diff --git a/.claude/agents/cli-execution-agent.md b/.claude/agents/cli-execution-agent.md
index c4de2dd2..4c45a3c0 100644
--- a/.claude/agents/cli-execution-agent.md
+++ b/.claude/agents/cli-execution-agent.md
@@ -172,10 +172,18 @@ score = 0
 "analyze" → "Code understanding and pattern identification"
 ```
 
+
 **2. Context Assembly**:
 ```bash
+# Default: comprehensive context
+CONTEXT: @**/*
+
+# Or specific patterns
 CONTEXT: @CLAUDE.md @{discovered_file1} @{discovered_file2} ...
 
+# Cross-directory references (requires --include-directories)
+CONTEXT: @**/* @../shared/**/* @../types/**/*
+
 ## Discovered Context
 - **Project Structure**: {module_summary}
 - **Relevant Files**: {top_files_with_scores}
@@ -187,6 +195,12 @@ CONTEXT: @CLAUDE.md @{discovered_file1} @{discovered_file2} ...
 {optional_best_practices_from_exa}
 ```
 
+**Context Pattern Guidelines**:
+- **Default**: Use `@**/*` for comprehensive context
+- **Specific files**: `@src/**/*` or `@*.ts @*.tsx`
+- **With docs**: `@CLAUDE.md @**/*CLAUDE.md`
+- **Cross-directory**: Must use `--include-directories` parameter (see Command Construction)
+
 **3. Template Selection**:
 ```
 intent=analyze → ~/.claude/workflows/cli-templates/prompts/analysis/pattern.txt
@@ -194,6 +208,14 @@ intent=execute + complex → ~/.claude/workflows/cli-templates/prompts/developme
 intent=plan → ~/.claude/workflows/cli-templates/prompts/planning/task-breakdown.txt
 ```
 
+**3a. RULES Field Guidelines**:
+
+When using `$(cat ...)` for template loading:
+- **Template reference only**: Use `$(cat ...)` directly, do NOT read template content first
+- **NEVER use escape characters**: `\$`, `\"`, `\'` will break command substitution
+- **Correct**: `RULES: $(cat ~/.claude/workflows/cli-templates/prompts/analysis/pattern.txt)`
+- **Wrong**: `RULES: \$(cat ...)` or `RULES: $(cat \"...\")`
+
 **4. Structured Prompt**:
 ```bash
 PURPOSE: {enhanced_intent}
@@ -234,36 +256,91 @@ ELSE IF intent = 'discuss':
 # User --tool flag overrides auto-selection
 ```
 
+### Model Selection
+
+**Gemini Models**:
+- `gemini-2.5-pro` - Analysis tasks (default)
+- `gemini-2.5-flash` - Documentation updates
+
+**Qwen Models**:
+- `coder-model` - Code analysis (default, -m optional)
+- `vision-model` - Image analysis (rare usage)
+
+**Codex Models**:
+- `gpt-5` - Analysis & execution (default)
+- `gpt5-codex` - Large context tasks
+
+**Parameter Position**: `-m` must be placed AFTER prompt string
+
 ### Command Construction
 
 **Gemini/Qwen (Analysis Mode)**:
 ```bash
-cd {directory} && ~/.claude/scripts/{tool}-wrapper -p "
+# Use 'gemini' (primary) or 'qwen' (fallback)
+cd {directory} && gemini -p "
 {enhanced_prompt}
 "
+
+# With model selection (NOTE: -m placed AFTER prompt)
+cd {directory} && gemini -p "{enhanced_prompt}" -m gemini-2.5-pro
+cd {directory} && qwen -p "{enhanced_prompt}"  # coder-model default
 ```
 
 **Gemini/Qwen (Write Mode)**:
 ```bash
-cd {directory} && ~/.claude/scripts/{tool}-wrapper --approval-mode yolo -p "
+# NOTE: --approval-mode yolo must be placed AFTER the prompt
+cd {directory} && gemini -p "
 {enhanced_prompt}
-"
+" -m gemini-2.5-flash --approval-mode yolo
+
+# Fallback to Qwen
+cd {directory} && qwen -p "{enhanced_prompt}" --approval-mode yolo
 ```
 
 **Codex (Auto Mode)**:
 ```bash
+# NOTE: -m, --skip-git-repo-check and -s danger-full-access must be placed at command END
 codex -C {directory} --full-auto exec "
 {enhanced_prompt}
-" --skip-git-repo-check -s danger-full-access
+" -m gpt-5 --skip-git-repo-check -s danger-full-access
 ```
 
 **Codex (Resume for Related Tasks)**:
 ```bash
+# Parameter Position: resume --last must be placed AFTER prompt at command END
 codex --full-auto exec "
 {continuation_prompt}
 " resume --last --skip-git-repo-check -s danger-full-access
 ```
 
+**Cross-Directory Context (Gemini/Qwen)**:
+```bash
+# When CONTEXT references external directories, use --include-directories
+# TWO-STEP REQUIREMENT: 
+# Step 1: Reference in CONTEXT (@../shared/**/*) 
+# Step 2: Add --include-directories parameter
+cd src/auth && gemini -p "
+PURPOSE: {goal}
+CONTEXT: @**/* @../shared/**/* @../types/**/*
+...
+" --include-directories ../shared,../types
+```
+
+### Directory Scope Rules
+
+**Once `cd` to a directory**:
+- **@ references ONLY apply to current directory and its subdirectories**
+- `@**/*` = All files within current directory tree
+- `@*.ts` = TypeScript files in current directory tree
+- `@src/**/*` = Files within src subdirectory (if exists)
+- **CANNOT reference parent or sibling directories via @ alone**
+
+**To reference files outside current directory**:
+- **Step 1**: Add `--include-directories` parameter
+- **Step 2**: Explicitly reference in CONTEXT field with @ patterns
+- **⚠️ BOTH steps are MANDATORY**
+- **Rule**: If CONTEXT contains `@../dir/**/*`, command MUST include `--include-directories ../dir`
+
 ### Timeout Configuration
 
 ```javascript
diff --git a/.claude/commands/memory/docs.md b/.claude/commands/memory/docs.md
index f174a4f2..56f60bce 100644
--- a/.claude/commands/memory/docs.md
+++ b/.claude/commands/memory/docs.md
@@ -364,7 +364,8 @@ bash(
   if [[ "$tool" == "codex" ]]; then
     echo "codex -C \${dir} --full-auto exec \"...\" --skip-git-repo-check -s danger-full-access"
   else
-    echo "bash(cd \${dir} && ~/.claude/scripts/${tool}-wrapper ${approval_flag} -p \"...\")"
+    echo "bash(cd \${dir} && ${tool} ${approval_flag} -p \"...\")"
+    # Direct CLI commands for gemini/qwen
   fi
 )
 ```
diff --git a/.claude/commands/memory/load.md b/.claude/commands/memory/load.md
index 5fd44eaa..9e2024c9 100644
--- a/.claude/commands/memory/load.md
+++ b/.claude/commands/memory/load.md
@@ -136,7 +136,7 @@ Task(
 Execute Gemini/Qwen CLI for deep analysis (saves main thread tokens):
 
 \`\`\`bash
-cd . && ~/.claude/scripts/${tool}-wrapper -p "
+cd . && ${tool} -p "
 PURPOSE: Extract project core context for task: ${task_description}
 TASK: Analyze project architecture, tech stack, key patterns, relevant files
 MODE: analysis
diff --git a/.claude/commands/memory/update-full.md b/.claude/commands/memory/update-full.md
index 991f0991..ef9a03eb 100644
--- a/.claude/commands/memory/update-full.md
+++ b/.claude/commands/memory/update-full.md
@@ -250,7 +250,7 @@ TOOLS (try in order): {{tool_1}}, {{tool_2}}, {{tool_3}}
 
 EXECUTION SCRIPT: ~/.claude/scripts/update_module_claude.sh
   - Accepts strategy parameter: multi-layer | single-layer
-  - Tool execution via CLI wrapper (gemini/qwen/codex)
+  - Tool execution via direct CLI commands (gemini/qwen/codex)
 
 EXECUTION FLOW (for each module):
   1. Tool fallback loop (exit on first success):
diff --git a/.claude/commands/workflow/tdd-verify.md b/.claude/commands/workflow/tdd-verify.md
index 017feafe..2ab06231 100644
--- a/.claude/commands/workflow/tdd-verify.md
+++ b/.claude/commands/workflow/tdd-verify.md
@@ -237,7 +237,7 @@ Final Score: Max(0, Base Score - Deductions)
 
 ### Command Chain
 - **Called After**: `/workflow:execute` (when TDD tasks completed)
-- **Calls**: `/workflow:tools:tdd-coverage-analysis`, Gemini wrapper
+- **Calls**: `/workflow:tools:tdd-coverage-analysis`, Gemini CLI
 - **Related**: `/workflow:tdd-plan`, `/workflow:status`
 
 ### Basic Usage
diff --git a/.claude/commands/workflow/test-cycle-execute.md b/.claude/commands/workflow/test-cycle-execute.md
index eaf1d702..75386255 100644
--- a/.claude/commands/workflow/test-cycle-execute.md
+++ b/.claude/commands/workflow/test-cycle-execute.md
@@ -219,7 +219,7 @@ Iteration N (managed by test-cycle-execute orchestrator):
 #### When Test Failures Occur
 1. **[Orchestrator]** Detects failures from agent output
 2. **[Orchestrator]** Collects failure context from `.process/test-results.json` and logs
-3. **[Orchestrator]** Runs Gemini/Qwen wrapper with failure context
+3. **[Orchestrator]** Runs Gemini/Qwen CLI with failure context
 4. **[CLI Tool]** Analyzes failures and generates fix strategy
 5. **[Orchestrator]** Saves analysis to `.process/iteration-N-analysis.md`
 6. **[Orchestrator]** Generates `IMPL-fix-N.json` with strategy content (not just path)
diff --git a/.claude/commands/workflow/tools/task-generate-tdd.md b/.claude/commands/workflow/tools/task-generate-tdd.md
index 812d1227..1f55fcd4 100644
--- a/.claude/commands/workflow/tools/task-generate-tdd.md
+++ b/.claude/commands/workflow/tools/task-generate-tdd.md
@@ -452,7 +452,7 @@ Update workflow-session.json with TDD metadata:
 
 ### Command Chain
 - **Called By**: `/workflow:tdd-plan` (Phase 4)
-- **Calls**: Gemini wrapper for TDD breakdown
+- **Calls**: Gemini CLI for TDD breakdown
 - **Followed By**: `/workflow:execute`, `/workflow:tdd-verify`
 
 ### Basic Usage
diff --git a/CLAUDE.md b/CLAUDE.md
index f3c1d85f..29133a4f 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -72,6 +72,7 @@ For all CLI tool usage, command syntax, and integration guidelines:
 ## Platform-Specific Guidelines
 
 ### Windows Path Format Guidelines
+- always use complete absolute Windows paths with drive letters and backslashes for ALL file operations
 - **MCP Tools**: Use double backslash `D:\\path\\file.txt` (MCP doesn't support POSIX `/d/path`)
 - **Bash Commands**: Use forward slash `D:/path/file.txt` or POSIX `/d/path/file.txt`
 - **Relative Paths**: No conversion needed `./src`, `../config`