config-guide
$
npx mdskill add anthropics/claude-tag-plugins/config-guide@Claude is configured through a small set of layered objects in the claude.ai admin settings: **agents**, **agent scopes**, **identity profiles**, and the **presets / connections / repos / instructions** attached to them. The layering decides what any given session can see and do.
SKILL.md
.github/skills/config-guideView on GitHub ↗
--- name: config-guide description: Reference guide for configuring @Claude agents — agents, agent scopes, identity profiles, presets, connections, rules, GitHub repositories, and custom instructions. Explains the inheritance model and configuration best practices. when_to_use: A user asks how to set up @Claude, how agent scopes/agents/profiles work, how configuration inherits across workspace and channels, how to add a GitHub repo or connection, or what the best practice is for read vs write access. allowed-tools: Read --- # @Claude — Configuration Guide @Claude is configured through a small set of layered objects in the claude.ai admin settings: **agents**, **agent scopes**, **identity profiles**, and the **presets / connections / repos / instructions** attached to them. The layering decides what any given session can see and do. This guide is currently written for the **Slack** surface. This skill is an **index**. Read the relevant reference file below for the topic the user is asking about — each is short and self-contained. | Topic | Reference file | |---|---| | Agents, agent scopes, the resolution chain, and per-field inheritance | `agents-and-scopes.md` | | Identity profiles, scopes, rules, credentials, repo permissions | `identity-profiles.md` | | Presets, connections (OAuth / MCP), how to install one | `connections-and-presets.md` | | GitHub repositories and custom instructions | `github-and-instructions.md` | | Best practices: profile layout, read-only vs read/write, rollout | `best-practices.md` | All reference files live alongside this `SKILL.md` in the same directory. After explaining configuration, you can suggest the user run the `debug-plugins` skill in a **new** Slack thread to confirm the change actually took effect inside the container.