codegpt-ai 2.10.0 → 2.12.0

package/README.md

@@ -6,169 +6,168 @@
 npm i -g codegpt-ai
 ```
 
-Then type `ai`.
+Then type `code`.
 
 ```
-  ██████╗ ██████╗ ██████╗ ███████╗ ██████╗ ██████╗ ████████╗
- ██╔════╝██╔═══██╗██╔══██╗██╔════╝██╔════╝ ██╔══██╗╚══██╔══╝
- ██║     ██║   ██║██║  ██║█████╗  ██║  ███╗██████╔╝   ██║
- ██║     ██║   ██║██║  ██║██╔══╝  ██║   ██║██╔═══╝    ██║
- ╚██████╗╚██████╔╝██████╔╝███████╗╚██████╔╝██║        ██║
-  ╚═════╝ ╚═════╝ ╚═════╝ ╚══════╝ ╚═════╝ ╚═╝        ╚═╝
+  ╔══════════════════════════════════════════════════╗
+  ║                                                  ║
+  ║      C o d e   G P T   v2.0                      ║
+  ║      local ai · powered by ollama                 ║
+  ║                                                  ║
+  ╚══════════════════════════════════════════════════╝
 ```
 
 ## What is it?
 
-A fully-featured AI chat CLI that runs locally on your machine using [Ollama](https://ollama.com). No API keys needed. No cloud. No subscription.
+CodeGPT is a full-featured AI CLI that runs entirely on your machine using [Ollama](https://ollama.com). 123 commands, 8 AI agents, 26 tools, no cloud needed.
 
-- **80+ slash commands** — chat, code, search, export, and more
-- **8 AI agents** — coder, debugger, reviewer, architect, pentester, explainer, optimizer, researcher
-- **29 AI tool integrations** — Claude, Codex, Gemini, Copilot, Cline, and more
-- **Multi-AI system** — swarm, vote, race, team chat with multiple AIs
-- **AI training** — train your own custom Ollama models from conversations
-- **Security** — PIN lock, audit log, sandboxed tools, shell blocklist
-- **Works anywhere** — Windows, macOS, Linux, Termux (Android)
+**Works on Windows, macOS, Linux, and Termux (Android).**
 
 ## Install
 
-### npm (recommended)
 ```bash
 npm i -g codegpt-ai
-ai
+code
 ```
 
-Works with just Node.js. If Python is installed, you get the full 80+ command experience. Without Python, you get a lightweight Node.js chat client.
+No Python? No problem — a built-in Node.js chat client runs automatically. With Python installed, you get the full 123-command experience.
+
+### Other install methods
 
-### pip
 ```bash
-pip install -e git+https://github.com/CCguvycu/codegpt.git#egg=codegpt
+# pip
+pip install git+https://github.com/CCguvycu/codegpt.git
 ai
-```
-
-### PowerShell (Windows)
-```powershell
-irm https://raw.githubusercontent.com/CCguvycu/codegpt/main/install.ps1 | iex
-```
 
-### Termux (Android)
-```bash
+# Termux (Android)
 curl -sL https://raw.githubusercontent.com/CCguvycu/codegpt/main/install-termux.sh | bash
+code
+
+# Windows (PowerShell)
+irm https://raw.githubusercontent.com/CCguvycu/codegpt/main/install.ps1 | iex
 ```
 
 ## Requirements
 
-- **Node.js 16+** (for npm install) OR **Python 3.10+** (for pip install)
-- **Ollama** — install from [ollama.com](https://ollama.com), then `ollama pull llama3.2`
-- Or connect to a remote Ollama server with `/connect IP`
-
-## Quick Start
+- **Node.js 16+** or **Python 3.10+**
+- **Ollama** — [ollama.com](https://ollama.com) then `ollama pull llama3.2`
+- Or connect to a remote Ollama with `/connect IP`
 
-```bash
-# Install
-npm i -g codegpt-ai
+## Features
 
-# Pull a model
-ollama pull llama3.2
+### 123 Commands
 
-# Start chatting
-ai
-```
+| Category | Commands |
+|----------|----------|
+| **Chat** | `/new` `/save` `/load` `/copy` `/regen` `/edit` `/history` `/export` |
+| **Model** | `/model` `/temp` `/think` `/tokens` `/params` `/compact` |
+| **AI Agents** | `/agent` `/all` `/vote` `/swarm` `/team` `/room` `/spectate` |
+| **AI Lab** | `/lab bench` `/lab chain` `/lab prompt` `/race` `/compare` |
+| **Tools** | `/tools` `/bg` `/split` `/grid` 26 AI CLI integrations |
+| **Connect** | `/connect` `/server` `/qr` `/scan` `/disconnect` |
+| **Files** | `/file` `/run` `/code` `/shell` `/browse` `/open` |
+| **Memory** | `/mem` `/train` `/pin` `/search` `/fork` `/rate` |
+| **Skills** | `/skill` `/skills` `/auto` `/cron` |
+| **Comms** | `/broadcast` `/inbox` `/dm` `/monitor` `/hub` |
+| **System** | `/github` `/weather` `/spotify` `/volume` `/sysinfo` |
+| **Security** | `/pin-set` `/lock` `/audit` `/security` `/permissions` |
+| **Profile** | `/profile` `/persona` `/usage` `/setname` `/setbio` |
 
-Type `/help` inside for all commands.
+### 8 AI Agents
 
-## Features
+Specialized agents with custom system prompts:
 
-### Chat
-| Command | Description |
-|---------|-------------|
-| `/new` | New conversation |
-| `/save` | Save conversation |
-| `/load` | Load saved conversation |
-| `/model` | Switch model |
-| `/persona` | Switch personality (hacker, teacher, roast, architect, minimal) |
-| `/think` | Toggle deep reasoning mode |
-| `/file path` | Read a file into context |
-| `/run` | Execute last code block |
-| `/compact` | Summarize conversation to save context |
-
-### AI Agents
-```
-/agent coder build a REST API
+```
+/agent coder build a REST API with auth
 /agent debugger why does this crash
-/agent reviewer check this code
+/agent reviewer check this code for bugs
 /agent architect design a microservices system
+/agent pentester find security vulnerabilities
+/agent optimizer make this code faster
+/agent explainer explain kubernetes simply
+/agent researcher deep-dive into WebSockets
 ```
 
-8 specialized agents: `coder` `debugger` `researcher` `reviewer` `architect` `pentester` `explainer` `optimizer`
+### Multi-AI System
 
-### Multi-AI
 ```
 /all what's the best database?       # All 8 agents answer in parallel
 /vote Flask or FastAPI?              # Agents vote with consensus
-/swarm build a CLI password manager  # 6-agent pipeline
-/team claude codex                   # Group chat with 2 AIs + you
-/race explain recursion              # Race all models for speed
+/swarm build a CLI password manager  # 6-agent collaborative pipeline
+/team claude codex                   # Group chat: you + 2 AIs
+/room coder reviewer architect       # Chat room with 3+ AIs
+/spectate claude gemini debate AI    # Watch AIs debate without you
 ```
 
-### AI Tools (29 integrations)
+### 26 AI Tool Integrations
+
+Launch any AI CLI from CodeGPT. Auto-installs on first use. All sandboxed.
+
 ```
-/tools                    # See all tools
-/claude                   # Launch Claude Code
-/codex                    # Launch Codex
-/gemini                   # Launch Gemini CLI
-/split claude codex       # Side-by-side split screen
-/grid claude codex gemini cline  # 2x2 grid
+/claude    /codex     /gemini    /copilot   /cline
+/aider     /shellgpt  /llm       /litellm   /opencommit
+/gorilla   /chatgpt   /cursor    /ollama    /jq
+/vercel    /netlify   /supabase  /railway   /wrangler
 ```
 
-All tools auto-install on first use. Sandboxed for security.
+Split screen multiple tools:
 
-### AI Training
 ```
-/rate good                # Rate a response
-/train collect            # Save chat as training data
-/train build mymodel      # Create custom Ollama model
-/model mymodel            # Use your trained model
+/split claude codex                  # Side by side
+/grid claude codex gemini cline      # 2x2 grid
 ```
 
-### Remote Connect
+### Custom Skills
+
+Create your own commands — like OpenClaw's self-extending skills:
+
 ```
-/connect 192.168.1.100    # Connect to PC's Ollama
-/qr                       # Show QR code to connect from phone
-/server                   # Check connection status
+/skill poet Write all responses as poetry
+/auto a brutal code reviewer that finds every bug
+/cron 5m /weather                    # Scheduled tasks
 ```
 
 ### Security
-```
-/pin-set                  # Set login PIN
-/audit                    # View security log
-/security                 # Security dashboard
-```
 
-### Integrations
+- **Permission system** — asks before every action with risk level (CRITICAL/HIGH/MEDIUM/LOW)
+- **PIN lock** — SHA-256 hashed, auto-locks after 10min idle
+- **Sandbox** — non-coding tools run in isolated directories, API keys stripped
+- **Rate limiting** — blocks rapid-fire command spam
+- **Input sanitization** — strips null bytes and control characters
+- **Shell blocklist** — blocks dangerous commands + injection patterns
+- **Audit log** — every action logged
+- **Pre-commit hook** — blocks secrets from being committed
+
+### Remote Connect
+
 ```
-/github repos             # Your GitHub repos
-/spotify play             # Control Spotify
-/weather London           # Get weather
-/sysinfo                  # System info
+/connect 192.168.1.100    # Connect to PC's Ollama from phone
+/qr                       # Show QR code to scan
+/server                   # Check connection status
 ```
 
-## CLI Args (non-interactive)
+### Token Counter
+
+Tracks lifetime tokens and messages across all sessions. Visible on startup.
+
+## CLI Args
 
 ```bash
-ai --ask "explain kubernetes"
-ai --agent coder "build a flask app"
-ai --team claude codex "discuss auth"
-ai --tools
-ai --models
-ai --status
-ai doctor
-ai update
-echo "what is rust?" | ai
+code                           # Interactive chat
+code --ask "explain recursion" # One-shot question
+code --agent coder "flask app" # Run an agent
+code --team claude codex "auth"# Two AIs respond
+code --tools                   # List tools
+code --models                  # List Ollama models
+code --status                  # Show status
+code doctor                    # System diagnostics
+code update                    # Self-update
+echo "question" | code         # Pipe input
 ```
 
 ## Aliases
 
-Type faster with 30+ shortcuts:
+30+ shortcuts for fast typing:
 
 | Short | Full | Short | Full |
 |-------|------|-------|------|
@@ -178,33 +177,31 @@ Type faster with 30+ shortcuts:
 | `/m` | `/model` | `/h` | `/help` |
 | `/f` | `/file` | `/con` | `/connect` |
 
-## Architecture
+## Personas
+
+6 built-in AI personalities:
 
 ```
-codegpt/
-  chat.py          6,200 lines   Main CLI (Python)
-  bin/chat.js       300 lines    Node.js fallback
-  bin/ai.js          50 lines    Entry point (routes to Python or Node)
-  ai_cli/                        Package (updater, doctor)
-  app.py                         TUI app (Textual)
-  bot.py                         Telegram bot
-  web.py                         Web app (Flask PWA)
-  server.py                      Backend API
+/persona hacker      # Cybersecurity expert, dark humor
+/persona teacher     # Patient, step-by-step explanations
+/persona roast       # Brutal code reviewer
+/persona architect   # System design, ASCII diagrams
+/persona minimal     # One-line answers, code only
+/persona default     # Standard helpful assistant
 ```
 
-## Data
+## Architecture
 
-All data stored locally at `~/.codegpt/`:
-- `profiles/` — user profile
-- `memory/` — persistent AI memories
-- `security/` — PIN hash, audit log
-- `training/` — training data, custom models
-- `chats/` — saved conversations
+```
+codegpt/
+  chat.py        6,500+ lines   Main CLI (Python)
+  bin/chat.js     300 lines     Node.js fallback
+  ai_cli/                       Package (updater, doctor)
+  app.py                        TUI app (Textual)
+  bot.py                        Telegram bot
+  web.py                        Web app (Flask PWA)
+```
 
 ## License
 
-MIT
-
-## Author
-
-Built by [ArukuX](https://github.com/CCguvycu)
+MIT — Built by [ArukuX](https://github.com/CCguvycu)

package/chat.py

@@ -660,40 +660,91 @@ def ask_permission(action, detail=""):
     risk_color = RISK_COLORS.get(risk, "yellow")
     risk_icon = RISK_ICONS.get(risk, "?")
 
-    # Risk warnings — explain what could happen
-    risk_warnings = {
-        "CRITICAL": "This can execute code, modify your system, or expose data.",
-        "HIGH": "This accesses external services or modifies important data.",
-        "MEDIUM": "This uses resources or changes session settings.",
-        "LOW": "This is a safe operation with minimal impact.",
+    # Detailed risk info per level
+    risk_info = {
+        "CRITICAL": {
+            "warning": "Can execute code, modify files, or expose data.",
+            "examples": "shell commands, code execution, installing software",
+            "tip": "Only allow if you trust the command.",
+        },
+        "HIGH": {
+            "warning": "Accesses external services or modifies data.",
+            "examples": "opening URLs, GitHub access, deleting data",
+            "tip": "Check what it's doing before allowing.",
+        },
+        "MEDIUM": {
+            "warning": "Uses resources or changes settings.",
+            "examples": "running agents, saving files, training models",
+            "tip": "Generally safe for normal use.",
+        },
+        "LOW": {
+            "warning": "Safe operation with minimal impact.",
+            "examples": "switching model, changing persona, forking chat",
+            "tip": "Low risk — consider always-allowing.",
+        },
     }
-    warning = risk_warnings.get(risk, "")
+    info = risk_info.get(risk, risk_info["MEDIUM"])
+    compact = is_compact()
 
-    # Clean minimal prompt — like Claude Code
+    # Permission prompt with full context
+    console.print()
+    console.print(Rule(style=risk_color.replace("bold ", ""), characters="─"))
+    console.print()
+    console.print(Text.from_markup(f"  [{risk_color}]{risk_icon} {risk}[/]  [{risk_color}]{action_desc}[/]"))
     console.print()
-    console.print(Text.from_markup(f"  [{risk_color}]{risk_icon} {action_desc}[/]"))
     if detail:
-        console.print(Text(f"    {detail[:70]}", style="dim"))
-    console.print(Text.from_markup(f"    [{risk_color}]{risk} — {warning}[/]"))
+        console.print(Text.from_markup(f"    [bright_cyan]what:[/]  {detail[:70]}"))
+    if not compact:
+        console.print(Text.from_markup(f"    [dim]risk:[/]  {info['warning']}"))
+        console.print(Text.from_markup(f"    [dim]e.g.:[/]  {info['examples']}"))
+        console.print(Text.from_markup(f"    [dim]tip:[/]   {info['tip']}"))
+    else:
+        console.print(Text.from_markup(f"    [{risk_color}]{info['warning']}[/]"))
     console.print()
 
     try:
         answer = prompt(
-            [("class:prompt", "  Allow? (y)es / (n)o / (a)lways > ")],
+            [("class:prompt", "  (y)es  (n)o  (a)lways  (?) info > ")],
             style=input_style,
         ).strip().lower()
     except (KeyboardInterrupt, EOFError):
+        audit_log("PERMISSION_DENIED", f"{action}: interrupted")
         return False
 
+    if answer in ("?", "info", "help"):
+        # Show detailed info then re-ask
+        console.print()
+        console.print(Text.from_markup(
+            f"  [bold]Action:[/]   {action_desc}\n"
+            f"  [bold]Risk:[/]     [{risk_color}]{risk}[/]\n"
+            f"  [bold]Warning:[/]  {info['warning']}\n"
+            f"  [bold]Examples:[/] {info['examples']}\n"
+            f"  [bold]Tip:[/]      {info['tip']}\n\n"
+            f"  [dim]What each option does:[/]\n"
+            f"    [bright_cyan]y[/]  Allow this one time\n"
+            f"    [bright_cyan]n[/]  Block this action\n"
+            f"    [bright_cyan]a[/]  Always allow (saved, never asked again)\n"
+        ))
+        try:
+            answer = prompt(
+                [("class:prompt", "  Allow? (y/n/a) > ")],
+                style=input_style,
+            ).strip().lower()
+        except (KeyboardInterrupt, EOFError):
+            return False
+
     if answer in ("a", "always"):
         PERMISSION_ALWAYS_ALLOW.add(action)
         save_permissions()
-        console.print(Text(f"  ✓ Always allowed", style="green"))
+        console.print(Text(f"  ✓ Always allowed — {action_desc}", style="green"))
+        audit_log("PERMISSION_ALWAYS", action)
         return True
     elif answer in ("y", "yes", ""):
+        audit_log("PERMISSION_GRANTED", action)
         return True
     else:
-        print_sys("Denied.")
+        console.print(Text(f"  ✗ Blocked — {action_desc}", style="red"))
+        audit_log("PERMISSION_DENIED", action)
         return False
 
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codegpt-ai",
-  "version": "2.10.0",
+  "version": "2.12.0",
   "description": "Local AI Assistant Hub — 80+ commands, 29 tools, 8 agents, training, security",
   "author": "ArukuX",
   "license": "MIT",