Use env !# to be more portable

README.md

@@ -1,6 +1,6 @@
-# ask - AI CLI tool
+# ask - Claude Code CLI wrapper
 
-A lightweight bash script for querying AI models via the OpenRouter API, optimized for direct, executable output.
+A lightweight bash script for querying Claude AI via the Claude Code CLI, optimized for direct, executable output with file editing capabilities.
 
 ## Quick start
 
@@ -12,16 +12,15 @@ cd ask
 chmod +x ask
 sudo cp ask /usr/local/bin/
 
-
-# Make sure you have you OpenRouter API key
-export OPENROUTER_API_KEY="your-api-key-here"
+# Install Claude Code CLI first
+# Visit: https://claude.com/claude-code
 
 # Test it
-> ask remove lines in file1 that appear in file2
+ask "Write a hello world in Python"
 
-grep -vFf file2 file1 > file3 && mv file3 file1
+print("Hello, World!")
 
-[inception/mercury-coder via Inception - 0.66s - 20.9 tok/s]
+[sonnet - 1.23s - 0 tool calls - 12 tokens - $0.0001]
 ```
 
 We also provide a handy install script.
@@ -31,36 +30,24 @@ We also provide a handy install script.
 ### Basic usage
 
 ```bash
-ask ffmpeg command to convert mp4 to gif
+ask "ffmpeg command to convert mp4 to gif"
 ```
 
 ### Model selection
 
 ```bash
-# Default model (Mercury Coder - optimized for code)
-ask find files larger than 20mb
+# Default model (Claude Sonnet)
+ask "find files larger than 20mb"
 
 # Shorthand flags for quick model switching
-ask -c "prompt"  # Mercury Coder (default, best for code)
-ask -g "prompt"  # Gemini 2.5 Flash (fast, general purpose)
-ask -s "prompt"  # Claude Sonnet 4 (complex reasoning)
-ask -k "prompt"  # Kimi K2 (long context)
-ask -q "prompt"  # Qwen 235B (large model)
-
-# Custom model by full name
-ask -m "openai/gpt-4o" "Explain this concept"
-```
+ask -s "prompt"  # Claude Sonnet (default, balanced)
+ask -o "prompt"  # Claude Opus (most capable, complex reasoning)
+ask -h "prompt"  # Claude Haiku (fastest, simple tasks)
 
-### Provider routing
-
-Specify provider order for fallback support:
-
-```bash
-ask --provider "cerebras,together" "Generate code"
+# Custom model by name
+ask -m haiku "What is 2+2?"
 ```
 
-This will try Cerebras first, then fall back to Together if needed.
-
 ### System prompts
 
 ```bash
@@ -71,36 +58,60 @@ ask --system "You are a pirate" "Tell me about sailing"
 ask -r "What is 2+2?"
 ```
 
-### Streaming mode
+### File editing
 
-Get responses as they're generated:
+File editing is **enabled by default** in the current working directory:
 
 ```bash
-ask --stream "Tell me a long story"
+# Claude can read and edit files automatically
+ask "Read README.md and update it with current features"
+ask "Fix the bug in server.js"
+
+# Disable file editing if you only want analysis
+ask --no-edits "Just analyze server.js, don't modify it"
+```
+
+### Running commands
+
+Claude can run bash commands to gather context:
+
+```bash
+# Claude will run git commands for you
+ask "Look at git diff and write a commit message"
+ask "Run git status and summarize changes"
+
+# Show statistics to see tool usage
+ask --stats "Run git status and summarize changes"
 ```
 
 ### Pipe input
 
 ```bash
+# Pipe as prompt
 echo "Fix this code: print('hello world)" | ask
-cat script.py | ask "Review this code"
+
+# Pipe as context with --context flag
+git diff | ask --context "Write a commit message for this diff"
+cat error.log | ask --context "Explain what caused this error"
 ```
 
 ## Options
 
 | Option | Description |
 |--------|-------------|
-| `-c` | Use Mercury Coder (default) |
-| `-g` | Use Google Gemini 2.5 Flash |
-| `-s` | Use Claude Sonnet 4 |
-| `-k` | Use Moonshotai Kimi K2 |
-| `-q` | Use Qwen3 235B |
-| `-m MODEL` | Use custom model |
-| `-r` | Disable system prompt |
-| `--stream` | Enable streaming output |
-| `--system` | Set custom system prompt |
-| `--provider` | Set provider order (comma-separated) |
-| `-h, --help` | Show help message |
+| `-s` | Use Claude Sonnet (default) |
+| `-o` | Use Claude Opus |
+| `-h` | Use Claude Haiku |
+| `-m MODEL` | Use custom Claude model |
+| `-r` | Disable system prompt (raw model behavior) |
+| `--system TEXT` | Set custom system prompt |
+| `--context` | Pipe stdin as context, combine with prompt argument |
+| `--no-edits` | Disable file editing (enabled by default) |
+| `--permission-mode MODE` | Set permission mode (e.g., acceptEdits) |
+| `--allowed-tools TOOLS` | Comma-separated list of allowed tools |
+| `--stats` | Show statistics (time, tool calls, tokens, cost) |
+| `--debug` | Show system context, prompts, and command (stderr) |
+| `--help` | Show help message |
 
 ## Common use cases
 
@@ -123,6 +134,29 @@ ask "Python function to calculate factorial"
 cat script.py | ask "Find potential bugs in this code"
 ```
 
+### File editing and code modification
+```bash
+# Claude can read and modify files directly
+ask "Read server.js and add error handling to the API routes"
+ask "Fix the typo in README.md line 42"
+ask "Refactor utils.py to use type hints"
+
+# Disable editing for read-only analysis
+ask --no-edits "Analyze the performance of this algorithm in sort.js"
+```
+
+### Git workflow
+```bash
+# Claude can run git commands and analyze changes
+ask "Look at git diff and write a commit message"
+ask "Run git status and tell me what needs attention"
+ask "Review the last 3 commits and summarize changes"
+
+# See what Claude is doing with --stats
+ask --stats "Run git log --oneline -5 and summarize the recent changes"
+# Output: [sonnet - 2.15s - 1 tool calls - 89 tokens - $0.0023]
+```
+
 ### Quick answers
 ```bash
 # Calculations
@@ -142,19 +176,34 @@ ask "List all Python files" | ask "Generate a script to check syntax of these fi
 # Use with other tools
 docker ps -a | ask "Which containers are using the most memory?"
 
+# Debug mode to see what's happening
+ask --debug "Why is my build failing?"
 ```
 
 ## Requirements
 
 ### Dependencies
 - `bash` - Shell interpreter
-- `curl` - HTTP requests to OpenRouter API
 - `jq` - JSON parsing for API responses
 - `bc` - Performance metrics calculation
-
-### API access
-- OpenRouter API key (get one at [openrouter.ai](https://openrouter.ai))
-- Set as environment variable: `OPENROUTER_API_KEY`
+- Claude Code CLI - The underlying AI engine
+
+### Installation
+1. Install Claude Code CLI from [claude.com/claude-code](https://claude.com/claude-code)
+2. Clone this repository
+3. Make the script executable: `chmod +x ask`
+4. Copy to your PATH: `sudo cp ask /usr/local/bin/`
+
+## Features
+
+- **Direct executable output** - Optimized for command generation and piping
+- **File editing** - Claude can read and modify files in your working directory
+- **Command execution** - Claude can run bash commands to gather context
+- **Smart system context** - Automatically detects OS, tools, git repos, and project type
+- **Model selection** - Choose between Sonnet, Opus, or Haiku
+- **Statistics** - Track tool usage, tokens, and costs with `--stats`
+- **Debug mode** - See exactly what prompts are being sent
+- **Caching** - System context is cached for 24 hours for performance
 
 ## License