Agent rule pattern

Architecture as narrative

A written walkthrough of how the system works.

26 of 100 projects

The shape of the rule

  1. Follow a request
  2. Name each owner
  3. Explain the boundaries
How to use it

Request lifecycles, data flows, and component boundaries help with cases the rules do not cover. The tradeoff is a longer file.

Follow a request or piece of data through the system. Name who owns each step and why the boundaries exist.

Examples from real projects

3 examples · Pinned source

Show the approval path across processes

An architecture diagram follows Claude Code through MCP, the CLI, the daemon, and the cloud API. It also shows where terminal and graphical approval interfaces join the flow.

CLAUDE.mdL32–37
Source excerpt starting at line 32.
### Architecture Flow```Claude Code → MCP Protocol → hlyr → JSON-RPC → hld → HumanLayer Cloud API                                         ↑         ↑                                    TUI ─┘         └─ WUI```

The seam drawn as a pipeline

Two ASCII flows show how a diff becomes a review document and how an intent becomes a surface projection. Each stage names the module that owns it and the facts a renderer is to consume rather than recompute.

AGENTS.mdL49
Source excerpt starting at line 49.
  parity manifests. Consume them; never re-derive those facts in a renderer.

Describe the content-processing path

The ingestion path runs from content detection and extraction through summarization, embeddings, chunking, and space relationships. The sequence shows where each behavior belongs.

CLAUDE.mdL70–75
Source excerpt starting at line 70.
All content goes through the `IngestContentWorkflow` which handles:- Content type detection and extraction- AI-powered summarization and automatic tagging- Vector embedding generation using Cloudflare AI- Chunking for semantic search optimization- Space relationship management

Projects using this pattern (26)

Browse in directory →
Show 18 more projects
← All agent rule patterns