claude-code-prompt-analysis

This skill analyzes how Claude Code automatically augments prompts across five distinct mechanisms: CLAUDE.md for project-wide context injection, Output Styles for session-wide system prompt changes, Slash Commands for deterministic workflow triggers, Skills for model-invoked capability extension, and Sub-Agents for isolated conversation delegation. By instrumenting network traffic, it reveals exactly where and how each mechanism injects instructions into API requests, enabling developers and operations teams to understand the injection points and scope of each approach. The skill includes a comparison matrix showing activation methods and scope differences, plus a decision guide for choosing between CLAUDE.md (team standards), Output Styles (personal preferences), and Skills versus Slash Commands (automation vs. control). This technical reference helps teams optimize workflows, structure project context effectively, and understand security implications before deploying AI agents in production.

Review the five mechanisms section to understand each augmentation layer, then consult the comparison matrix to determine which mechanism fits your use case. The decision guide helps you choose between CLAUDE.md (committed standards), Output Styles (session behavior), Skills (autonomous activation), Slash Commands (explicit control), or Sub-Agents (multi-step delegation).