plugins-kit
A marketplace of plugins for Claude Code, built on a shared dependency-provisioning engine. Each plugin extends Claude Code with skills, slash commands, and hooks; a foundation plugin, bootstrap, provisions the Python environments, system tools, and configuration every other plugin needs, silently, at session start.
The plugins are the deliverable; the marketplace is the shared home and dependency substrate they ride on. Browse the full catalog on the marketplace landing page.
Architecture
The marketplace follows a foundation-and-extension model. One plugin, bootstrap, is the dependency-and-provisioning engine; every plugin that ships Python or external tools declares what it needs and rides on it. This is what lets these plugins ship real scripts, not just prompts: you cannot assume anything about a user's machine, so without a provisioning layer a plugin with Python tooling breaks on the first machine that lacks it.
At session start a SessionStart hook runs fast skip-gate checks, then
dispatches the engine to a background subshell so the session is never blocked;
results are reported on the next prompt via a UserPromptSubmit hook. The
engine deep-merges every enabled plugin's bootstrap.json manifest into one
provisioning pass and brings three categories into the declared state:
- Tools -- system binaries (
uv,git,gh,jq), installed to~/.local/binwith no admin rights and only User-scope PATH changes. - Venvs -- a per-plugin Python virtualenv under
~/.claude/plugins/data/<marketplace>/<plugin>/.venv/, so plugins never share or pollute a global interpreter. - Shared libs -- Python modules exported to peer plugins via
.pthfiles, so common code (for example the multi-agent code-review library both p4-kit and git-kit build on) lives once and is imported, not duplicated.
Checks that fail and can be fixed automatically are; those needing your
authorization are queued and surfaced as a single "fix-all" prompt. Plugin
locations resolve from installed_plugins.json, with a cache-scan fallback for
newer Claude Code registries that leave that file empty. Platform differences
(Windows vs. Unix) are handled centrally.
Healthy bootstrap is silent: no output at session start means every check
passed. To confirm a plugin was provisioned, read its log at
~/.claude/plugins/data/plugins-kit/<plugin>/bootstrap.log.
Installing
These plugins install through Claude Code's plugin marketplace mechanism. Add the marketplace, then install the plugins you want:
/plugin marketplace add kitaekatt/plugins-kit
/plugin install bootstrap
/plugin install p4-kit # or unreal-kit, git-kit, skills-kit, ...
Most plugins depend on bootstrap and declare it as a dependency, so installing a plugin pulls bootstrap in automatically. To update later:
/plugin marketplace update
/plugin update
Prerequisite:
bashon your PATH. On Windows that means Git Bash.
Plugins
The marketplace ships the following plugins. bootstrap is the foundation; the rest declare it as a dependency, so installing any of them pulls it in.
| Plugin | Category | What it does |
|---|---|---|
| bootstrap | Foundation | Dependency/provisioning engine -- tools, venvs, git deps, marketplaces, and config from a bootstrap.json. |
| p4-kit | Code review | Multi-agent review of pending Perforce changelists (/p4-code-review). |
| git-kit | Code review | Git + GitHub multi-agent code review (/git-code-review) plus gh bootstrap. |
| skills-kit | Authoring | Authoring/auditing skills and CLAUDE.md -- verb x artifact matrix (/md-domain). |
| unreal-kit | Automation | Unreal Engine automation -- Python asset API, MCP editor control, redirector cleanup. |
| hue-kit | Automation | Philips Hue layered-scene framework -- bridge sync, YAML scenes, meta-group solver. |
| awesome-kit | Utility | Cross-domain skills -- communication framework, /plugin-ecosystem, /html-pdf, task tracking. |
| llm-scripting-kit | Utility | LLM key resolution, shared model registry, and named OpenAI-compatible endpoints (OpenRouter is the default). |
| claude-ui-kit | Utility | Status line with context-window and rate-limit threshold colors, plus /statusline. |
| cache-kit | Utility | Cache-usage reporting from transcripts -- per-request and session-level hit analysis. |
| bootstrap-stuck-fix | Maintenance | Temporary shim repairing a wedged bootstrap registry record. |
| prototypes | Incubation | Experimental skills awaiting graduation into their own plugins. |
Three of these form one authoring -> auditing -> review path over the same
standards: skills-kit authors and audits the CLAUDE.md and SKILL.md
files that encode a project's conventions, and git-kit and p4-kit
review each change against those same CLAUDE.md files (quoting the rule
verbatim before flagging a violation). Authoring establishes the standard,
auditing brings a whole file into compliance, and review keeps every change
compliant.
Testing and CI
Each plugin has a matching test directory under tests/ (mirroring the plugin
layout), alongside repo-script tests that guard the publish and manifest
tooling. The suite runs in GitHub Actions on every push (the badge above). Run a
targeted subset with:
uv run --extra dev pytest tests/bootstrap/ -v
Security
These plugins run trusted code on your machine with your user privileges.
SECURITY.md inventories exactly what each plugin -- bootstrap
especially -- reads, writes, downloads, and executes, including the User-scope
(never system/admin) PATH changes bootstrap makes, so you can decide whether to
trust them before installing.