claude-code-skill

Files (20)

• .gitignore
• LICENSE
• README.md
• SKILL.md
• basic-mcp.ts · examples/
• chat-store.ts · examples/
• mcp_config.example.json
• package.json
• index.ts · src/
• actions.ts · src/mcp/
• client.ts · src/mcp/
• index.ts · src/mcp/
• logger.ts · src/mcp/
• types.ts · src/mcp/
• utils.ts · src/mcp/
• index.ts · src/store/
• indexeddb-storage.ts · src/store/
• persist.ts · src/store/
• sync.ts · src/store/
• tsconfig.json

SKILL.md (first lines)

# Claude Code Skill

Control Claude Code via MCP (Model Context Protocol). This skill unleashes the full power of Claude Code for solanaos agents, including persistent sessions, agent teams, and advanced tool control.


---

## ⚡ Quick Start

```bash
# Start a persistent Claude session for your project
claude-code-skill session-start myproject -d ~/project \
  --permission-mode acceptEdits \
  --allowed-tools "Bash,Read,Edit,Write,Glob,Grep"

# Send a plan (Claude will execute precisely)
claude-code-skill session-send myproject --stream '<精确的执行计划>'

# Send with high effort (ultrathink) for complex reasoning
claude-code-skill session-send myproject --stream --ultrathink 'Refactor the auth module'

# Enter plan mode — Claude creates a plan first, then executes
claude-code-skill session-send myproject --stream --plan 'Implement rate limiting'

# Check progress
claude-code-skill session-status myproject

# Compact session when context gets large
claude-code-skill session-compact myproject

# Switch model mid-session
claude-code-skill session-model myproject opus
```

## 🎯 When to Use This Skill

### Use Persistent Sessions When:
- ✅ Multi-step tasks requiring multiple tool calls
- ✅ Iterative development (write code → test → fix → repeat)
- ✅ Long conversations needing full context
- ✅ Agent needs to work autonomously
- ✅ You want streaming real-time feedback

### Use Direct MCP Tools When:
- ✅ Single command execution
- ✅ Quick file read/write
- ✅ One-off searches
- ✅ No context needed between operations

## 📚 Command Reference

### Basic MCP Operations

```bash
# Connect to Claude Code MCP
claude-code-skill connect
claude-code-skill status
claude-code-skill tools

# Direct tool calls (no persistent session)
claude-code-skill bash "npm test"
claude-code-skill read /path/to/file.ts
claude-code-skill glob "**/*.ts" -p ~/project
claude-code-skill grep "TODO" -p ~/project -c
claude-code-skill call Write -a '{"file_path":"/tmp/test.txt","content":"Hello"}'

# Disconnect
claude-code-skill disconnect
```

### Persistent Sessions (Agent Loop)

#### Starting Sessions

```bash
# Basic start
claude-code-skill session-start myproject -d ~/project

# With custom API endpoint (for Gemini/GPT proxy)
claude-code-skill session-start gemini-task -d ~/project \
  --base-url http://127.0.0.1:8082 \
  --model gemini-2.0-flash

# With permission mode (plan = preview changes before applying)
claude-code-skill session-start review -d ~/project --permission-mode plan

# With tool whitelist (auto-approve these tools)
claude-code-skill session-start safe -d ~/project \
  --allowed-tools "Bash(git:*),Read,Glob,Grep"

# With budget limit
claude-code-skill session-start limited -d ~/project --max-budget 1.50

# Full configuration
claude-code-skill session-start advanced -d ~/project \
  --permission-mode acceptEdits \
  --allowed-tools "Bash,Read,Edit,Write" \
  --disallowed-tools "Task" \
  --max-budget 5.00 \
  --model claude-opus-4-5 \
  --append-system-prompt "Always write tests" \
  --add-dir "/tmp,/var/log"
```

**Permission Modes:**
| Mode | Description |
|------|-------------|
| `acceptEdits` | Auto-accept file edits (default) |
| `plan` | Preview changes before applying |
| `default` | Ask for each operation |
| `bypassPermissions` | Skip all prompts (dangerous!) |
| `delegate` | Delegate decisions to parent |
| `dontAsk` | Never ask, reject by default |

#### Sending Messages

```bash
# Basic send (blocks until complete)
claude-code-skill session-send myproject "Write unit tests for auth.ts"

From x402agent/solana-clawd, MIT.

More from x402agent/solana-clawd