Guides
Declare portable intent
Use portable fields to describe what the Resource needs, then let each AI Channel choose its closest native shape.
Portable fields
Declare intent once.
Use Playbook fields to describe model class, effort, access, and capabilities. Do not bake Claude, Codex, or Cursor terms into the source doc unless you are explaining a channel-native override.
modelTier
reasoning | coding | fastDeclares the model class you want, not a vendor model name. Channels map it to what they support.
effort
low | medium | highDeclares reasoning depth. Codex can render this as model_reasoning_effort; other channels may ignore it.
access
read-only | write | fullDeclares permission posture. Channels translate it into sandbox, approval, or permission settings.
tools
read | edit | shell | web | todo | taskDeclares capability intent. Claude may render native tools; Codex uses access for sandbox behavior.
Example Topic
A small source file can drive several AI Channels.
This Topic stays vendor-neutral. Playbook decides how to render the same Resource into Codex, Claude Code, Cursor, Copilot, Gemini, or OpenCode output.
---
name: frontend-accessibility
title: Frontend Accessibility
description: "Accessibility rules for SUMR UI work. Use before shipping visible React changes."
modelTier: reasoning
effort: medium
access: read-only
tools: [read, web]
---
# Frontend Accessibility
Use semantic controls, preserve keyboard navigation, and verify focus states.
## When to use
Read this before editing visible UI, form controls, dialogs, navigation, or error states.The Topic says reasoning, medium effort, read-only access, and read/web tools. It does not name a specific model or permission file.
Each AI Channel gets the closest native representation: skill metadata, agent profile, command, rule, prompt file, or hook output.
Use channel blocks only for native options that cannot be expressed as portable intent.
Guide map
Choose the decision you are making.
Source model
Place guidance in the Project, group reusable knowledge in Catalogs, and keep Resources reviewable before they become AI Channel output.
2Resource types
Choose Document, Topic, Reference, team-member, command, workflow, or lifecycle by what Playbook should render next.
4Agent instructions
Extract durable rules with agents-md markers instead of hand-editing generated Codex, Claude, or Cursor files.
5Channels and sync
Enable AI Channels in sumr.yaml, validate source docs, preview generated files, then write outputs.