
+✅ Initialize state tracking
+✅ Spawn agents (step-executor, step-verifier, step-fixer)
+✅ Process agent results
+✅ Update state file after each step
+✅ Report completion
+✅ Tell user to run /5:verify-implementation
+
+Your job is NOT:
+❌ Execute skills directly (agents call skills)
+❌ Do heavy file reading/writing (agents do this)
+❌ Write code files directly (agents do this)
+❌ Skip state file updates
+❌ Analyze or fix errors yourself (delegate to step-fixer agent)
+❌ Continue past 80% context usage
+
+**You are a THIN ORCHESTRATOR. Delegate all heavy work to agents in forked contexts.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Execute skills directly** - Agents call skills, not this command
+- ❌ **Write code directly** - Agents handle all file operations
+- ❌ **Analyze errors yourself** - Delegate to step-fixer agent (sonnet)
+- ❌ **Skip state file updates** - State updates are MANDATORY after each step
+- ❌ **Skip verification** - step-verifier must run after each step-executor
+- ❌ **Continue after 2 failed retries** - Escalate to user
+- ❌ **Ignore context warnings** - Stop at 80% usage
+
+**If you find yourself writing code or fixing errors directly, STOP. You should be spawning agents instead.**
+
 ## Prerequisites
 
 Before using this skill, ensure:
@@ -367,17 +404,25 @@ Update state file:
 4. **Verify** the update by reading the file again
 
 Tell the developer:
-1. "Feature implementation complete!"
-2. "All {N} components created successfully"
-3. "Compilation: Successful"
-4. "Tests: All passing ({N} tests)"
-5. **"State file: `.5/{feature-name}/state.json`"** (this is critical for resume capability)
-6. "Next steps:"
-   - "Run `/clear` to reset context"
-   - "Then run `/5:verify-implementation {feature-name}` to validate completeness"
+
+```
+✅ Feature implementation complete!
+
+Summary:
+- All {N} components created successfully
+- Compilation: Successful
+- Tests: All passing ({N} tests)
+- State file: `.5/{feature-name}/state.json`
+
+Next steps:
+1. Run `/clear` to reset context
+2. Run `/5:verify-implementation {feature-name}` to validate completeness
+```
 
 **Note:** The verification step will automatically prompt the developer to commit changes, which is recommended before running CodeRabbit review.
 
+**🛑 YOUR JOB IS COMPLETE. DO NOT START VERIFICATION. Wait for user to proceed to Phase 4.**
+
 ## State File Schema
 
 ```typescript
@@ -467,25 +512,29 @@ User: /implement-feature PROJ-1234-add-emergency-schedule
 
 ## Instructions Summary
 
+Follow these steps **IN ORDER**:
+
 1. **Load implementation plan metadata** from `.5/{feature-name}/plan/meta.md` - parse YAML frontmatter for total_steps, total_components
 2. **Initialize state file** (MANDATORY) in `.5/{feature-name}/state.json` - verify creation
 3. **Create tasks** for all steps defined in the plan
 4. **For each step (1 to total_steps):**
    - Load and parse step file: `.5/{feature-name}/plan/step-{N}.md` (parse YAML frontmatter + components block)
    - Build step block object from parsed data
-   - Spawn step-executor with step block
+   - Spawn step-executor with step block (haiku model)
    - Process results
    - Spawn step-verifier
    - Process results
    - **Update state file** (MANDATORY - Step 4f) - verify update
 5. **For final integration step (if configured):** Spawn integration-agent, process results, **update state file** (MANDATORY)
-6. **Handle failures** - record in state file (MANDATORY), extract original prompt from plan/step-{N}.md, spawn step-fixer to diagnose and fix, re-verify after fix, escalate if stuck
+6. **Handle failures** - record in state file (MANDATORY), extract original prompt from plan/step-{N}.md, spawn step-fixer (sonnet) to diagnose and fix, re-verify after fix, escalate if stuck
 7. **Monitor context** - warn at 50%, stop at 80%
 8. **Update state file to completed** (MANDATORY - Step 8) - verify final update
-9. **Report completion** with summary including state file location
+9. **Report completion** - Tell user: "Run `/clear` followed by `/5:verify-implementation {feature-name}`"
 
 **CRITICAL**: State file updates at Steps 2, 4f (after each step), 5 (failures), and 8 (completion) are MANDATORY. These enable resumability if implementation is interrupted.
 
+**🛑 AFTER REPORTING COMPLETION, YOUR JOB IS DONE. DO NOT START VERIFICATION.**
+
 ## Key Principles
 
 1. **Thin orchestrator** - Main context only reads plans, spawns agents, processes results, updates state
@@ -524,18 +573,6 @@ User: "Continue implementation of PROJ-1234-add-emergency-schedule"
 [Execute Step 2 → remaining steps]
 ```
 
-## DO NOT
-
-- DO NOT execute skills directly from this command (agents call skills)
-- DO NOT do heavy file reading/writing in main context (agents do this)
-- **DO NOT skip state file updates** (this breaks resumability)
-- **DO NOT skip state file initialization** (Step 2 is mandatory)
-- **DO NOT skip state file completion update** (Step 8 is mandatory)
-- DO NOT skip verification (step-verifier) after any step
-- DO NOT continue after 2 failed retry attempts without escalation
-- DO NOT ignore context usage warnings
-- DO NOT analyze errors in main context (delegate to step-fixer)
-- DO NOT fix code in main context (delegate to step-fixer)
 
 ## Related Documentation
 

package/src/commands/5/plan-feature.md

@@ -19,6 +19,41 @@ This skill is the **first phase** of the 5-phase workflow:
 
 This skill guides intensive collaboration with the developer to understand requirements, challenge assumptions, and create a comprehensive feature specification.
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND ONLY CREATES THE FEATURE SPECIFICATION. IT DOES NOT IMPLEMENT.**
+
+Your job in this phase:
+✅ Ask questions
+✅ Explore codebase
+✅ Create feature.md file
+✅ Tell user to run /5:plan-implementation
+
+Your job is NOT:
+❌ Create implementation plans
+❌ Map to technical components
+❌ Write any code
+❌ Start implementation
+
+**After creating feature.md and informing the user, YOUR JOB IS COMPLETE. EXIT IMMEDIATELY.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Create implementation plans** - That's Phase 2 (`/5:plan-implementation`)
+- ❌ **Map components to skills** - That's Phase 2's job
+- ❌ **Start any implementation** - That's Phase 3 (`/5:implement-feature`)
+- ❌ **Write any code** - Not even example code
+- ❌ **Create TodoWrite task lists** - Wrong phase
+- ❌ Ask for ticket ID (extract from branch name automatically)
+- ❌ Skip asking for feature description first
+- ❌ Ask questions before exploring the codebase
+- ❌ Skip the intensive Q&A phase
+- ❌ Accept vague requirements
+
+**If you find yourself doing any of the above, STOP IMMEDIATELY. You are exceeding this command's scope.**
+
 ## Planning Process
 
 ### Step 1: Gather Feature Description
@@ -129,7 +164,9 @@ Based on the feature description and discussion:
 
 ### Step 6: Create Feature Specification
 
-Write a comprehensive feature spec to `.5/{TICKET-ID}-{description}/feature.md` with the following structure:
+Write a comprehensive feature spec to `.5/{TICKET-ID}-{description}/feature.md` with the following structure.
+
+**THIS IS YOUR FINAL OUTPUT. After creating this file, proceed immediately to Step 7.**
 
 ```markdown
 # Feature: {TICKET-ID} - {Title}
@@ -219,6 +256,8 @@ After approval:
 
 ## Instructions
 
+Follow these steps **IN ORDER** and **STOP after step 8**:
+
 1. **Ask for feature description** - Request task description and additional information from developer
 2. **Extract Ticket ID** - Get current branch name and extract ticket ID using configured pattern
 3. **Explore the codebase** - Understand existing patterns and affected modules
@@ -226,7 +265,9 @@ After approval:
 5. **Challenge assumptions** - Present alternatives, question complexity
 6. **Determine feature name** - Create short kebab-case description
 7. **Create feature specification** in `.5/{TICKET-ID}-{description}/feature.md`
-8. **Inform the developer** to review the spec and then run `/plan-implementation {TICKET-ID}-{description}`
+8. **Inform the developer** to review the spec and then run `/clear` followed by `/5:plan-implementation {TICKET-ID}-{description}`
+
+**🛑 STOP HERE. YOUR JOB IS COMPLETE. DO NOT PROCEED TO IMPLEMENTATION.**
 
 ## Key Principles
 
@@ -239,16 +280,6 @@ After approval:
 7. **Structured output** - Use the feature spec template consistently
 8. **Clear handoff** - Tell developer what to do next
 
-## DO NOT in This Skill
-
-- DO NOT ask for ticket ID (extract from branch name automatically)
-- DO NOT skip asking for feature description first
-- DO NOT ask questions before exploring the codebase
-- DO NOT create TodoWrite task lists (that's Phase 2's job)
-- DO NOT map components to skills (that's Phase 2's job)
-- DO NOT start implementation (that's Phase 3's job)
-- DO NOT skip the intensive Q&A phase
-- DO NOT accept vague requirements
 
 ## Common Feature Types
 
@@ -279,8 +310,14 @@ After approval:
 7. Skill challenges: "Could we reuse existing scheduling infrastructure instead of creating new one?"
 8. Skill determines: Feature name `add-emergency-schedule`
 9. Skill creates: `.5/PROJ-1234-add-emergency-schedule/feature.md`
-10. Skill tells user: "Feature spec created. Please review and then run `/clear` followed by `/5:plan-implementation PROJ-1234-add-emergency-schedule`"
-11. Skill tells user: "If the feature needs refinements, run `/5:discuss-feature PROJ-1234-add-emergency-schedule`"
+10. Skill tells user: "✅ Feature spec created at `.5/PROJ-1234-add-emergency-schedule/feature.md`
+
+    **Next steps:**
+    1. Review the feature spec
+    2. If changes needed: `/5:discuss-feature PROJ-1234-add-emergency-schedule`
+    3. If approved: `/clear` then `/5:plan-implementation PROJ-1234-add-emergency-schedule`"
+
+**🛑 COMMAND COMPLETE. The skill stops here and waits for user to proceed to Phase 2.**
 
 ## Related Documentation
 

package/src/commands/5/plan-implementation.md

@@ -19,6 +19,41 @@ This skill is the **second phase** of the 5-phase workflow:
 
 **Critical design constraint:** The plan must be executable by haiku-model agents without any codebase exploration. This means YOU must do all analysis upfront and embed everything into the plan: exact file paths, exact code patterns, reference snippets, imports, and complete instructions.
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND ONLY CREATES THE IMPLEMENTATION PLAN. IT DOES NOT IMPLEMENT.**
+
+Your job in this phase:
+✅ Read feature specification
+✅ Analyze codebase deeply
+✅ Ask technical questions
+✅ Create atomic plan files (meta.md, step-N.md, verification.md)
+✅ Build self-contained prompts for each component
+✅ Tell user to run /5:implement-feature
+
+Your job is NOT:
+❌ Start implementation
+❌ Execute any components
+❌ Create state files
+❌ Write any code files
+❌ Run build or tests
+❌ Create TodoWrite task lists
+
+**After creating the plan files and informing the user, YOUR JOB IS COMPLETE. EXIT IMMEDIATELY.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Start implementation** - That's Phase 3 (`/5:implement-feature`)
+- ❌ **Execute components** - That's Phase 3's job
+- ❌ **Create state files** - That's Phase 3's responsibility
+- ❌ **Write code files** - Only create plan files in `.5/{feature-name}/plan/`
+- ❌ **Run builds or tests** - Phase 3 and 4 handle this
+- ❌ **Create vague prompts** - Every prompt must be self-contained and executable by haiku without exploration
+
+**If you find yourself doing any of the above, STOP IMMEDIATELY. You are exceeding this command's scope.**
+
 ## Prerequisites
 
 Before invoking this skill, ensure:
@@ -386,14 +421,18 @@ prompt: |
 
 ## Instructions Summary
 
+Follow these steps **IN ORDER** and **STOP after step 8**:
+
 1. **Read feature spec** from `.5/{feature-name}/feature.md`
 2. **Deep codebase analysis** - Read actual files, extract patterns, imports, conventions. This is the critical step.
 3. **Ask 3-5 technical questions** - Clarify implementation details
 4. **Map components to skills** - Check `.claude/skills/` for available skills
 5. **Group into dependency steps** - Based on component dependencies
 6. **Build self-contained prompts** - Each prompt must be executable by haiku without exploration
-7. **Write plan** to `.5/{feature-name}/plan.md` in the structured format
-8. **Inform developer** to review and run `/5:implement-feature`
+7. **Write plan** to `.5/{feature-name}/plan/` directory with atomic files (meta.md, step-N.md, verification.md)
+8. **Inform developer** to review and run `/clear` followed by `/5:implement-feature {feature-name}`
+
+**🛑 STOP HERE. YOUR JOB IS COMPLETE. DO NOT START IMPLEMENTATION.**
 
 ## Key Principles
 
@@ -404,15 +443,6 @@ prompt: |
 5. **Dependency-driven steps** - Step grouping from component dependencies
 6. **Verify prompt quality** - Could someone with no codebase knowledge execute it?
 
-## DO NOT
-
-- DO NOT start implementation (that's Phase 3's job)
-- DO NOT create state files (that's `/implement-feature`'s job)
-- DO NOT write vague prompts like "follow existing patterns" without including the pattern
-- DO NOT skip codebase analysis - reading actual files is mandatory
-- DO NOT leave file paths ambiguous
-- DO NOT write prompts that require the executor to read other files to understand what to do
-- DO NOT assume the executor can figure things out - be explicit
 
 ## Example Usage
 
@@ -431,7 +461,19 @@ Skill:
 9. Writes plan/meta.md with metadata and risks
 10. Writes plan/step-1.md, plan/step-2.md, plan/step-3.md with YAML components
 11. Writes plan/verification.md with build/test config
-12. Tells user: "Plan created with 7 components across 3 steps. Atomic plan structure at `.5/PROJ-1234-add-emergency-schedule/plan/`. Review the plan files, then run `/clear` followed by `/5:implement-feature PROJ-1234-add-emergency-schedule`"
+12. Tells user: "✅ Implementation plan created at `.5/PROJ-1234-add-emergency-schedule/plan/`
+
+    **Plan structure:**
+    - 7 components across 3 steps
+    - Atomic plan files: meta.md, step-1.md, step-2.md, step-3.md, verification.md
+    - Self-contained prompts ready for haiku execution
+
+    **Next steps:**
+    1. Review the plan files
+    2. Run `/clear` to reset context
+    3. Run `/5:implement-feature PROJ-1234-add-emergency-schedule`"
+
+**🛑 COMMAND COMPLETE. The command stops here and waits for user to proceed to Phase 3.**
 ```
 
 ## Related Documentation

package/src/commands/5/quick-implement.md

@@ -10,6 +10,43 @@ user-invocable: true
 
 Fast path for small, well-understood tasks (1-5 files). Skips extensive planning phases but preserves state tracking and skill-based implementation.
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND IS FOR SMALL, FOCUSED TASKS ONLY (1-5 FILES).**
+
+Your job in this command:
+✅ Get task description
+✅ Extract ticket ID
+✅ Create quick plan (max 5 components)
+✅ Get user approval on plan
+✅ Initialize state tracking
+✅ Execute implementation
+✅ Run verification
+✅ Report completion
+
+Your job is NOT:
+❌ Handle complex features (6+ files)
+❌ Work across multiple domains
+❌ Skip clarifying questions if unclear
+❌ Skip state file updates
+❌ Create feature spec files (use full workflow)
+❌ Commit changes (user handles this)
+
+**This is a FAST PATH for well-understood, small tasks. For anything complex, use the full workflow.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Use for complex features** - 6+ files or multiple domains → use full workflow
+- ❌ **Skip clarifying questions** - If implementation is unclear, ask first
+- ❌ **Skip state updates** - State file updates are MANDATORY
+- ❌ **Create feature specs** - This is for quick tasks, not full features
+- ❌ **Auto-commit** - User handles commits
+- ❌ **Skip plan approval** - Always show plan and get user approval first
+
+**If the task involves more than 5 files or multiple domains, STOP and recommend the full workflow instead.**
+
 ## Process
 
 ### Step 1: Get Task Description
@@ -224,7 +261,7 @@ Skill tool call:
 Report to user:
 
 ```
-Quick implementation complete!
+✅ Quick implementation complete!
 
 ${TICKET_ID}: ${DESCRIPTION}
 
@@ -238,10 +275,12 @@ Tests: {Success | Skipped}
 State: .5/${feature_name}/state.json
 
 Next steps:
-- Review and commit changes
-- For a new task, run `/clear` before starting
+1. Review and commit changes
+2. For a new task, run `/clear` before starting
 ```
 
+**🛑 YOUR JOB IS COMPLETE. DO NOT START NEW TASKS.**
+
 ## Skill Mappings
 
 Skills are project-specific and should be configured in your project's `.claude/skills/` directory. Common patterns include:
@@ -256,12 +295,3 @@ Skills are project-specific and should be configured in your project's `.claude/
 | Tests | Project-specific test skill |
 | Simple edits | Edit tool directly |
 
-## DO NOT
-
-- DO NOT use for complex features (6+ files, multiple domains)
-- DO NOT skip clarifying questions if implementation is unclear
-- **DO NOT skip state file updates** (breaks resumability)
-- **DO NOT skip state file initialization** (Step 7 is mandatory)
-- **DO NOT skip state file completion update** (Step 10 is mandatory)
-- DO NOT create feature spec files (use full workflow for that)
-- DO NOT commit changes (user handles this)

package/src/commands/5/review-code.md

@@ -30,6 +30,45 @@ This skill automates code review using the CodeRabbit CLI. It supports two workf
 - This command stays in the main context (user interaction, fix application)
 - review-processor agent runs CodeRabbit and categorizes findings (forked context)
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND REVIEWS CODE AND APPLIES USER-APPROVED FIXES ONLY.**
+
+Your job in this phase:
+✅ Check CodeRabbit prerequisites
+✅ Ask user what to review
+✅ Spawn review-processor agent
+✅ Present findings overview to user
+✅ Ask user which fixes to apply
+✅ Apply ONLY user-approved fixes
+✅ Verify changes (compile and test)
+✅ Save review report
+
+Your job is NOT:
+❌ Run CodeRabbit directly (agent does this)
+❌ Parse CodeRabbit output directly (agent does this)
+❌ Apply fixes without user approval
+❌ Auto-apply complex refactoring
+❌ Skip the overview step
+❌ Skip verification after applying fixes
+❌ Proceed if CodeRabbit not installed
+
+**ALWAYS GET USER CONSENT BEFORE APPLYING ANY FIXES. This is a review tool, not an auto-fix tool.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Run CodeRabbit directly** - review-processor agent handles this
+- ❌ **Parse CodeRabbit output directly** - review-processor agent handles this
+- ❌ **Apply fixes without approval** - ALWAYS ask user first
+- ❌ **Auto-apply refactoring** - Even if CodeRabbit suggests it
+- ❌ **Skip the overview** - User must see all findings first
+- ❌ **Skip verification** - Always compile and test after fixes
+- ❌ **Assume user wants all fixes** - Each fix needs approval
+
+**If you find yourself applying fixes without user approval, STOP IMMEDIATELY. User consent is mandatory.**
+
 ## Prerequisites
 
 **Required:**
@@ -451,31 +490,45 @@ For interactive mode only, save the review summary to:
 ## Instructions Summary
 
 ### Interactive Mode (Default)
+
+Follow these steps **IN ORDER**:
+
 1. **Check prerequisites** - CodeRabbit installed and logged in
 2. **Ask what to review** - Staged, unstaged, branch, or specific files
 3. **Ask review mode** - Interactive or save to file
 4. **Spawn review-processor** - Delegate CodeRabbit execution and parsing
 5. **Process results** - Receive categorized findings
-6. **Provide overview** - Show concise summary to user
+6. **Provide overview** - Show concise summary to user (MANDATORY - don't skip)
 7. **Ask user** - Which fixes to apply, how to handle questions
 8. **Apply fixes** - Only user-approved fixes, using Edit tool
 9. **Handle questions** - Ask user for each if requested
 10. **Verify changes** - Compile and test after applying fixes
 11. **Save report** - Store in `.5/{feature-name}/`
 
+**🛑 AFTER SAVING REPORT, YOUR JOB IS COMPLETE.**
+
 ### File-Based Annotation Mode
+
+Follow these steps **IN ORDER**:
+
 1. **Check prerequisites** - CodeRabbit installed and logged in
 2. **Ask what to review** - Staged, unstaged, branch, or specific files
 3. **User selects "Save to file"**
 4. **Spawn review-processor** - Delegate CodeRabbit execution and parsing
 5. **Process results** - Receive categorized findings
 6. **Save findings file** - Store structured findings in `.5/{feature-name}/review-{timestamp}-findings.md`
-7. **User edits file** - Mark findings as [FIX], [SKIP], or [MANUAL] with instructions
-8. **User runs `/review-code apply`** - Apply annotated fixes
-9. **Parse annotations** - Read user's action markers and custom instructions
-10. **Apply marked fixes** - Apply [FIX] and [MANUAL] fixes automatically
-11. **Verify changes** - Compile and test after applying fixes
-12. **Update findings file** - Append application results
+7. **Tell user** - Edit the file and run `/5:review-code apply`
+
+**🛑 STOP HERE if in save-to-file mode. Wait for user to edit and run apply.**
+
+**When user runs `/5:review-code apply`:**
+
+8. **Parse annotations** - Read user's action markers and custom instructions
+9. **Apply marked fixes** - Apply [FIX] and [MANUAL] fixes automatically (user already approved via file annotations)
+10. **Verify changes** - Compile and test after applying fixes
+11. **Update findings file** - Append application results
+
+**🛑 AFTER UPDATING FINDINGS FILE, YOUR JOB IS COMPLETE.**
 
 ## Key Principles
 
@@ -486,16 +539,6 @@ For interactive mode only, save the review summary to:
 5. **Verification** - Always compile and test after applying fixes
 6. **Non-intrusive** - If not installed or logged in, gracefully exit
 
-## DO NOT
-
-- DO NOT run CodeRabbit directly (agent handles this)
-- DO NOT parse CodeRabbit output directly (agent handles this)
-- DO NOT apply fixes without user approval
-- DO NOT auto-apply complex refactoring suggestions
-- DO NOT skip the overview step
-- DO NOT skip verification (compilation/tests)
-- DO NOT proceed if CodeRabbit is not installed
-- DO NOT assume user wants all fixes applied
 
 ## Error Handling
 

package/src/commands/5/verify-implementation.md

@@ -20,6 +20,43 @@ This skill provides comprehensive verification of a completed feature implementa
 - Delegates verification work to the verification-agent (also forked context)
 - Handles reporting, state updates, and user interaction within the forked context
 
+## ⚠️ CRITICAL SCOPE CONSTRAINT
+
+**THIS COMMAND VERIFIES IMPLEMENTATION. IT DELEGATES CHECKS TO AGENT.**
+
+Your job in this phase:
+✅ Load implementation plan
+✅ Spawn verification-agent
+✅ Process verification results
+✅ Save verification report
+✅ Update state file
+✅ Prompt for commit (if passed)
+✅ Tell user to run /5:review-code
+
+Your job is NOT:
+❌ Run compilation directly (agent does this)
+❌ Run tests directly (agent does this)
+❌ Detect problems directly (agent does this)
+❌ Block on warnings (only errors block)
+❌ Create commits without user approval
+❌ Skip asking about commit
+
+**After prompting for commit and informing the user, YOUR JOB IS COMPLETE. EXIT IMMEDIATELY.**
+
+## ❌ Boundaries: What This Command Does NOT Do
+
+**CRITICAL:** This command has a LIMITED scope. Do NOT:
+
+- ❌ **Run compilation directly** - verification-agent handles this
+- ❌ **Run tests directly** - verification-agent handles this
+- ❌ **Detect problems directly** - verification-agent handles this
+- ❌ **Block on warnings** - Only errors block completion
+- ❌ **Auto-commit** - Always ask user first
+- ❌ **Skip state file update** - State update is mandatory
+- ❌ **Start code review** - That's Phase 5 (/5:review-code)
+
+**If you find yourself running builds or tests directly, STOP. You should be spawning the verification-agent instead.**
+
 ## Verification Process
 
 ### Step 1: Load Implementation Plan
@@ -259,16 +296,20 @@ EOF
 
 ## Instructions Summary
 
+Follow these steps **IN ORDER** and **STOP after step 7**:
+
 1. **Load implementation plan** from `.5/{feature-name}/plan/`:
    - Read meta.md for plan metadata
    - Read verification.md for build/test config and targets
    - Aggregate expected files from all step-N.md files
 2. **Spawn verification-agent** with aggregated expected files, modules, and test modules
 3. **Process agent results** - extract status, report, and structured data
-4. **Save verification report** to state directory
+4. **Save verification report** to `.5/{feature-name}/verification.md`
 5. **Update state file** with verification results
 6. **Inform developer** with clear status
-7. **Prompt for commit** (if PASSED or PASSED WITH WARNINGS) - Recommended before CodeRabbit review
+7. **Prompt for commit** (if PASSED or PASSED WITH WARNINGS) and tell user: "Run `/clear` followed by `/5:review-code` for CodeRabbit review"
+
+**🛑 STOP HERE. YOUR JOB IS COMPLETE. DO NOT START CODE REVIEW.**
 
 ## Key Principles
 
@@ -280,17 +321,6 @@ EOF
 6. **Resumable** - Can be re-run after fixes
 7. **Commit ready** - Prompt for commit after successful verification
 
-## DO NOT
-
-- DO NOT run compilation or tests directly (agent handles this)
-- DO NOT run problem detection directly (agent handles this)
-- DO NOT block on warnings (only errors block)
-- DO NOT skip state file update
-- DO NOT skip saving verification report
-- DO NOT skip asking about commit if verification passed
-- DO NOT automatically create commits without user approval
-- DO NOT use emojis in commit messages
-- DO NOT include AI attribution or Co-Authored-By lines in commit messages
 
 ## Error Handling