Claude Code Plugin Blueprint
A practical guide to structuring plugins, configuring manifests, and organizing Claude Code components.
The material documents plugin paths, MCP, hooks, environment-variable references, and use of CLAUDE_PLUGIN_ROOT; however, it lacks user-confirmation requirements, least-privilege guidance, data-flow disclosure, and rollback guidance for auto-started MCP servers, automatic hooks, API-key propagation, and external notifications, so 10 points are deducted. Official provenance and Apache-2.0 licensing improve traceability but do not establish safety.
The directory, manifest, and component guidance is organized and includes troubleshooting. However, there are no tests for key paths, no compatibility matrix, and no defined diagnostic output for abnormal input. The claim that no restart is required conflicts with troubleshooting advice to restart for reloads, so the static score is conservatively limited to 6.
Trigger scenarios are specific and cover plugin creation, scaffolding, directory organization, plugin.json, path variables, and component configuration, with partially indicated inputs and boundaries. However, Claude Code version limits, platform differences, Chinese-language support, and mainland-China reachability are not addressed, and several behaviors are unverified, so 6 points are deducted.
SKILL.md, README, references, and examples provide useful progressive disclosure, naming guidance, versioning advice, maintenance steps, licensing information, and publishing guidance. However, the skill lacks its own changelog, dependency versions, installation instructions, maintainer contact, and explicit compatibility governance, so 5 points are deducted. Repository-level Apache-2.0 licensing and Anthropic maintenance provide only limited governance evidence.
The skill supplies directly reusable directory templates, manifest examples, organization patterns, and troubleshooting guidance, so its core scaffolding value is plausible. However, examples lack static evidence of tests or real execution; advanced examples depend on external MCP servers, cloud credentials, and scripts, requiring users to validate actual Claude Code behavior. Nine points are deducted and the score remains within the static ceiling.
The source is auditable and includes structured documentation, examples, and some repository-level CI maintenance signals. However, there are no committed tests covering this skill's key paths, independent review, or execution evidence; several behavior claims remain author documentation only. Six points are deducted.
- Do not treat documented auto-discovery, hook, or MCP behavior as verified; test against the target Claude Code version before installation.
- The advanced examples involve Kubernetes, Terraform, GitHub, Slack, PagerDuty, Datadog, and environment credentials; review permissions, secret exposure, external communication, and rollback plans before deployment.
- No Chinese-language guidance or mainland-China reachability guarantee is provided, so Chinese users may need additional explanation and localization testing.
What it does & when to use it
This skill from Anthropic’s official plugin directory explains how to create and maintain Claude Code plugins. It covers the standard layout and configuration of commands, agents, skills, hooks, MCP servers, and helper scripts. It also explains automatic discovery, portable references through ${CLAUDE_PLUGIN_ROOT}, naming conventions, maintenance practices, and troubleshooting. It is aimed at developers building or organizing Claude Code plugins.
Explains the standard plugin directory layout; documents required and recommended plugin.json fields and custom component paths; describes the file formats, locations, and discovery behavior for commands, agents, skills, hooks, MCP servers, and scripts; provides naming guidance, ${CLAUDE_PLUGIN_ROOT} examples, minimal and full-featured plugin patterns, and troubleshooting steps for loading, path resolution, discovery, and plugin conflicts.
- A developer creating a Claude Code plugin from scratch and needing a standard directory layout.
- A maintainer configuring plugin.json metadata, component paths, or versioning.
- A plugin author organizing commands, agents, skills, hooks, and MCP configuration.
- A Claude Code user troubleshooting missing components, path errors, or discovery failures.
Pros & cons
- Covers the main Claude Code plugin component types and their standard locations.
- Includes examples for plugin.json, hooks, and MCP configuration.
- Clearly explains how default directories relate to custom manifest paths.
- Includes naming, organization, maintenance, and troubleshooting guidance.
- Its guidance is specific to Claude Code’s plugin architecture rather than a generic Agent Skills packaging standard.
- The source provides no standalone installation command, test suite, or verified platform matrix.
- It is instructional guidance and does not state that it creates or edits plugin files automatically.
How to install
The skill is located at plugins/plugin-dev/skills/plugin-structure/SKILL.md in anthropics/claude-plugins-official. The collection can be installed through Claude Code’s plugin system with /plugin install plugin-dev@claude-plugins-official, or discovered through /plugin > Discover. The source does not document a standalone installation method for this skill.
How to use
In Claude Code, ask a question matching its documented triggers, such as “create a plugin”, “scaffold a plugin”, “understand plugin structure”, “set up plugin.json”, or “use ${CLAUDE_PLUGIN_ROOT}”. It then provides guidance on layout, manifest configuration, component organization, and plugin architecture practices.