User Prompt

# User Prompt — Agent Mode

> Claude Code injects skills, MCP server instructions, deferred tool list, mode flags, and project context as `<system-reminder>` blocks alongside the user's message — **6 content blocks in a single user turn**. The system prompt is identical between agent and plan modes; mode-specific behavior lives in this user-prompt layer.

The user message contains six layers, each as a separate text content block within a single user turn:

| # | Block | Length | Purpose |
|---|-------|--------|---------|
| 1 | Deferred tools list | ~4,350 chars | Names of 117 tools available via `ToolSearch` (schemas not preloaded) |
| 2 | MCP server instructions | ~3,284 chars | Per-server usage guidance (Figma, tldraw, context7…) |
| 3 | Skills reminder | ~4,252 chars | Catalog of 15 invocable skills |
| 4 | **Auto Mode** flag | ~1,209 chars | NEW in v2.1.126 — declares autonomous-execution policy |
| 5 | Project context | ~366 chars | userEmail, currentDate, optional CLAUDE.md |
| 6 | User task | (varies) | The actual user message |

> **Note on agent mode:** The plan-mode reminder is *absent* in agent mode. Plan mode adds a 7th block ("Plan mode is active...") between blocks 4 and 5.

---

## Block 1 — Deferred Tools