Claude Code Integration: Configure Permissions to Run Pipeline | AI App Factory Tutorial

What You'll Learn

Configure Claude Code security permissions without using --dangerously-skip-permissions
Understand the permission whitelist automatically generated by Factory
Execute the complete 7-stage pipeline in Claude Code
Master cross-platform permission configuration (Windows/macOS/Linux)

The Problem

When using Factory for the first time, you may encounter:

Permission blocked: Claude Code prompts "no permission to read file"
Using dangerous parameters: Forced to add --dangerously-skip-permissions to bypass security checks
Manual configuration hassle: Unsure which operations should be allowed
Cross-platform issues: Inconsistent path permissions between Windows and Unix

Actually, Factory will automatically generate complete permission configurations. You just need to use them correctly.

When to Use This

When you need to run Factory pipelines in Claude Code:

After using factory init to initialize a project (auto-starts)
When using factory run to continue the pipeline
When manually launching Claude Code

Why recommend Claude Code?

Claude Code is Anthropic's official AI programming assistant, deeply integrated with Factory's permission system. Compared to other AI assistants, Claude Code's permission management is more precise and more secure.

Core Concepts

Factory's permission configuration uses a whitelist mechanism: only explicitly allowed operations are permitted; everything else is prohibited.

Permission Whitelist Categories

Category	Allowed Operations	Purpose
File Operations	Read/Write/Edit/Glob	Read and modify project files
Git Operations	git add/commit/push etc.	Version control
Directory Operations	ls/cd/tree/pwd	Browse directory structure
Build Tools	npm/yarn/pnpm	Install dependencies, run scripts
TypeScript	tsc/npx tsc	Type checking
Database	npx prisma	Database migration and management
Python	python/pip	UI design system
Testing	vitest/jest/test	Run tests
Factory CLI	factory init/run/continue	Pipeline commands
Docker	docker compose	Containerized deployment
Web Operations	WebFetch(domain:...)	Fetch API documentation
Skills	superpowers/ui-ux-pro-max	Plugin skills

Why not use `--dangerously-skip-permissions`?

Method	Security	Recommended
`--dangerously-skip-permissions`	❌ Allows Claude to perform any operation (including deleting files)	Not recommended
Whitelist configuration	✅ Only allows explicit operations, prompts error for overreach	Recommended

While whitelist configuration is initially complex to set up, once generated it's automatically reused and more secure.

🎒 Prerequisites

Before starting, please confirm:

[x] Completed Installation and Configuration (start/installation/)
[x] Completed Initialize Factory Project (start/init-project/)
[x] Installed Claude Code: https://claude.ai/code
[x] Project directory initialized (.factory/ directory exists)

Check if Claude Code is installed

Run the following command in terminal to confirm:

bash

claude --version

If prompted "command not found", please install Claude Code first.

Follow Along

Step 1: Initialize Project (Auto-generate Permissions)

Why: factory init will automatically generate .claude/settings.local.json, containing the complete permission whitelist.

Run in project directory:

bash

# Create new directory and enter
mkdir my-factory-project && cd my-factory-project

# Initialize Factory project
factory init

You should see:

✓ Factory project initialized!
✓ Claude Code is starting...
  (Please wait for window to open)

Claude Code window will automatically open and display the following prompt:

Please read .factory/pipeline.yaml and .factory/agents/orchestrator.checkpoint.md,
start the pipeline, help me transform product idea fragments into a runnable application,
next I will input idea fragments. Note: The skills/ and policies/ files
referenced by Agent need to first search the .factory/ directory, then search the root directory.

What happened:

Created .factory/ directory, containing pipeline configuration
Generated .claude/settings.local.json (permission whitelist)
Automatically launched Claude Code and passed startup prompt

Step 2: Verify Permission Configuration

Why: Confirm the permission file has been correctly generated to avoid runtime permission issues.

Check the generated permission file:

bash

# View permission file content
cat .claude/settings.local.json

You should see (partial content):

json

{
  "permissions": {
    "allow": [
      "Read(/path/to/project/**)",
      "Write(/path/to/project/**)",
      "Glob(/path/to/project/**)",
      "Bash(git add:*)",
      "Bash(git commit:*)",
      "Bash(npm install:*)",
      "Bash(npx prisma generate:*)",
      "Skill(superpowers:brainstorming)",
      "Skill(ui-ux-pro-max)",
      "WebFetch(domain:github.com)",
      "WebFetch(domain:npmjs.org)"
    ]
  },
  "features": {
    "autoSave": true,
    "telemetry": false
  }
}

Path explanation

Paths in permissions are automatically adjusted based on your operating system:

Windows: Read(//c/Users/...) (both lowercase and uppercase drive letters supported)
macOS/Linux: Read(/Users/...) (absolute path)

Step 3: Start Pipeline in Claude Code

Why: Claude Code has configured permissions and can directly read Agent definitions and Skill files.

In the opened Claude Code window, input your product idea:

I want to create a mobile expense tracking app to help young people quickly record daily expenses,
avoid overspending at the end of the month. Main features are recording amount, selecting categories (food, transport, entertainment, other),
viewing this month's total expenses.

You should see:

Claude Code will execute the following steps (automatically):

Read .factory/pipeline.yaml
Read .factory/agents/orchestrator.checkpoint.md
Start Bootstrap phase, structuring your idea into input/idea.md
Pause after completion, waiting for your confirmation

Checkpoint ✅: Confirm Bootstrap phase completed

bash

# View generated structured idea
cat input/idea.md

Step 4: Continue Pipeline

Why: After each phase completes, manual confirmation is required to avoid error accumulation.

In Claude Code, reply:

Continue

Claude Code will automatically enter the next phase (PRD), and repeat the "execute → pause → confirm" process until all 7 phases are completed.

Use factory run to restart

If the Claude Code window is closed, you can run in terminal:

bash

factory run

This will redisplay the Claude Code execution instructions.

Step 5: Cross-platform Permission Handling (Windows Users)

Why: Windows path permissions require special handling to ensure Claude Code can correctly access project files.

If you're using Windows, factory init will automatically generate permissions supporting drive letters:

json

{
  "permissions": {
    "allow": [
      "Read(//c/Users/yourname/project/**)",
      "Read(//C/Users/yourname/project/**)",
      "Write(//c/Users/yourname/project/**)",
      "Write(//C/Users/yourname/project/**)"
    ]
  }
}

Checkpoint ✅: Windows users verify permissions

powershell

# PowerShell
Get-Content .claude\settings.local.json | Select-String -Pattern "Read|Write"

If you see both //c/ and //C/ path formats, it's correctly configured.

Checkpoint ✅

After completing the above steps, you should be able to:

[x] Find the .claude/settings.local.json file
[x] See the complete permission whitelist (including Read/Write/Bash/Skill/WebFetch)
[x] Successfully start the Bootstrap phase in Claude Code
[x] View input/idea.md to confirm idea has been structured
[x] Continue executing the pipeline to the next phase

If you encounter permission errors, please see the "Troubleshooting" section below.

Troubleshooting

Problem 1: Permission Blocked

Error message:

Permission denied: Read(path/to/file)

Causes:

Permission file generation failed or path is incorrect
Claude Code is using old permission cache

Solutions:

Check if permission file exists:

bash

ls -la .claude/settings.local.json

Regenerate permissions:

bash

# Delete old permission file
rm .claude/settings.local.json

# Re-initialize (will regenerate)
factory init --force

Restart Claude Code to clear cache.

Problem 2: `--dangerously-skip-permissions` Warning

Error message:

Using --dangerously-skip-permissions is not recommended.

Causes:

.claude/settings.local.json not found
Permission file format error

Solutions:

Check permission file format (JSON syntax):

bash

# Verify JSON format
python -m json.tool .claude/settings.local.json

If prompted with syntax error, delete the file and rerun factory init.

Problem 3: Windows Path Permissions Not Working

Error message:

Permission denied: Read(C:\Users\yourname\project\file.js)

Causes:

Drive letter path missing in permission configuration
Incorrect path format (Windows needs to use //c/ format)

Solutions:

Manually edit .claude\settings.local.json, add drive letter paths:

json

{
  "permissions": {
    "allow": [
      "Read(//c/Users/yourname/project/**)",
      "Write(//c/Users/yourname/project/**)"
    ]
  }
}

Note that both drive letter cases must be supported (//c/ and //C/).

Problem 4: Skills Permission Blocked

Error message:

Permission denied: Skill(superpowers:brainstorming)

Causes:

Required Claude Code plugins not installed (superpowers, ui-ux-pro-max)
Plugin version incompatibility

Solutions:

Add plugin marketplace:

bash

# Add superpowers plugin marketplace
claude plugin marketplace add obra/superpowers-marketplace

Install superpowers plugin:

bash

claude plugin install superpowers@superpowers-marketplace

Add ui-ux-pro-max plugin marketplace:

bash

claude plugin marketplace add nextlevelbuilder/ui-ux-pro-max-skill

Install ui-ux-pro-max plugin:

bash

claude plugin install ui-ux-pro-max@ui-ux-pro-max-skill

Rerun the pipeline.

Factory will automatically try to install plugins

The factory init command will automatically try to install these plugins. If it fails, please install manually.

Summary

Permission whitelist is more secure than --dangerously-skip-permissions
factory init automatically generates .claude/settings.local.json
Permission configuration includes file operations, Git, build tools, database, web operations and other categories
Cross-platform support: Windows uses //c/ paths, Unix uses absolute paths
Manual plugin installation: If automatic installation fails, you need to manually install superpowers and ui-ux-pro-max in Claude Code

Next Up

In the next lesson, we'll learn OpenCode and Other AI Assistants.
You will learn:
How to run Factory pipelines in OpenCode
Integration methods for other AI assistants like Cursor, GitHub Copilot
Permission configuration differences for different assistants

Appendix: Source Code Reference

Click to expand source code locations

Last updated: 2026-01-29

Function	File Path	Lines
Permission configuration generation	`cli/utils/claude-settings.js`	1-292
Auto-launch Claude Code	`cli/commands/init.js`	119-147
AI assistant detection	`cli/commands/run.js`	105-124
Claude Code instruction generation	`cli/commands/run.js`	138-156
Cross-platform path handling	`cli/utils/claude-settings.js`	14-67

Key functions:

generatePermissions(projectDir): Generate complete permission whitelist, including Read/Write/Bash/Skill/WebFetch operations
generateClaudeSettings(projectDir): Generate and write .claude/settings.local.json file
launchClaudeCode(projectDir): Launch Claude Code window and pass startup prompt
detectAIAssistant(): Detect currently running AI assistant type (Claude Code/Cursor/OpenCode)

Key constants:

Windows path pattern: Read(//c/**), Write(//c/**) (supports lowercase and uppercase drive letters)
Unix path pattern: Read(/path/to/project/**), Write(/path/to/project/**)
Skills permissions: 'Skill(superpowers:brainstorming)', 'Skill(ui-ux-pro-max)'

Permission whitelist categories:

File operations: Read/Write/Glob (supports wildcards)
Git operations: git add/commit/push/pull etc. (complete Git command set)
Build tools: npm/yarn/pnpm install/build/test/dev
TypeScript: tsc/npx tsc/npx type-check
Database: npx prisma validate/generate/migrate/push
Python: python/pip install (for ui-ux-pro-max)
Testing: vitest/jest/test
Factory CLI: factory init/run/continue/status/reset
Docker: docker compose/ps/build/run
Web operations: WebFetch(domain:github.com) etc. (specified domain whitelist)

Claude Code Integration: Configure Permissions to Run Pipeline | AI App Factory Tutorial ​

What You'll Learn ​

The Problem ​

When to Use This ​

Core Concepts ​

Permission Whitelist Categories ​

Why not use --dangerously-skip-permissions? ​

🎒 Prerequisites ​

Follow Along ​

Step 1: Initialize Project (Auto-generate Permissions) ​

Step 2: Verify Permission Configuration ​

Step 3: Start Pipeline in Claude Code ​

Step 4: Continue Pipeline ​

Step 5: Cross-platform Permission Handling (Windows Users) ​

Checkpoint ✅ ​

Troubleshooting ​

Problem 1: Permission Blocked ​

Problem 2: --dangerously-skip-permissions Warning ​

Problem 3: Windows Path Permissions Not Working ​

Problem 4: Skills Permission Blocked ​

Summary ​

Next Up ​

Appendix: Source Code Reference ​

Claude Code Integration: Configure Permissions to Run Pipeline | AI App Factory Tutorial

What You'll Learn

The Problem

When to Use This

Core Concepts

Permission Whitelist Categories

Why not use `--dangerously-skip-permissions`?

🎒 Prerequisites

Follow Along

Step 1: Initialize Project (Auto-generate Permissions)

Step 2: Verify Permission Configuration

Step 3: Start Pipeline in Claude Code

Step 4: Continue Pipeline

Step 5: Cross-platform Permission Handling (Windows Users)

Checkpoint ✅

Troubleshooting

Problem 1: Permission Blocked

Problem 2: `--dangerously-skip-permissions` Warning

Problem 3: Windows Path Permissions Not Working

Problem 4: Skills Permission Blocked

Summary

Next Up

Appendix: Source Code Reference