catlog22/Claude-Code-Workflow
1
0
Fork 0
mirror of https://github.com/catlog22/Claude-Code-Workflow.git synced 2026-02-13 02:41:50 +08:00
Code Issues Packages Projects Releases Wiki Activity

chore: 删除 intelligent-tools-strategy.md 中的参考链接以简化文档

Browse Source
This commit is contained in:
catlog22
2025-10-05 09:08:33 +08:00
parent 3bf15ced59
commit 7d9adf5a55
8 changed files with 484 additions and 289 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
.claude/commands/cli/analyze.md
View File

@@ -17,7 +17,6 @@ allowed-tools: SlashCommand(*), Bash(*), TodoWrite(*), Read(*), Glob(*)
Execute CLI tool analysis on codebase patterns, architecture, or code quality. Execute CLI tool analysis on codebase patterns, architecture, or code quality.
**Supported Tools**: codex, gemini (default), qwen **Supported Tools**: codex, gemini (default), qwen
**Reference**: @~/.claude/workflows/intelligent-tools-strategy.md for complete tool details
## Execution Flow ## Execution Flow

1
.claude/commands/cli/chat.md
View File

@@ -16,7 +16,6 @@ allowed-tools: SlashCommand(*), Bash(*)
Direct interaction with CLI tools for codebase analysis and Q&A. Direct interaction with CLI tools for codebase analysis and Q&A.
**Supported Tools**: codex, gemini (default), qwen **Supported Tools**: codex, gemini (default), qwen
**Reference**: @~/.claude/workflows/intelligent-tools-strategy.md for complete tool details
## Parameters ## Parameters

264
.claude/commands/cli/codex-execute.md
View File

@@ -17,18 +17,23 @@ allowed-tools: SlashCommand(*), Bash(*), TodoWrite(*), Read(*), Glob(*)
Automated task decomposition and sequential execution with Codex, using `codex exec "..." resume --last` mechanism for continuity between subtasks. Automated task decomposition and sequential execution with Codex, using `codex exec "..." resume --last` mechanism for continuity between subtasks.
**Input**: User description or task ID (automatically loads from `.task/[ID].json` if applicable) **Input**: User description or task ID (automatically loads from `.task/[ID].json` if applicable)
**Reference**: @~/.claude/workflows/intelligent-tools-strategy.md for Codex details
## Core Workflow ## Core Workflow
``` ```
Task Input → Decompose into Subtasks → TodoWrite Tracking → Task Input → Analyze Dependencies → Create Task Flow Diagram →
For Each Subtask: Decompose into Subtask Groups → TodoWrite Tracking →
0. Stage existing changes (git add -A) if valid git repo For Each Subtask Group:
1. Execute with Codex For First Subtask in Group:
2. [Optional] Git verification 0. Stage existing changes (git add -A) if valid git repo
3. Mark complete in TodoWrite 1. Execute with Codex (new session)
4. Resume next subtask with `codex resume --last` 2. [Optional] Git verification
3. Mark complete in TodoWrite
For Related Subtasks in Same Group:
0. Stage changes from previous subtask
1. Execute with `codex exec "..." resume --last` (continue session)
2. [Optional] Git verification
3. Mark complete in TodoWrite
→ Final Summary → Final Summary
``` ```
@@ -41,17 +46,49 @@ For Each Subtask:
## Execution Flow ## Execution Flow
### Phase 1: Input Processing & Decomposition ### Phase 1: Input Processing & Task Flow Analysis
1. **Parse Input**: 1. **Parse Input**:
- Check if input matches task ID pattern (e.g., `IMPL-001`, `TASK-123`) - Check if input matches task ID pattern (e.g., `IMPL-001`, `TASK-123`)
- If yes: Load from `.task/[ID].json` and extract requirements - If yes: Load from `.task/[ID].json` and extract requirements
- If no: Use input as task description directly - If no: Use input as task description directly
2. **Analyze & Decompose**: 2. **Analyze Dependencies & Create Task Flow Diagram**:
- Analyze task complexity and scope - Analyze task complexity and scope
- Break down into 3-8 subtasks - Identify dependencies and relationships between subtasks
- Create TodoWrite tracker with all subtasks - Create visual task flow diagram showing:
- Independent task groups (parallel execution possible)
- Sequential dependencies (must use resume)
- Branching logic (conditional paths)
- Display flow diagram for user review
**Task Flow Diagram Format**:
```
[Group A: Auth Core]
A1: Create user model ──┐
A2: Add validation ─┤─► [resume] ─► A3: Database schema
│
[Group B: API Layer] │
B1: Auth endpoints ─────┘─► [new session]
B2: Middleware ────────────► [resume] ─► B3: Error handling
[Group C: Testing]
C1: Unit tests ─────────────► [new session]
C2: Integration tests ──────► [resume]
```
**Diagram Symbols**:
- `──►` Sequential dependency (must resume previous session)
- `─┐` Branch point (multiple paths)
- `─┘` Merge point (wait for completion)
- `[resume]` Use `codex exec "..." resume --last`
- `[new session]` Start fresh Codex session
3. **Decompose into Subtask Groups**:
- Group related subtasks that share context
- Break down into 3-8 subtasks total
- Assign each subtask to a group
- Create TodoWrite tracker with groups
- Display decomposition for user review - Display decomposition for user review
**Decomposition Criteria**: **Decomposition Criteria**:
@@ -59,8 +96,9 @@ For Each Subtask:
- Clear, testable outcomes - Clear, testable outcomes
- Explicit dependencies - Explicit dependencies
- Focused file scope (1-5 files per subtask) - Focused file scope (1-5 files per subtask)
- **Group coherence**: Subtasks in same group share context/files
### Phase 2: Sequential Execution ### Phase 2: Group-Based Execution
**Pre-Execution Git Staging** (if valid git repository): **Pre-Execution Git Staging** (if valid git repository):
```bash ```bash
@@ -70,37 +108,61 @@ git add -A
git status --short git status --short
``` ```
**For First Subtask**: **For First Subtask in Each Group** (New Session):
```bash ```bash
# Initial execution (no resume needed) # Start new Codex session for independent task group
codex -C [dir] --full-auto exec " codex -C [dir] --full-auto exec "
PURPOSE: [task goal] PURPOSE: [group goal]
TASK: [subtask 1 description] TASK: [subtask description - first in group]
CONTEXT: @{relevant_files} @{CLAUDE.md} CONTEXT: @{relevant_files} @{CLAUDE.md}
EXPECTED: [specific deliverables] EXPECTED: [specific deliverables]
RULES: [constraints] RULES: [constraints]
Subtask 1 of N: [subtask title] Group [X]: [group name] - Subtask 1 of N in this group
" --skip-git-repo-check -s danger-full-access " --skip-git-repo-check -s danger-full-access
``` ```
**For Subsequent Subtasks** (using resume --last): **For Related Subtasks in Same Group** (Resume Session):
```bash ```bash
# Stage changes from previous subtask (if valid git repository) # Stage changes from previous subtask (if valid git repository)
git add -A git add -A
# Resume previous session for context continuity # Resume session ONLY for subtasks in same group
codex exec " codex exec "
CONTINUE TO NEXT SUBTASK: CONTINUE IN SAME GROUP:
Subtask N of M: [subtask title] Group [X]: [group name] - Subtask N of M
PURPOSE: [continuation goal] PURPOSE: [continuation goal within group]
TASK: [subtask N description] TASK: [subtask N description]
CONTEXT: Previous work completed, now focus on @{new_relevant_files} CONTEXT: Previous work in this group completed, now focus on @{new_relevant_files}
EXPECTED: [specific deliverables] EXPECTED: [specific deliverables]
RULES: Build on previous subtask, maintain consistency RULES: Build on previous subtask in group, maintain consistency
" resume --last --skip-git-repo-check -s danger-full-access " resume --last --skip-git-repo-check -s danger-full-access
``` ```
**For First Subtask in Different Group** (New Session):
```bash
# Stage changes from previous group
git add -A
# Start NEW session for different group (no resume)
codex -C [dir] --full-auto exec "
PURPOSE: [new group goal]
TASK: [subtask description - first in new group]
CONTEXT: @{different_files} @{CLAUDE.md}
EXPECTED: [specific deliverables]
RULES: [constraints]
Group [Y]: [new group name] - Subtask 1 of N in this group
" --skip-git-repo-check -s danger-full-access
```
**Resume Decision Logic**:
```
if (subtask.group == previous_subtask.group):
use `codex exec "..." resume --last` # Continue session
else:
use `codex -C [dir] exec "..."` # New session
```
### Phase 3: Verification (if --verify-git enabled) ### Phase 3: Verification (if --verify-git enabled)
After each subtask completion: After each subtask completion:
@@ -121,16 +183,26 @@ git ls-files --others --exclude-standard
- No merge conflicts or errors - No merge conflicts or errors
- Code compiles/runs (if applicable) - Code compiles/runs (if applicable)
### Phase 4: TodoWrite Tracking ### Phase 4: TodoWrite Tracking with Groups
**Initial Setup**: **Initial Setup with Task Flow**:
```javascript ```javascript
TodoWrite({ TodoWrite({
todos: [ todos: [
{ content: "Subtask 1: [description]", status: "in_progress", activeForm: "Executing subtask 1" }, // Display task flow diagram first
{ content: "Subtask 2: [description]", status: "pending", activeForm: "Executing subtask 2" }, { content: "Task Flow Analysis Complete - See diagram above", status: "completed", activeForm: "Analyzing task flow" },
{ content: "Subtask 3: [description]", status: "pending", activeForm: "Executing subtask 3" },
// ... more subtasks // Group A subtasks (will use resume within group)
{ content: "[Group A] Subtask 1: [description]", status: "in_progress", activeForm: "Executing Group A subtask 1" },
{ content: "[Group A] Subtask 2: [description] [resume]", status: "pending", activeForm: "Executing Group A subtask 2" },
// Group B subtasks (new session, then resume within group)
{ content: "[Group B] Subtask 1: [description] [new session]", status: "pending", activeForm: "Executing Group B subtask 1" },
{ content: "[Group B] Subtask 2: [description] [resume]", status: "pending", activeForm: "Executing Group B subtask 2" },
// Group C subtasks (new session)
{ content: "[Group C] Subtask 1: [description] [new session]", status: "pending", activeForm: "Executing Group C subtask 1" },
{ content: "Final verification and summary", status: "pending", activeForm: "Verifying and summarizing" } { content: "Final verification and summary", status: "pending", activeForm: "Verifying and summarizing" }
] ]
}) })
@@ -140,8 +212,9 @@ TodoWrite({
```javascript ```javascript
TodoWrite({ TodoWrite({
todos: [ todos: [
{ content: "Subtask 1: [description]", status: "completed", activeForm: "Executing subtask 1" }, { content: "Task Flow Analysis Complete - See diagram above", status: "completed", activeForm: "Analyzing task flow" },
{ content: "Subtask 2: [description]", status: "in_progress", activeForm: "Executing subtask 2" }, { content: "[Group A] Subtask 1: [description]", status: "completed", activeForm: "Executing Group A subtask 1" },
{ content: "[Group A] Subtask 2: [description] [resume]", status: "in_progress", activeForm: "Executing Group A subtask 2" },
// ... update status // ... update status
] ]
}) })
@@ -149,18 +222,36 @@ TodoWrite({
## Codex Resume Mechanism ## Codex Resume Mechanism
**Why `codex resume --last`?** **Why Group-Based Resume?**
- Maintains conversation context across subtasks - **Within Group**: Maintains conversation context for related subtasks
- Codex remembers previous decisions and patterns - Codex remembers previous decisions and patterns
- Reduces context repetition - Reduces context repetition
- Ensures consistency in implementation style - Ensures consistency in implementation style
- **Between Groups**: Fresh session for independent tasks
- Avoids context pollution from unrelated work
- Prevents confusion when switching domains
- Maintains focused attention on current group
**How It Works**: **How It Works**:
1. First subtask creates new Codex session 1. **First subtask in Group A**: Creates new Codex session
2. After completion, session is saved 2. **Subsequent subtasks in Group A**: Use `codex resume --last` to continue session
3. Subsequent subtasks use `codex resume --last` to continue 3. **First subtask in Group B**: Creates NEW Codex session (no resume)
4. Each subtask builds on previous context 4. **Subsequent subtasks in Group B**: Use `codex resume --last` within Group B
5. Final subtask completes full task 5. Each group builds on its own context, isolated from other groups
**When to Resume vs New Session**:
```
✅ RESUME (same group):
- Subtasks share files/modules
- Logical continuation of previous work
- Same architectural domain
❌ NEW SESSION (different group):
- Independent task area
- Different files/modules
- Switching architectural domains
- Testing after implementation
```
**Image Support**: **Image Support**:
```bash ```bash
@@ -199,25 +290,47 @@ codex exec "CONTINUE TO NEXT SUBTASK: ..." resume --last --skip-git-repo-check -
**During Execution**: **During Execution**:
``` ```
📋 Task Decomposition: 📊 Task Flow Diagram:
1. [Subtask 1 description] [Group A: Auth Core]
2. [Subtask 2 description] A1: Create user model ──┐
... A2: Add validation ─┤─► [resume] ─► A3: Database schema
│
[Group B: API Layer] │
B1: Auth endpoints ─────┘─► [new session]
B2: Middleware ────────────► [resume] ─► B3: Error handling
▶️ Executing Subtask 1/N: [title] [Group C: Testing]
Codex session started... C1: Unit tests ─────────────► [new session]
C2: Integration tests ──────► [resume]
📋 Task Decomposition:
[Group A] 1. Create user model
[Group A] 2. Add validation logic [resume]
[Group A] 3. Implement database schema [resume]
[Group B] 4. Create auth endpoints [new session]
[Group B] 5. Add middleware [resume]
[Group B] 6. Error handling [resume]
[Group C] 7. Unit tests [new session]
[Group C] 8. Integration tests [resume]
▶️ [Group A] Executing Subtask 1/8: Create user model
Starting new Codex session for Group A...
[Codex output] [Codex output]
✅ Subtask 1 completed ✅ Subtask 1 completed
🔍 Git Verification: 🔍 Git Verification:
M src/file1.ts M src/models/user.ts
A src/file2.ts
✅ Changes verified ✅ Changes verified
▶️ Executing Subtask 2/N: [title] ▶️ [Group A] Executing Subtask 2/8: Add validation logic
Resuming Codex session... Resuming Codex session (same group)...
[Codex output] [Codex output]
✅ Subtask 2 completed ✅ Subtask 2 completed
▶️ [Group B] Executing Subtask 4/8: Create auth endpoints
Starting NEW Codex session for Group B...
[Codex output]
✅ Subtask 4 completed
... ...
✅ All Subtasks Completed ✅ All Subtasks Completed
@@ -248,16 +361,30 @@ codex exec "CONTINUE TO NEXT SUBTASK: ..." resume --last --skip-git-repo-check -
## Examples ## Examples
**Example 1: Simple Task** **Example 1: Simple Task with Groups**
```bash ```bash
/cli:codex-execute "implement user authentication system" /cli:codex-execute "implement user authentication system"
# Decomposes into: # Task Flow Diagram:
# 1. Create user model and database schema # [Group A: Data Layer]
# 2. Implement JWT token generation # A1: Create user model ──► [resume] ──► A2: Database schema
# 3. Create authentication middleware #
# 4. Add login/logout endpoints # [Group B: Auth Logic]
# 5. Write tests for auth flow # B1: JWT token generation ──► [new session]
# B2: Authentication middleware ──► [resume]
#
# [Group C: API Endpoints]
# C1: Login/logout endpoints ──► [new session]
#
# [Group D: Testing]
# D1: Unit tests ──► [new session]
# D2: Integration tests ──► [resume]
# Execution:
# Group A: A1 (new) → A2 (resume)
# Group B: B1 (new) → B2 (resume)
# Group C: C1 (new)
# Group D: D1 (new) → D2 (resume)
``` ```
**Example 2: With Git Verification** **Example 2: With Git Verification**
@@ -280,13 +407,16 @@ codex exec "CONTINUE TO NEXT SUBTASK: ..." resume --last --skip-git-repo-check -
## Best Practices ## Best Practices
1. **Subtask Granularity**: Keep subtasks small and focused 1. **Task Flow First**: Always create visual flow diagram before execution
2. **Clear Boundaries**: Each subtask should have well-defined input/output 2. **Group Related Work**: Cluster subtasks by domain/files for efficient resume
3. **Git Hygiene**: Use `--verify-git` for critical refactoring 3. **Subtask Granularity**: Keep subtasks small and focused (5-15 min each)
4. **Pre-Execution Staging**: Stage changes before each subtask to clearly see codex modifications 4. **Clear Boundaries**: Each subtask should have well-defined input/output
5. **Context Continuity**: Let `codex resume --last` maintain context 5. **Git Hygiene**: Use `--verify-git` for critical refactoring
6. **Recovery Points**: TodoWrite provides clear progress tracking 6. **Pre-Execution Staging**: Stage changes before each subtask to clearly see codex modifications
7. **Image References**: Attach design files for UI tasks 7. **Smart Resume**: Use `resume --last` ONLY within same group
8. **Fresh Sessions**: Start new session when switching to different group/domain
9. **Recovery Points**: TodoWrite with group labels provides clear progress tracking
10. **Image References**: Attach design files for UI tasks (first subtask in group)
## Input Processing ## Input Processing

1
.claude/commands/cli/execute.md
View File

@@ -18,7 +18,6 @@ allowed-tools: SlashCommand(*), Bash(*)
Execute implementation tasks with YOLO permissions (auto-approves all confirmations). Execute implementation tasks with YOLO permissions (auto-approves all confirmations).
**Supported Tools**: codex, gemini (default), qwen **Supported Tools**: codex, gemini (default), qwen
**Reference**: @~/.claude/workflows/intelligent-tools-strategy.md for complete tool details
## YOLO Permissions ## YOLO Permissions

2
.claude/commands/cli/mode/bug-index.md
View File

@@ -37,8 +37,6 @@ Execute systematic bug analysis and fix suggestions using CLI tools with diagnos
## Command Template ## Command Template
**Core Guidelines**: @~/.claude/workflows/intelligent-tools-strategy.md
```bash ```bash
cd [directory] && ~/.claude/scripts/gemini-wrapper --all-files -p " cd [directory] && ~/.claude/scripts/gemini-wrapper --all-files -p "
PURPOSE: [bug analysis goal] PURPOSE: [bug analysis goal]

2
.claude/commands/cli/mode/code-analysis.md
View File

@@ -48,8 +48,6 @@ The code-analysis template provides:
## Command Templates ## Command Templates
**Core Guidelines**: @~/.claude/workflows/intelligent-tools-strategy.md
### Gemini (Default) ### Gemini (Default)
```bash ```bash
cd [directory] && ~/.claude/scripts/gemini-wrapper --all-files -p " cd [directory] && ~/.claude/scripts/gemini-wrapper --all-files -p "

2
.claude/commands/cli/mode/plan.md
View File

@@ -37,8 +37,6 @@ Execute planning and architecture analysis using CLI tools with specialized temp
## Command Template ## Command Template
**Core Guidelines**: @~/.claude/workflows/intelligent-tools-strategy.md
```bash ```bash
cd [directory] && ~/.claude/scripts/gemini-wrapper --all-files -p " cd [directory] && ~/.claude/scripts/gemini-wrapper --all-files -p "
PURPOSE: [planning goal from topic] PURPOSE: [planning goal from topic]

498
.claude/workflows/intelligent-tools-strategy.md
View File

@@ -6,11 +6,22 @@ type: strategic-guideline
# Intelligent Tools Selection Strategy # Intelligent Tools Selection Strategy
## 📋 Table of Contents
1. [Core Framework](#-core-framework)
2. [Tool Specifications](#-tool-specifications)
3. [Command Templates](#-command-templates)
4. [Tool Selection Guide](#-tool-selection-guide)
5. [Usage Patterns](#-usage-patterns)
6. [Best Practices](#-best-practices)
---
## ⚡ Core Framework ## ⚡ Core Framework
**Gemini**: Analysis, understanding, exploration & documentation ### Tool Overview
**Qwen**: Architecture analysis, code generation & implementation - **Gemini**: Analysis, understanding, exploration & documentation
**Codex**: Development, implementation & automation - **Qwen**: Architecture analysis, code generation & implementation
- **Codex**: Development, implementation & automation
### Decision Principles ### Decision Principles
- **Use tools early and often** - Tools are faster, more thorough, and reliable than manual approaches - **Use tools early and often** - Tools are faster, more thorough, and reliable than manual approaches
@@ -27,30 +38,92 @@ type: strategic-guideline
4. **Not sure?** → Use multiple tools in parallel 4. **Not sure?** → Use multiple tools in parallel
5. **Small task?** → Still use tools - they're faster than manual work 5. **Small task?** → Still use tools - they're faster than manual work
### Core Execution Rules ---
- **Dynamic Timeout (20-120min)**: Allocate execution time based on task complexity
- Simple tasks (analysis, search): 20-40min (1200000-2400000ms)
- Medium tasks (refactoring, documentation): 40-60min (2400000-3600000ms)
- Complex tasks (implementation, migration): 60-120min (3600000-7200000ms)
- **Codex Multiplier**: Codex commands use 1.5x of allocated time
- **Apply to All Tools**: All bash() wrapped commands including Gemini, Qwen wrapper and Codex executions
- **Command Examples**: `bash(~/.claude/scripts/gemini-wrapper -p "prompt")`, `bash(codex -C directory --full-auto exec "task")`
- **Auto-detect**: Analyze PURPOSE and TASK fields to determine appropriate timeout
### Permission Framework ## 🎯 Tool Specifications
- **⚠️ WRITE PROTECTION**: Local codebase write/modify requires EXPLICIT user confirmation
- **Analysis Mode (default)**: Read-only, safe for auto-execution
- **Write Mode**: Requires user explicitly states MODE=write or MODE=auto in prompt
- **Exception**: User provides clear instructions like "modify", "create", "implement"
- **Gemini/Qwen Write Access**: Use `--approval-mode yolo` ONLY when MODE=write explicitly specified
- **Parameter Position**: Place AFTER the wrapper command: `gemini-wrapper --approval-mode yolo -p "..."`
- **Codex Write Access**: Use `-s danger-full-access` and `--skip-git-repo-check` ONLY when MODE=auto explicitly specified
- **Parameter Position**: Place AFTER the prompt string at command END: `codex ... exec "..." --skip-git-repo-check -s danger-full-access`
- **Default Behavior**: All tools default to analysis/read-only mode without explicit write permission
## 🎯 Universal Command Template ### Gemini
- **Command**: `~/.claude/scripts/gemini-wrapper`
- **Strengths**: Large context window, pattern recognition
- **Best For**: Analysis, documentation generation, code exploration
- **Permissions**: Default read-only analysis, MODE=write requires explicit specification (auto-enables --approval-mode yolo)
- **Default MODE**: `analysis` (read-only)
- **⚠️ Write Trigger**: Only when user explicitly requests "generate documentation", "modify code", or specifies MODE=write
### Standard Format (REQUIRED) #### MODE Options
- `analysis` (default) - Read-only analysis and documentation generation
- `write` - ⚠️ Create/modify codebase files (requires explicit specification, auto-enables --approval-mode yolo)
### Qwen
- **Command**: `~/.claude/scripts/qwen-wrapper`
- **Strengths**: Architecture analysis, pattern recognition
- **Best For**: System design analysis, architectural review
- **Permissions**: Architecture analysis only, no automatic code generation
- **Default MODE**: `analysis` (read-only)
- **⚠️ Write Trigger**: Explicitly prohibited from auto-calling write mode
#### MODE Options
- `analysis` (default) - Architecture analysis only, no code generation/modification (read-only)
- `write` - ⚠️ Code generation (requires explicit specification, disabled by default)
### Codex
- **Command**: `codex --full-auto exec`
- **Strengths**: Autonomous development, mathematical reasoning
- **Best For**: Implementation, testing, automation
- **Permissions**: Requires explicit MODE=auto or MODE=write specification
- **Default MODE**: No default, must be explicitly specified
- **⚠️ Write Trigger**: Only when user explicitly requests "implement", "modify", "generate code" AND specifies MODE
#### MODE Options
- `auto` - ⚠️ Autonomous development with full file operations (requires explicit specification, enables -s danger-full-access)
- `write` - ⚠️ Test generation and file modification (requires explicit specification)
- **Default**: No default mode, MODE must be explicitly specified
#### Session Management
- `codex resume` - Resume previous interactive session (picker by default)
- `codex exec "task" resume --last` - Continue most recent session with new task (maintains context)
- `codex -i <image_file>` - Attach image(s) to initial prompt (useful for UI/design references)
- **Multi-task Pattern**: First task uses `exec`, subsequent tasks use `exec "..." resume --last` for context continuity
- **Parameter Position**: `resume --last` must be placed AFTER the prompt string at command END
- **Example**:
```bash
# First task - establish session
codex -C project --full-auto exec "Implement auth module" --skip-git-repo-check -s danger-full-access
# Subsequent tasks - continue same session
codex --full-auto exec "Add JWT validation" resume --last --skip-git-repo-check -s danger-full-access
codex --full-auto exec "Write auth tests" resume --last --skip-git-repo-check -s danger-full-access
```
#### Auto-Resume Decision Rules
**When to use `resume --last`**:
- Current task is related to/extends previous Codex task in conversation memory
- Current task requires context from previous implementation
- Current task is part of multi-step workflow (e.g., implement → enhance → test)
- Session memory indicates recent Codex execution on same module/feature
**When NOT to use `resume --last`**:
- First Codex task in conversation
- New independent task unrelated to previous work
- Switching to different module/feature area
- No recent Codex task in conversation memory
---
## 🎯 Command Templates
### Universal Template Structure
Every command MUST follow this structure:
- [ ] **PURPOSE** - Clear goal and intent
- [ ] **TASK** - Specific execution task
- [ ] **MODE** - Execution mode and permission level
- [ ] **CONTEXT** - File references and memory context from previous sessions
- [ ] **EXPECTED** - Clear expected results
- [ ] **RULES** - Template reference and constraints
### Standard Command Formats
#### Gemini Commands
```bash ```bash
# Gemini Analysis (read-only, default) # Gemini Analysis (read-only, default)
cd [directory] && ~/.claude/scripts/gemini-wrapper -p " cd [directory] && ~/.claude/scripts/gemini-wrapper -p "
@@ -72,7 +145,10 @@ CONTEXT: [file references and memory context]
EXPECTED: [expected output] EXPECTED: [expected output]
RULES: [template reference and constraints] RULES: [template reference and constraints]
" "
```
#### Qwen Commands
```bash
# Qwen Architecture Analysis (read-only, default) # Qwen Architecture Analysis (read-only, default)
cd [directory] && ~/.claude/scripts/qwen-wrapper -p " cd [directory] && ~/.claude/scripts/qwen-wrapper -p "
PURPOSE: [clear architecture goal] PURPOSE: [clear architecture goal]
@@ -93,7 +169,10 @@ CONTEXT: [file references and memory context]
EXPECTED: [expected deliverables] EXPECTED: [expected deliverables]
RULES: [template reference and constraints] RULES: [template reference and constraints]
" "
```
#### Codex Commands
```bash
# Codex Development (requires explicit MODE=auto) # Codex Development (requires explicit MODE=auto)
# NOTE: --skip-git-repo-check and -s danger-full-access must be placed at command END # NOTE: --skip-git-repo-check and -s danger-full-access must be placed at command END
codex -C [directory] --full-auto exec " codex -C [directory] --full-auto exec "
@@ -117,32 +196,7 @@ RULES: [template reference and constraints]
" --skip-git-repo-check -s danger-full-access " --skip-git-repo-check -s danger-full-access
``` ```
### Template Structure ### Directory Context Configuration
- [ ] **PURPOSE** - Clear goal and intent
- [ ] **TASK** - Specific execution task
- [ ] **MODE** - Execution mode and permission level
- [ ] **CONTEXT** - File references and memory context from previous sessions
- [ ] **EXPECTED** - Clear expected results
- [ ] **RULES** - Template reference and constraints
### MODE Field Definition
The MODE field controls execution behavior and file permissions:
**For Gemini**:
- `analysis` (default) - Read-only analysis and documentation generation
- `write` - ⚠️ Create/modify codebase files (requires explicit specification, auto-enables --approval-mode yolo)
**For Qwen**:
- `analysis` (default) - Architecture analysis only, no code generation/modification (read-only)
- `write` - ⚠️ Code generation (requires explicit specification, disabled by default)
**For Codex**:
- `auto` - ⚠️ Autonomous development with full file operations (requires explicit specification, enables -s danger-full-access)
- `write` - ⚠️ Test generation and file modification (requires explicit specification)
- **Default**: No default mode, MODE must be explicitly specified
### Directory Context
Tools execute in current working directory: Tools execute in current working directory:
- **Gemini**: `cd path/to/project && ~/.claude/scripts/gemini-wrapper -p "prompt"` - **Gemini**: `cd path/to/project && ~/.claude/scripts/gemini-wrapper -p "prompt"`
- **Qwen**: `cd path/to/project && ~/.claude/scripts/qwen-wrapper -p "prompt"` - **Qwen**: `cd path/to/project && ~/.claude/scripts/qwen-wrapper -p "prompt"`
@@ -150,7 +204,7 @@ Tools execute in current working directory:
- **Path types**: Supports both relative (`../project`) and absolute (`/full/path`) paths - **Path types**: Supports both relative (`../project`) and absolute (`/full/path`) paths
- **Token analysis**: For gemini-wrapper and qwen-wrapper, token counting happens in current directory - **Token analysis**: For gemini-wrapper and qwen-wrapper, token counting happens in current directory
### Rules Field Format ### RULES Field Format
```bash ```bash
RULES: $(cat "~/.claude/workflows/cli-templates/prompts/[category]/[template].txt") | [constraints] RULES: $(cat "~/.claude/workflows/cli-templates/prompts/[category]/[template].txt") | [constraints]
``` ```
@@ -161,167 +215,7 @@ RULES: $(cat "~/.claude/workflows/cli-templates/prompts/[category]/[template].tx
- No template: `Focus on security patterns, include dependency analysis` - No template: `Focus on security patterns, include dependency analysis`
- File patterns: `@{src/**/*.ts,CLAUDE.md} - Stay within scope` - File patterns: `@{src/**/*.ts,CLAUDE.md} - Stay within scope`
## 📊 Tool Selection Matrix ### File Pattern Reference
| Task Type | Tool | Use Case | Template |
|-----------|------|----------|-----------|
| **Analysis** | Gemini | Code exploration, architecture review, patterns | `analysis/pattern.txt` |
| **Architecture** | Qwen | System design, code generation, architectural analysis | `analysis/architecture.txt` |
| **Code Generation** | Qwen | Implementation patterns, code scaffolding, component creation | `development/feature.txt` |
| **Development** | Codex | Feature implementation, bug fixes, testing | `development/feature.txt` |
| **Planning** | Multiple | Task breakdown, migration planning | `planning/task-breakdown.txt` |
| **Documentation** | Multiple | Code docs, API specs, guides | `analysis/quality.txt` |
| **Security** | Codex | Vulnerability assessment, fixes | `analysis/security.txt` |
| **Refactoring** | Multiple | Gemini for analysis, Qwen/Codex for execution | `development/refactor.txt` |
## 📁 Template System
**Base Structure**: `~/.claude/workflows/cli-templates/`
### Available Templates
```
prompts/
├── analysis/
│ ├── pattern.txt - Code pattern analysis
│ ├── architecture.txt - System architecture review
│ ├── security.txt - Security assessment
│ └── quality.txt - Code quality review
├── development/
│ ├── feature.txt - Feature implementation
│ ├── refactor.txt - Refactoring tasks
│ └── testing.txt - Test generation
└── planning/
└── task-breakdown.txt - Task decomposition
planning-roles/
├── system-architect.md - System design perspective
├── security-expert.md - Security architecture
└── feature-planner.md - Feature specification
tech-stacks/
├── typescript-dev.md - TypeScript guidelines
├── python-dev.md - Python conventions
└── react-dev.md - React architecture
```
## 🚀 Usage Patterns
### Workflow Integration (REQUIRED)
When planning any coding task, **ALWAYS** integrate CLI tools:
1. **Understanding Phase**: Use Gemini for analysis
2. **Architecture Phase**: Use Qwen for design and code generation
3. **Implementation Phase**: Use Qwen/Codex for development
4. **Quality Phase**: Use Codex for testing and validation
### Common Scenarios
```bash
# Gemini - Code Analysis
~/.claude/scripts/gemini-wrapper -p "
PURPOSE: Understand codebase architecture
TASK: Analyze project structure and identify patterns
MODE: analysis
CONTEXT: @{src/**/*.ts,CLAUDE.md} Previous analysis of auth system
EXPECTED: Architecture overview and integration points
RULES: $(cat '~/.claude/workflows/cli-templates/prompts/analysis/architecture.txt') | Focus on integration points
"
# Gemini - Generate Documentation
~/.claude/scripts/gemini-wrapper -p "
PURPOSE: Generate API documentation
TASK: Create comprehensive API reference from code
MODE: write
CONTEXT: @{src/api/**/*}
EXPECTED: API.md with all endpoints documented
RULES: Follow project documentation standards
"
# Qwen - Architecture Analysis
cd src/auth && ~/.claude/scripts/qwen-wrapper -p "
PURPOSE: Analyze authentication system architecture
TASK: Review JWT-based auth system design
MODE: analysis
CONTEXT: @{src/auth/**/*} Existing patterns and requirements
EXPECTED: Architecture analysis report with recommendations
RULES: $(cat '~/.claude/workflows/cli-templates/prompts/analysis/architecture.txt') | Focus on security
"
# Codex - Feature Development
codex -C path/to/project --full-auto exec "
PURPOSE: Implement user authentication
TASK: Create JWT-based authentication system
MODE: auto
CONTEXT: @{src/auth/**/*} Database schema from session memory
EXPECTED: Complete auth module with tests
RULES: $(cat '~/.claude/workflows/cli-templates/prompts/development/feature.txt') | Follow security best practices
" --skip-git-repo-check -s danger-full-access
# Codex - Test Generation
codex -C src/auth --full-auto exec "
PURPOSE: Increase test coverage
TASK: Generate comprehensive tests for auth module
MODE: write
CONTEXT: @{**/*.ts} Exclude existing tests
EXPECTED: Complete test suite with 80%+ coverage
RULES: Use Jest, follow existing patterns
" --skip-git-repo-check -s danger-full-access
```
## 📋 Planning Checklist
For every development task:
- [ ] **Purpose defined** - Clear goal and intent
- [ ] **Mode selected** - Execution mode and permission level determined
- [ ] **Context gathered** - File references and session memory documented
- [ ] **Gemini analysis** completed for understanding
- [ ] **Template selected** - Appropriate template chosen
- [ ] **Constraints specified** - File patterns, scope, requirements
- [ ] **Implementation approach** - Tool selection and workflow
- [ ] **Quality measures** - Testing and validation plan
- [ ] **Tool configuration** - Review `.gemini/CLAUDE.md` or `.codex/Agent.md` if needed
## 🎯 Key Features
### Gemini
- **Command**: `~/.claude/scripts/gemini-wrapper`
- **Strengths**: Large context window, pattern recognition
- **Best For**: Analysis, documentation generation, code exploration
- **Permissions**: Default read-only analysis, MODE=write requires explicit specification (auto-enables --approval-mode yolo)
- **Default MODE**: `analysis` (read-only)
- **⚠️ Write Trigger**: Only when user explicitly requests "generate documentation", "modify code", or specifies MODE=write
### Qwen
- **Command**: `~/.claude/scripts/qwen-wrapper`
- **Strengths**: Architecture analysis, pattern recognition
- **Best For**: System design analysis, architectural review
- **Permissions**: Architecture analysis only, no automatic code generation
- **Default MODE**: `analysis` (read-only)
- **⚠️ Write Trigger**: Explicitly prohibited from auto-calling write mode
### Codex
- **Command**: `codex --full-auto exec`
- **Strengths**: Autonomous development, mathematical reasoning
- **Best For**: Implementation, testing, automation
- **Permissions**: Requires explicit MODE=auto or MODE=write specification
- **Default MODE**: No default, must be explicitly specified
- **⚠️ Write Trigger**: Only when user explicitly requests "implement", "modify", "generate code" AND specifies MODE
- **Session Management**:
- `codex resume` - Resume previous interactive session (picker by default)
- `codex exec "task" resume --last` - Continue most recent session with new task (maintains context)
- `codex -i <image_file>` - Attach image(s) to initial prompt (useful for UI/design references)
- **Multi-task Pattern**: First task uses `exec`, subsequent tasks use `exec "..." resume --last` for context continuity
- **Parameter Position**: `resume --last` must be placed AFTER the prompt string at command END
- **Example**:
```bash
# First task - establish session
codex -C project --full-auto exec "Implement auth module" --skip-git-repo-check -s danger-full-access
# Subsequent tasks - continue same session
codex --full-auto exec "Add JWT validation" resume --last --skip-git-repo-check -s danger-full-access
codex --full-auto exec "Write auth tests" resume --last --skip-git-repo-check -s danger-full-access
```
### File Patterns
- All files: `@{**/*}` - All files: `@{**/*}`
- Source files: `@{src/**/*}` - Source files: `@{src/**/*}`
- TypeScript: `@{*.ts,*.tsx}` - TypeScript: `@{*.ts,*.tsx}`
@@ -354,8 +248,150 @@ RULES: Focus on type safety and component composition
" "
``` ```
---
## 📊 Tool Selection Guide
### Selection Matrix
| Task Type | Tool | Use Case | Template |
|-----------|------|----------|-----------|
| **Analysis** | Gemini | Code exploration, architecture review, patterns | `analysis/pattern.txt` |
| **Architecture** | Qwen | System design, code generation, architectural analysis | `analysis/architecture.txt` |
| **Code Generation** | Qwen | Implementation patterns, code scaffolding, component creation | `development/feature.txt` |
| **Development** | Codex | Feature implementation, bug fixes, testing | `development/feature.txt` |
| **Planning** | Multiple | Task breakdown, migration planning | `planning/task-breakdown.txt` |
| **Documentation** | Multiple | Code docs, API specs, guides | `analysis/quality.txt` |
| **Security** | Codex | Vulnerability assessment, fixes | `analysis/security.txt` |
| **Refactoring** | Multiple | Gemini for analysis, Qwen/Codex for execution | `development/refactor.txt` |
### Template System
**Base Structure**: `~/.claude/workflows/cli-templates/`
#### Available Templates
```
prompts/
├── analysis/
│ ├── pattern.txt - Code pattern analysis
│ ├── architecture.txt - System architecture review
│ ├── security.txt - Security assessment
│ └── quality.txt - Code quality review
├── development/
│ ├── feature.txt - Feature implementation
│ ├── refactor.txt - Refactoring tasks
│ └── testing.txt - Test generation
└── planning/
└── task-breakdown.txt - Task decomposition
planning-roles/
├── system-architect.md - System design perspective
├── security-expert.md - Security architecture
└── feature-planner.md - Feature specification
tech-stacks/
├── typescript-dev.md - TypeScript guidelines
├── python-dev.md - Python conventions
└── react-dev.md - React architecture
```
---
## 🚀 Usage Patterns
### Workflow Integration (REQUIRED)
When planning any coding task, **ALWAYS** integrate CLI tools:
1. **Understanding Phase**: Use Gemini for analysis
2. **Architecture Phase**: Use Qwen for design and code generation
3. **Implementation Phase**: Use Qwen/Codex for development
4. **Quality Phase**: Use Codex for testing and validation
### Common Scenarios
#### Code Analysis
```bash
~/.claude/scripts/gemini-wrapper -p "
PURPOSE: Understand codebase architecture
TASK: Analyze project structure and identify patterns
MODE: analysis
CONTEXT: @{src/**/*.ts,CLAUDE.md} Previous analysis of auth system
EXPECTED: Architecture overview and integration points
RULES: $(cat '~/.claude/workflows/cli-templates/prompts/analysis/architecture.txt') | Focus on integration points
"
```
#### Documentation Generation
```bash
~/.claude/scripts/gemini-wrapper -p "
PURPOSE: Generate API documentation
TASK: Create comprehensive API reference from code
MODE: write
CONTEXT: @{src/api/**/*}
EXPECTED: API.md with all endpoints documented
RULES: Follow project documentation standards
"
```
#### Architecture Analysis
```bash
cd src/auth && ~/.claude/scripts/qwen-wrapper -p "
PURPOSE: Analyze authentication system architecture
TASK: Review JWT-based auth system design
MODE: analysis
CONTEXT: @{src/auth/**/*} Existing patterns and requirements
EXPECTED: Architecture analysis report with recommendations
RULES: $(cat '~/.claude/workflows/cli-templates/prompts/analysis/architecture.txt') | Focus on security
"
```
#### Feature Development (Multi-task with Resume)
```bash
# First task - establish session
codex -C path/to/project --full-auto exec "
PURPOSE: Implement user authentication
TASK: Create JWT-based authentication system
MODE: auto
CONTEXT: @{src/auth/**/*} Database schema from session memory
EXPECTED: Complete auth module with tests
RULES: $(cat '~/.claude/workflows/cli-templates/prompts/development/feature.txt') | Follow security best practices
" --skip-git-repo-check -s danger-full-access
# Continue in same session - Add JWT validation
codex --full-auto exec "
PURPOSE: Enhance authentication security
TASK: Add JWT token validation and refresh logic
MODE: auto
CONTEXT: Previous auth implementation from current session
EXPECTED: JWT validation middleware and token refresh endpoints
RULES: Follow JWT best practices, maintain session context
" resume --last --skip-git-repo-check -s danger-full-access
# Continue in same session - Add tests
codex --full-auto exec "
PURPOSE: Increase test coverage
TASK: Generate comprehensive tests for auth module
MODE: write
CONTEXT: Auth implementation from current session
EXPECTED: Complete test suite with 80%+ coverage
RULES: Use Jest, follow existing patterns
" resume --last --skip-git-repo-check -s danger-full-access
```
#### Interactive Session Resume
```bash
# Resume previous session with picker
codex resume
# Or resume most recent session directly
codex resume --last
```
---
## 🔧 Best Practices ## 🔧 Best Practices
### General Guidelines
- **Start with templates** - Use predefined templates for consistency - **Start with templates** - Use predefined templates for consistency
- **Be specific** - Clear PURPOSE, TASK, and EXPECTED fields - **Be specific** - Clear PURPOSE, TASK, and EXPECTED fields
- **Include constraints** - File patterns, scope, requirements in RULES - **Include constraints** - File patterns, scope, requirements in RULES
@@ -403,3 +439,41 @@ EXPECTED: Code improvements and fixes
RULES: Maintain backward compatibility RULES: Maintain backward compatibility
" --skip-git-repo-check -s danger-full-access " --skip-git-repo-check -s danger-full-access
``` ```
### Planning Checklist
For every development task:
- [ ] **Purpose defined** - Clear goal and intent
- [ ] **Mode selected** - Execution mode and permission level determined
- [ ] **Context gathered** - File references and session memory documented
- [ ] **Gemini analysis** completed for understanding
- [ ] **Template selected** - Appropriate template chosen
- [ ] **Constraints specified** - File patterns, scope, requirements
- [ ] **Implementation approach** - Tool selection and workflow
- [ ] **Quality measures** - Testing and validation plan
- [ ] **Tool configuration** - Review `.gemini/CLAUDE.md` or `.codex/Agent.md` if needed
---
## ⚙️ Execution Configuration
### Core Execution Rules
- **Dynamic Timeout (20-120min)**: Allocate execution time based on task complexity
- Simple tasks (analysis, search): 20-40min (1200000-2400000ms)
- Medium tasks (refactoring, documentation): 40-60min (2400000-3600000ms)
- Complex tasks (implementation, migration): 60-120min (3600000-7200000ms)
- **Codex Multiplier**: Codex commands use 1.5x of allocated time
- **Apply to All Tools**: All bash() wrapped commands including Gemini, Qwen wrapper and Codex executions
- **Command Examples**: `bash(~/.claude/scripts/gemini-wrapper -p "prompt")`, `bash(codex -C directory --full-auto exec "task")`
- **Auto-detect**: Analyze PURPOSE and TASK fields to determine appropriate timeout
### Permission Framework
- **⚠️ WRITE PROTECTION**: Local codebase write/modify requires EXPLICIT user confirmation
- **Analysis Mode (default)**: Read-only, safe for auto-execution
- **Write Mode**: Requires user explicitly states MODE=write or MODE=auto in prompt
- **Exception**: User provides clear instructions like "modify", "create", "implement"
- **Gemini/Qwen Write Access**: Use `--approval-mode yolo` ONLY when MODE=write explicitly specified
- **Parameter Position**: Place AFTER the wrapper command: `gemini-wrapper --approval-mode yolo -p "..."`
- **Codex Write Access**: Use `-s danger-full-access` and `--skip-git-repo-check` ONLY when MODE=auto explicitly specified
- **Parameter Position**: Place AFTER the prompt string at command END: `codex ... exec "..." --skip-git-repo-check -s danger-full-access`
- **Default Behavior**: All tools default to analysis/read-only mode without explicit write permission