claude-clilisted
Install: claude install-skill Git-Fg/taches-principled
# claude-cli
The `claude-cli-wrapper` plugin exposes a flat, agent-callable surface over the
Claude Code CLI. This skill is a user-facing map of the 6 tools and when to
reach for which one. For the MCP design principles these tools embody (flat
schema, pass-through, output contract), see `mcp-server-design`. For
implementation details, see `mcp-server-implement`.
---
## §1. When this skill fires
**Use this skill when the user says any of:**
- "Run Claude Code on this prompt" / "spawn a Claude Code agent"
- "Continue / resume / fork a Claude Code session"
- "List my Claude Code sessions"
- "Review this PR with Claude Code"
- "Spawn a background Claude agent"
- "Set the model / effort / permissions for Claude Code"
- "Wrap Claude Code as a tool for my agent"
- "Drive Claude Code programmatically"
**DO NOT use this skill for:**
- "How do I design an MCP server" → `mcp-server-design`
- "How do I implement an MCP server in Rust" → `mcp-server-implement`
- "How do I write a good JSON Schema" → `mcp-tool-surface`
- "How do I write a Claude Code hook" / "slash command" → not covered here
## CONTRAST
- NOT for: spawning in-process subagents within the current Claude Code session — use subagent-orchestration
- NOT for: designing an MCP server from scratch — use mcp-server-design
- NOT for: writing a single Claude Code hook — see the official hooks docs
## §2. The 6-tool surface
| Tool | Operational domain | When to use |
|---|---|---|
| `claude_execute` | Run a prompt, get str