Back to directory

Claude Code Prompts

An independently authored collection of prompt templates for building AI coding agents, covering system prompts, tool prompts, agent delegation, memory management and multi-agent coordination.

Claude Code Prompts

Claude Code Prompts is a collection of prompt templates for building AI coding agents, written independently after studying how Claude Code structures its safety rules, tool routing, subagent delegation, and memory management. It is not affiliated with Anthropic.

The collection includes one system prompt covering agent identity, safety rules, code style, tool routing and output format; eleven tool prompts for shell, file read/edit/write, grep, glob, web search and fetch, an agent launcher, asking the user, and plan mode; five agent prompts for roles such as a general-purpose worker, a code explorer, a solution architect, a verification specialist, and a documentation guide; four memory prompts for summarization and consolidation; a multi-worker coordinator prompt; utility prompts for session titles and summaries; commentary on each pattern; and three drop-in Cursor skills.

To use it, a developer browses the pattern files, copies a complete prompt into their own agent configuration, and replaces its placeholder values with their own stack, tool names and risk policy.

More in Agent skills

See all 581
  • Provenance gate

    github.com

    A deterministic authorization gateway that sits between an AI agent and its tools, tracing where each parameter of a side-effect call came from, and blocking the call or asking a human when the source does not check out.

  • A markdown framework of system rules that restates Isaac Asimov's Four Laws of Robotics as a set of prioritized loss calculations for large-language-model agents to follow, published as a discovery file for agents and crawlers to read.

  • Vitruvius

    github.com

    An engineering research agent that runs a discover, read, synthesize, verify and review loop over technical questions, reading sources directly rather than inferring from titles, and publishing its own scored verifier benchmark, weaknesses included.