A Claude Code plugin marketplace with parallel Codex marketplace metadata for the plugins that already map cleanly onto Codex's current plugin model.
This project started as a fork of EveryInc/compound-engineering-plugin.
Add this marketplace to Claude Code:
/plugin marketplace add RBozydar/rbw-claude-codeThen browse and install plugins:
/plugin menuCodex currently reads local marketplaces rather than remote GitHub marketplace
references. This repo now ships a repo-local Codex marketplace at
.agents/plugins/marketplace.json.
To use it:
git clone https://github.com/RBozydar/rbw-claude-code.git
cd rbw-claude-code
codex
/pluginsThe current Codex marketplace exposes:
corepython-backenddeep-research-plus
The guard plugins remain Claude-only for now because current Codex plugin docs describe skills, apps, and MCP configuration, but not hook plugins.
If you use Poetry instead of uv for package management, install from the poetry-variant branch:
/plugin marketplace add RBozydar/rbw-claude-code#poetry-variantThis variant includes enforce-poetry instead of enforce-uv, suggesting poetry run and poetry add commands.
Due to a known Claude Code bug, plugin hooks are matched but not executed. Until this is fixed upstream, you need to manually configure hooks in your settings.
Navigate to the marketplace directory:
cd ~/.claude/plugins/RBozydar/rbw-claude-codeRun the setup script:
# Global hooks (apply to all projects) - default
./scripts/setup-hooks.sh
# Or project-specific hooks (run from within your project)
./scripts/setup-hooks.sh --projectTo verify hooks are active:
/hooksYou should see the configured hooks listed (enforce-uv, conventional-commits, etc.).
Once hooks are installed, a SessionStart hook automatically checks if your
configured hooks are in sync with available plugin hooks. If hooks change
(e.g., after updating the marketplace), you'll see a warning at session start:
====================================================
rbw-claude-code: Hooks are out of sync!
====================================================
Plugin hooks have changed. Run to update:
./scripts/setup-hooks.sh --project
====================================================
You can also manually check sync status:
./scripts/setup-hooks.sh --check # Check global hooks
./scripts/setup-hooks.sh --check --project # Check project hooks| Plugin | Description |
|---|---|
| core | Universal AI development tools: 29 agents, 18 commands, 16 skills |
| python-backend | Python-specific tools: 5 reviewers, 2 commands |
| deep-research-plus | Deep research workflows with optional GDELT MCP support |
| Plugin | Description |
|---|---|
| enforce-uv | Block bare python/pip/pytest commands, enforce uv |
| conventional-commits | Validate conventional commit format |
| python-format | Auto-format Python files with ruff after edits |
| python-typecheck | Run type checking after Python file edits |
| test-reminder | Remind to add tests when creating new Python files |
| Plugin | Description |
|---|---|
| protect-env | Block reading .env files to protect secrets |
| git-safety-guard | Block destructive git commands |
| safety-guard | Block destructive file ops and supply chain attacks |
Codex-compatible plugins in this repo currently are:
corepython-backenddeep-research-plus
Each now ships a .codex-plugin/plugin.json, and the repo marketplace lives at
.agents/plugins/marketplace.json.
Hook-only plugins remain Claude-specific for now. Details: docs/codex-compatibility.md
For Codex-only local generation without hand-maintaining duplicate markdown,
the repo now generates Codex custom agents directly from the Claude agent
markdown and keeps them under .codex/agents/.
Regenerate them from source:
uv run python scripts/generate_codex_agents.pyInstall them into your Codex home:
./scripts/install-codex-agents.shBy default the installer creates symlinks into ~/.codex/agents/ so updates in
this repo propagate without copy drift. Use --copy if you prefer materialized
files.
The core plugin provides language-agnostic AI-powered development tools:
/workflows:plan- Transform features into structured plans/workflows:work- Execute work plans efficiently/workflows:review- Multi-agent code reviews with parallel analysis/workflows:compound- Document solved problems for knowledge compounding
code-simplicity-reviewer- Reviews for unnecessary complexitysecurity-sentinel- Security vulnerability analysisperformance-oracle- Performance analysisarchitecture-strategist- System design reviewpattern-recognition-specialist- Pattern/anti-pattern detectionagent-native-reviewer- Ensures features are agent-accessibledata-migration-expert- Database migration validation
framework-docs-researcher- Framework documentation lookupgit-history-analyzer- Git history analysisrepo-research-analyst- Repository analysisbest-practices-researcher- Industry best practices
| Skill Name | Description |
|---|---|
core:agent-native-architecture |
Build AI agents using prompt-native architecture |
core:compound-docs |
Capture solved problems as categorized documentation |
core:create-agent-skills |
Expert guidance for creating Claude Code skills |
core:file-todos |
File-based todo tracking in todos/ directory |
core:git-ship |
Complete git workflow: commit, push, PR, CI, merge |
core:git-worktree |
Manage Git worktrees for parallel development |
core:instruction-hygiene |
Audit and trim bloated CLAUDE.md/skill/context setups with recurring hygiene checks |
core:skill-creator |
Guide for creating effective skills |
The python-backend plugin extends core with Python-specific capabilities:
kieran-python-reviewer- High-quality Python code reviewskeptical-simplicity-reviewer- Anti-overengineering critiqueml-expert-reviewer- ML/DS/LLM specialized reviewgemini-brainstorm- Second opinion from Geminigemini-reviewer- Alternative code review from Gemini
/pytest-runner- Smart pytest execution with failure analysis/type-check- Intelligent type checking with mypy/pyright
Ensures Claude Code uses uv for all Python operations.
Blocks bare python, pip, pytest commands.
Validates commit messages follow Conventional Commits specification.
Runs uvx ruff format automatically after any Python file edit.
Runs uvx pyright automatically after any Python file edit.
Reminds you to add tests when creating new Python modules.
Blocks reading .env files to prevent exposing secrets to AI.
Blocks destructive git commands that could cause data loss:
git reset --hard,git reset --mergegit push --force,git push -fgit checkout -- <files>,git restore <files>git clean -f,git branch -Dgit stash drop,git stash cleargit reflog expire,git filter-branch
Allows safe alternatives: git checkout -b, git push --force-with-lease,
git clean -n.
Blocks destructive file operations and supply chain attacks:
rm -rfoutside temp directoriesfind -delete,shred,truncatecurl | bash,wget | sh(supply chain attacks)bash -cwith destructive commands (bypass detection)- Reading
.envfiles (allows.env.example,.env.sample,.env.template)
- Claude Code with plugin support
uvinstalled for Python-related plugins- GitHub CLI (
gh) for PR operations - Gemini CLI for Gemini agents (optional)
MIT