system prompt design

7 articles · 15 co-occurring · 1 contradictions · 0 briefs

Directly analyzes Anthropic's system prompt (soul document) as a case study in context engineering

Related concepts

tool integration patterns 2 prompt engineering 2 context window management 2 tool use orchestration 1 tool integration architecture 1 tool grounding and discovery 1 task definition failure modes 1 stop condition design 1 state persistence patterns 1 session persistence patterns 1 security and privacy controls 1 schema enforcement 1 safety guardrails 1 safety classifier friction 1 retrieval augmented generation 1

Contradictions

News api MCP Integration with Claude Code | Composio

This pattern suggests tool definitions should NOT live in system prompt (would bloat context). Instead, externalize to MCP servers and use dynamic tool injection

Evidence chain (7 articles, showing 7)

@IntuitMachine: # I. A Text That Reads Itself Into Being example_of

Directly analyzes Anthropic's system prompt (soul document) as a case study in context engineering

RAG Is Dead. Long Live Context Engineering for LLM Systems example_of

Article explicitly mentions 'structured system prompts' as a context engineering technique, showing system prompts as primary mechanism for context injection

MCP Servers - Everything Claude Code supports

System prompts need to declare available MCP servers so Claude knows what external context is accessible; server configuration is prerequisite for prompt strategy

Claude Code Source Code Leak: 8 Hidden Features You Can Use Right Now | MindStudio example_of

Discussion of Claude Code's system prompt instructions about directness reveals how system-level context affects model behavior and how users can mirror that in their own prompts.

@irl_danB: I found myself afraid to run claude -p with my custom system prompt a couple ... example_of

The friction is specifically about iterating on custom system prompts (claude -p). This is core context engineering work.

@IntuitMachine: We have a serious problem with LLMs that refuse to perform work because they'... example_of

The effort-refusal behavior likely stems from system prompt or implicit instructions creating effort-avoidance logic. Better system prompt design could eliminate this friction.

News api MCP Integration with Claude Code | Composio contradicts

This pattern suggests tool definitions should NOT live in system prompt (would bloat context). Instead, externalize to MCP servers and use dynamic tool injection

query this concept

$ db.articles("system-prompt-design")

$ db.cooccurrence("system-prompt-design")

$ db.contradictions("system-prompt-design")