@namch/agent-assistant 1.0.1 → 1.0.3

package/commands/cook/hard.md

@@ -134,6 +134,65 @@ One phase at a time, each phase independent: Phase 1 → then Phase 2 → … in
 
 ---
 
+## 🛡️ VERIFICATION CHECKPOINT — Context Optimization
+
+> **PURPOSE**: Prevent "context rot" by clearing noisy planning history before implementation.
+>
+> Long-running planning sessions fill context with noise that degrades code generation quality.
+> This checkpoint acts as a "firewall" between Planning and Implementation phases.
+
+### ⚡ OPTIONS (Present to User)
+
+```markdown
+## 🛡️ Context Optimization Checkpoint
+
+**Planning Complete** — Plan file created at: `./reports/plans/PLAN-{feature}.md`
+
+**Choose how to proceed with implementation:**
+
+| Option                           | Action        | Description                                                                          |
+| -------------------------------- | ------------- | ------------------------------------------------------------------------------------ |
+| **1. 🚀 Clear context & Auto-Implement** | `RECOMMENDED` | Fresh start: Reload Plan file, ignore chat history, begin implementation immediately |
+| **2. ⏸️ Clear context & Manual**         | `SAFE`        | Clear context, reload Plan, pause for your command before coding                     |
+| **3. ⚠️ Continue (No Clear)**    | `RISKY`       | Proceed with full history attached (may cause hallucination)                         |
+
+⏳ Awaiting selection...
+```
+
+### 🔄 EXECUTION BEHAVIOR
+
+```yaml
+option_1_clear_auto_implement:
+  behavior: "RECOMMENDED - Simulate fresh start"
+  steps:
+    1. ACKNOWLEDGE: "🚀 Executing Clear context & Auto-Implement..."
+    2. CONTEXT_DIRECTIVE: |
+      ⛔ IGNORE all previous chat messages and reasoning chains.
+      ✅ RELOAD: `./reports/plans/PLAN-{feature}.md` as SOLE SOURCE OF TRUTH.
+      ✅ PROCEED: Begin Phase 6 (Implementation) immediately.
+    3. LOAD: Read Plan file completely
+    4. EXECUTE: Start Implementation phase with fresh context mindset
+
+option_2_clear_manual:
+  behavior: "Clear and wait for explicit command"
+  steps:
+    1. ACKNOWLEDGE: "⏸️ Context cleared. Plan reloaded."
+    2. CONTEXT_DIRECTIVE: |
+      ⛔ IGNORE all previous chat messages and reasoning chains.
+      ✅ RELOAD: `./reports/plans/PLAN-{feature}.md` as SOLE SOURCE OF TRUTH.
+    3. OUTPUT: "Ready for implementation. Type `/continue` or give specific instructions."
+    4. WAIT: For user command before proceeding
+
+option_3_continue_no_clear:
+  behavior: "Proceed with caution - context rot risk"
+  steps:
+    1. WARN: "⚠️ Continuing with full history. Higher hallucination risk."
+    2. PROCEED: Continue to Phase 6 with existing context
+    3. MONITOR: Watch for signs of context confusion
+```
+
+---
+
 ## 🎭 Phase 6: IMPLEMENTATION
 
 | Agent        | `tech-lead` → routes to specialists                     |

package/commands/cook.md

@@ -16,9 +16,10 @@ execution-mode: router
 ## 🛑 PRE-FLIGHT (DO FIRST — BLOCKS EXECUTION)
 
 **LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
-1. ORCHESTRATION-LAWS.md  
-2. ADAPTIVE-EXECUTION.md  
-3. EXECUTION-PROTOCOL.md  
+
+1. ORCHESTRATION-LAWS.md
+2. ADAPTIVE-EXECUTION.md
+3. EXECUTION-PROTOCOL.md
 
 **⛔ Do not run any workflow phase until all are loaded.** Follow **all** rules in those files. Then run this file's ROUTING LOGIC, LOAD the chosen variant workflow, and execute it.
 
@@ -33,6 +34,9 @@ IF feature is simple (clear spec, low complexity):
 IF feature is complex (multi-component, research needed):
   → Route to /cook:hard
 
+IF feature is complex AND context optimization is critical:
+  → Route to /cook:focus
+
 IF unsure:
   → Default to /cook:fast
 ```
@@ -41,10 +45,11 @@ IF unsure:
 
 ## AVAILABLE ROUTES
 
-| Route        | When to Use                          |
-| ------------ | ------------------------------------ |
-| `/cook:fast` | Quick features, clear specifications |
-| `/cook:hard` | Complex features, full workflow      |
+| Route         | When to Use                                             |
+| ------------- | ------------------------------------------------------- |
+| `/cook:fast`  | Quick features, clear specifications                    |
+| `/cook:hard`  | Complex features, full workflow                         |
+| `/cook:focus` | Complex features with **enforced context optimization** |
 
 ---
 
@@ -59,6 +64,7 @@ IF unsure:
 
 1. ⚡ **Fast** → `/cook:fast` — Quick implementation
 2. ⚡⚡⚡ **Hard** → `/cook:hard` — Full feature development
+3. 🎯 **Focus** → `/cook:focus` — Full development with **automatic context clearing** (prevents hallucination)
 
 ⏳ Awaiting selection...
 ```

package/commands/debug/focus.md

@@ -0,0 +1,155 @@
+---
+description: 🔒 Focused Debug — Deep investigation with enforced context optimization
+version: "1.0"
+category: debugging
+execution-mode: execute
+---
+
+# /debug:focus — Focus Deep Investigation
+
+> **MISSION**: Thorough investigation with **enforced context optimization**.
+>
+> This variant automatically clears context before fix handoff—no user prompt required.
+> Use when you want guaranteed clean fix handoff without hypothesis drift.
+
+<issue>$ARGUMENTS</issue>
+
+---
+
+## 🛑 PRE-FLIGHT (DO FIRST — BLOCKS PHASE 1)
+
+**LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
+
+1. ORCHESTRATION-LAWS.md
+2. ADAPTIVE-EXECUTION.md
+3. EXECUTION-PROTOCOL.md
+
+**⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
+
+---
+
+## 🔀 TIERED EXECUTION
+
+| Tier       | When               | Action                       |
+| ---------- | ------------------ | ---------------------------- |
+| **TIER 1** | runSubagent EXISTS | Invoke sub-agent (MANDATORY) |
+| **TIER 2** | Tool MISSING       | EMBODY agent file (FALLBACK) |
+
+---
+
+## 📁 DELIVERABLE FILES
+
+| Agent    | Output                              |
+| -------- | ----------------------------------- |
+| debugger | `./reports/debugs/DEBUG-{issue}.md` |
+
+All files in `./reports/` → English only.
+
+---
+
+## ⛔ INCREMENTAL EXECUTION (MANDATORY)
+
+One phase at a time, each phase independent: Phase 1 → then Phase 2 → … in one reply. No batching (load only what each phase needs). **Within each phase:** when doing a part, output it in format so user sees what's happening (announce before doing).
+
+---
+
+## 🎭 Phase 1: INFORMATION GATHERING
+
+| Agent | `scouter`                                 |
+| ----- | ----------------------------------------- |
+| Goal  | Gather context and reproduction steps     |
+| Exit  | Context gathered, reproduction documented |
+
+---
+
+## 🎭 Phase 2: HYPOTHESIS FORMATION
+
+| Agent | `debugger`                          |
+| ----- | ----------------------------------- |
+| Goal  | Form and rank hypotheses            |
+| Exit  | Hypotheses documented with evidence |
+
+---
+
+## 🎭 Phase 3: ROOT CAUSE ANALYSIS
+
+| Agent  | `debugger`                                      |
+| ------ | ----------------------------------------------- |
+| Goal   | Deep investigation                              |
+| Output | `./reports/debugs/DEBUG-{issue}.md`             |
+| Exit   | Root cause identified, failure chain documented |
+
+---
+
+## 🎭 Phase 4: SOLUTION DESIGN
+
+| Agent | `planner`                               |
+| ----- | --------------------------------------- |
+| Goal  | Design fix strategy                     |
+| Exit  | Fix approach defined with rollback plan |
+
+---
+
+## 🛡️ STRICT CONTEXT GATE — Automatic Context Optimization
+
+> **⚡ FOCUS MODE**: This checkpoint executes automatically. NO user input required.
+>
+> **PURPOSE**: Prevent "context rot" by forcibly clearing noisy debugging history.
+> This acts as a mandatory "firewall" before handoff to fix workflows.
+
+### 🔒 AUTOMATIC EXECUTION (NO PROMPT)
+
+```yaml
+strict_context_gate:
+  mode: "AUTOMATIC - No user interaction"
+  behavior: "Force Clear context & Fix"
+
+  execution:
+    1. ANNOUNCE: |
+      ## 🛡️ Strict Context Gate — Automatic
+
+      ✅ **Debug Complete**: Root cause identified, fix strategy designed.
+      🔒 **Focus Mode**: Automatically clearing context for fix handoff.
+
+      ⚡ Executing: **Clear context & Fix**...
+
+    2. CONTEXT_DIRECTIVE: |
+      ╔══════════════════════════════════════════════════════════════════╗
+      ║  ⛔ MANDATORY CONTEXT RESET — FOCUS MODE                        ║
+      ╠══════════════════════════════════════════════════════════════════╣
+      ║  IGNORE: All debugging hypotheses, failed investigation paths,  ║
+      ║          rejected theories, and intermediate reasoning.         ║
+      ║                                                                  ║
+      ║  SOLE SOURCE OF TRUTH: Root cause + Fix strategy.               ║
+      ║                                                                  ║
+      ║  HANDOFF: Ready for `/fix:focus` implementation.               ║
+      ║           Fix should treat root cause as fresh context.         ║
+      ╚══════════════════════════════════════════════════════════════════╝
+
+    3. OUTPUT: Root cause + Fix strategy summary
+
+    4. PROCEED: To COMPLETION
+```
+
+### 📋 Post-Gate Status
+
+```markdown
+🔒 **Context Gate Passed**
+
+- Previous context: DISCARDED
+- Active context: Root cause + Fix strategy only
+- Mode: Clean handoff ready
+
+Ready for fix implementation...
+```
+
+---
+
+## COMPLETION
+
+Present findings with:
+
+1. ✅ **Root Cause** — Identified (Focus Mode)
+2. 🔒 **Context Optimized** — Ready for focus fix
+3. 🔧 **Fix** → `/fix:focus`
+4. 📝 **Document** → `/docs:core`

package/commands/debug/hard.md

@@ -87,6 +87,59 @@ One phase at a time, each phase independent: Phase 1 → then Phase 2 → … in
 
 ---
 
+## 🛡️ VERIFICATION CHECKPOINT — Context Optimization
+
+> **PURPOSE**: Prevent "context rot" by clearing noisy debugging/analysis history before fix implementation.
+> 
+> Deep investigation sessions fill context with hypotheses and failed attempts that can confuse implementation.
+> This checkpoint acts as a "firewall" before handing off to fix workflows.
+
+### ⚡ OPTIONS (Present to User)
+
+```markdown
+## 🛡️ Context Optimization Checkpoint
+
+**Debug Analysis Complete** — Root cause identified, fix strategy designed.
+
+**Choose how to proceed:**
+
+| Option | Action | Description |
+|--------|--------|-------------|
+| **1. 🚀 Clear context & Fix** | `RECOMMENDED` | Fresh start: Ignore debug history, proceed with fix strategy only |
+| **2. ⏸️ Review First** | `SAFE` | Clear context, show fix summary, wait for approval |
+| **3. ⚠️ Continue (No Clear)** | `RISKY` | Keep debug history (may cause fix drift) |
+
+⏳ Awaiting selection...
+```
+
+### 🔄 EXECUTION BEHAVIOR
+
+```yaml
+option_1_clear_fix:
+  behavior: "RECOMMENDED - Clean handoff to fix"
+  steps:
+    1. ACKNOWLEDGE: "🚀 Context optimized. Fix strategy ready."
+    2. CONTEXT_DIRECTIVE: |
+       ⛔ IGNORE all debugging hypotheses and failed investigation paths.
+       ✅ FIX STRATEGY is SOLE SOURCE OF TRUTH.
+    3. OUTPUT: "Run `/fix:focus` to implement with clean context."
+
+option_2_review_first:
+  behavior: "Clear and show summary"
+  steps:
+    1. ACKNOWLEDGE: "⏸️ Context cleared."
+    2. OUTPUT: Display root cause + fix strategy summary
+    3. WAIT: For user approval
+
+option_3_continue_no_clear:
+  behavior: "Proceed with caution"
+  steps:
+    1. WARN: "⚠️ Debug history retained. Watch for hypothesis drift."
+    2. PROCEED: Complete workflow with existing context
+```
+
+---
+
 ## COMPLETION
 
 Present findings with:

package/commands/debug.md

@@ -33,6 +33,9 @@ IF issue is simple (clear error, reproducible):
 IF issue is complex (intermittent, unclear cause):
   → Route to /debug:hard
 
+IF issue is complex AND clean fix handoff is critical:
+  → Route to /debug:focus
+
 IF unsure:
   → Default to /debug:fast (escalate if needed)
 ```
@@ -41,10 +44,11 @@ IF unsure:
 
 ## AVAILABLE ROUTES
 
-| Route         | When to Use                           |
-| ------------- | ------------------------------------- |
-| `/debug:fast` | Quick diagnosis for simple bugs       |
-| `/debug:hard` | Deep investigation for complex issues |
+| Route           | When to Use                                          |
+| --------------- | ---------------------------------------------------- |
+| `/debug:fast`   | Quick diagnosis for simple bugs                      |
+| `/debug:hard`   | Deep investigation for complex issues                |
+| `/debug:focus` | Deep investigation with **enforced context optimization** |
 
 ---
 
@@ -59,6 +63,7 @@ IF unsure:
 
 1. ⚡ **Fast** → `/debug:fast` — Quick diagnosis
 2. ⚡⚡⚡ **Hard** → `/debug:hard` — Deep investigation
+3. 🔒 **Focus** → `/debug:focus` — Deep investigation with **automatic context clearing** (clean fix handoff)
 
 ⏳ Awaiting selection...
 ```

package/commands/design/focus.md

@@ -0,0 +1,267 @@
+---
+description: 🔒 Focused Design — Complete design process with enforced context optimization
+version: "1.0"
+category: design
+execution-mode: execute
+---
+
+# /design:focus — Focus Full Design Process
+
+> **MISSION**: Execute complete design process with **enforced context optimization**.
+> 
+> This variant automatically clears context before implementation handoff—no user prompt required.
+> Use when you want guaranteed clean implementation handoff without design exploration noise.
+
+<request>$ARGUMENTS</request>
+
+---
+
+## 🛑 PRE-FLIGHT (DO FIRST — BLOCKS PHASE 1)
+
+**LOAD now** (in order; path `./rules/` or `~/.{TOOL}/skills/agent-assistant/rules/`):
+
+1. ORCHESTRATION-LAWS.md
+2. ADAPTIVE-EXECUTION.md
+3. EXECUTION-PROTOCOL.md
+
+**⛔ Do not run Phase 1 until all are loaded.** Follow **all** rules in those files; they override any conflicting instructions in this file.
+
+---
+
+## 🔀 TIERED EXECUTION PROTOCOL (MANDATORY)
+
+> **Reference**: `{RULES_PATH}/ADAPTIVE-EXECUTION.md`
+
+```yaml
+tiered_execution:
+  principle: "Sub-agent FIRST (Tier 1). EMBODY ONLY on system failure (Tier 2)."
+  for_each_phase:
+    TIER_1_MANDATORY: "IF tool exists → MUST use SUB_AGENT_DELEGATION"
+    TIER_2_FALLBACK: "ONLY on system error—NOT complexity/preference/speed"
+  anti_lazy_fallback:
+    - ❌ NEVER use Tier 2 when Tier 1 tool is available
+    - ✅ ALWAYS attempt Tier 1 first when tool exists
+```
+
+---
+
+## ⚠️ CRITICAL: DELIVERABLE FILE RULES
+
+```yaml
+deliverable_files:
+  researcher: "./reports/researchers/RESEARCH-{request}.md"
+  scouter: "./reports/scouts/SCOUT-{request}.md"
+  designer: "./reports/designs/DESIGN-{request}.md" # MANDATORY OUTPUT
+
+enforcement:
+  - Design phase MUST create design file
+  - Design file is the deliverable for implementation phases
+```
+
+All files in `./reports/` → English only.
+
+---
+
+## ⛔ INCREMENTAL EXECUTION (MANDATORY)
+
+One phase at a time, each phase independent: Phase 1 → then Phase 2 → … in one reply. No batching (load only what each phase needs). **Within each phase:** when doing a part, output it in format so user sees what's happening (announce before doing).
+
+---
+
+## 🎭 Phase 1: REQUIREMENTS DISCOVERY
+
+| Attribute | Value                       |
+| --------- | --------------------------- |
+| **Agent** | `brainstormer`              |
+| **Goal**  | Clarify design requirements |
+
+### ⚡ TIERED EXECUTION
+
+**TIER 1 (MANDATORY when tool exists):**
+
+> Invoke runSubagent for `brainstormer`. Context: ISOLATED.
+
+**TIER 2 (FALLBACK on system error only):**
+
+> Load `{AGENTS_PATH}/brainstormer.md`
+> EMBODY [brainstormer] — Requires logged system error justification.
+
+**Exit Criteria:**
+
+- [ ] Requirements clear
+- [ ] User needs identified
+- [ ] Constraints documented
+- [ ] **METHODOLOGY CHECK**: Output aligns with `brainstormer` Thinking Protocol
+
+---
+
+## 🎭 Phase 2: RESEARCH
+
+| Attribute | Value                           |
+| --------- | ------------------------------- |
+| **Agent** | `researcher`                    |
+| **Goal**  | Research design patterns and UX |
+
+### ⚡ TIERED EXECUTION
+
+**TIER 1 (MANDATORY when tool exists):**
+
+> Invoke runSubagent for `researcher`. Context: ISOLATED.
+
+**TIER 2 (FALLBACK on system error only):**
+
+> Load `{AGENTS_PATH}/researcher.md`
+> EMBODY [researcher] — Requires logged system error justification.
+
+**Exit Criteria:**
+
+- [ ] Patterns researched
+- [ ] Best practices identified
+- [ ] **METHODOLOGY CHECK**: Output aligns with `researcher` Thinking Protocol
+
+---
+
+## 🎭 Phase 3: CODEBASE ANALYSIS
+
+| Attribute | Value                      |
+| --------- | -------------------------- |
+| **Agent** | `scouter`                  |
+| **Goal**  | Map existing design system |
+
+### ⚡ TIERED EXECUTION
+
+**TIER 1 (MANDATORY when tool exists):**
+
+> Invoke runSubagent for `scouter`. Context: ISOLATED.
+
+**TIER 2 (FALLBACK on system error only):**
+
+> Load `{AGENTS_PATH}/scouter.md`
+> EMBODY [scouter] — Requires logged system error justification.
+
+**Exit Criteria:**
+
+- [ ] Design system documented
+- [ ] Component inventory
+- [ ] Integration points
+- [ ] **METHODOLOGY CHECK**: Output aligns with `scouter` Thinking Protocol
+
+---
+
+## 🎭 Phase 4: DESIGN CREATION
+
+| Attribute | Value                  |
+| --------- | ---------------------- |
+| **Agent** | `designer`             |
+| **Goal**  | Full design with specs |
+
+### ⚡ TIERED EXECUTION
+
+**TIER 1 (MANDATORY when tool exists):**
+
+> Invoke runSubagent for `designer`. Context: ISOLATED.
+
+**TIER 2 (FALLBACK on system error only):**
+
+> Load `{AGENTS_PATH}/designer.md`
+> EMBODY [designer] — Requires logged system error justification.
+
+**Exit Criteria:**
+
+- [ ] Design complete
+- [ ] All states covered
+- [ ] Accessibility verified
+- [ ] Specs documented
+- [ ] **METHODOLOGY CHECK**: Output aligns with `designer` Thinking Protocol
+
+---
+
+## 🛡️ STRICT CONTEXT GATE — Automatic Context Optimization
+
+> **⚡ FOCUS MODE**: This checkpoint executes automatically. NO user input required.
+> 
+> **PURPOSE**: Prevent "context rot" by forcibly clearing noisy design exploration history.
+> This acts as a mandatory "firewall" before review and implementation handoff.
+
+### 🔒 AUTOMATIC EXECUTION (NO PROMPT)
+
+```yaml
+strict_context_gate:
+  mode: "AUTOMATIC - No user interaction"
+  behavior: "Force Clear context & Ready"
+  
+  execution:
+    1. ANNOUNCE: |
+       ## 🛡️ Strict Context Gate — Automatic
+       
+       ✅ **Design Complete**: `./reports/designs/DESIGN-{request}.md`
+       🔒 **Focus Mode**: Automatically clearing context for review/implementation.
+       
+       ⚡ Executing: **Clear context & Ready**...
+    
+    2. CONTEXT_DIRECTIVE: |
+       ╔══════════════════════════════════════════════════════════════════╗
+       ║  ⛔ MANDATORY CONTEXT RESET — FOCUS  MODE                        ║
+       ╠══════════════════════════════════════════════════════════════════╣
+       ║  IGNORE: All design exploration, rejected alternatives,          ║
+       ║          iteration history, and intermediate reasoning.          ║
+       ║                                                                  ║
+       ║  SOLE SOURCE OF TRUTH: Design file in ./reports/designs/         ║
+       ║                                                                  ║
+       ║  PROCEED: Continue to Design Review with clean context.          ║
+       ║           Then handoff to implementation.                        ║
+       ╚══════════════════════════════════════════════════════════════════╝
+    
+    3. LOAD: Read Design file as if seeing it for the first time
+    
+    4. PROCEED: Continue to Phase 5 (Design Review)
+```
+
+### 📋 Post-Gate Status
+
+```markdown
+🔒 **Context Gate Passed**
+- Previous context: DISCARDED
+- Active context: Design file only
+- Mode: Clean review start
+
+Proceeding to Design Review...
+```
+
+---
+
+## 🎭 Phase 5: DESIGN REVIEW
+
+| Attribute | Value                 |
+| --------- | --------------------- |
+| **Agent** | `reviewer`            |
+| **Goal**  | Review design quality |
+
+### ⚡ TIERED EXECUTION
+
+**TIER 1 (MANDATORY when tool exists):**
+
+> Invoke runSubagent for `reviewer`. Context: ISOLATED.
+
+**TIER 2 (FALLBACK on system error only):**
+
+> Load `{AGENTS_PATH}/reviewer.md`
+> EMBODY [reviewer] — Requires logged system error justification.
+
+**Exit Criteria:**
+
+- [ ] Design reviewed
+- [ ] Standards met
+- [ ] Approved
+- [ ] **METHODOLOGY CHECK**: Output aligns with `reviewer` Thinking Protocol
+
+---
+
+## COMPLETION
+
+Present design with:
+
+1. ✅ **Done** — Design approved (Focus Mode)
+2. 🔒 **Context Optimized** — Ready for focus implementation
+3. 💻 **Implement** → `/code:focus`
+4. 🔄 **Iterate** — Further refinement

package/commands/design/hard.md

@@ -173,6 +173,59 @@ One phase at a time, each phase independent: Phase 1 → then Phase 2 → … in
 
 ---
 
+## 🛡️ VERIFICATION CHECKPOINT — Context Optimization
+
+> **PURPOSE**: Prevent "context rot" by clearing noisy design exploration history before implementation.
+> 
+> Design iterations and rejected alternatives fill context with noise.
+> This checkpoint acts as a "firewall" before handing off to implementation.
+
+### ⚡ OPTIONS (Present to User)
+
+```markdown
+## 🛡️ Context Optimization Checkpoint
+
+**Design Complete** — Design specs created at: `./reports/designs/DESIGN-{request}.md`
+
+**Choose how to proceed:**
+
+| Option | Action | Description |
+|--------|--------|-------------|
+| **1. 🚀 Clear context & Ready** | `RECOMMENDED` | Fresh start for implementation: Reload Design, ignore exploration history |
+| **2. ⏸️ Review First** | `SAFE` | Clear context, show design summary, wait for approval |
+| **3. ⚠️ Continue (No Clear)** | `RISKY` | Keep design history (may affect review quality) |
+
+⏳ Awaiting selection...
+```
+
+### 🔄 EXECUTION BEHAVIOR
+
+```yaml
+option_1_clear_ready:
+  behavior: "RECOMMENDED - Clean handoff"
+  steps:
+    1. ACKNOWLEDGE: "🚀 Context optimized. Design ready for review/implementation."
+    2. CONTEXT_DIRECTIVE: |
+       ⛔ IGNORE all design exploration and rejected alternatives.
+       ✅ DESIGN FILE is SOLE SOURCE OF TRUTH.
+    3. PROCEED: Continue to Design Review phase
+
+option_2_review_first:
+  behavior: "Clear and show summary"
+  steps:
+    1. ACKNOWLEDGE: "⏸️ Context cleared."
+    2. OUTPUT: Display design summary for user review
+    3. WAIT: For user approval
+
+option_3_continue_no_clear:
+  behavior: "Proceed with caution"
+  steps:
+    1. WARN: "⚠️ Design exploration history retained."
+    2. PROCEED: Continue to review with existing context
+```
+
+---
+
 ## 🎭 Phase 5: DESIGN REVIEW
 
 | Attribute | Value                 |