By it-bens
Debugging escalation plugin that automatically consults OpenAI Codex (GPT-5) when stuck after three failed attempts with no progress. Features codex-escalation agent with pre-flight validation, context gathering, and progressive escalation (Codex -> User). Includes /codex-check command for setup verification. Requires Codex CLI and OpenAI account. Uses native MCP server integration for fresh analytical perspective and root cause analysis.
npx claudepluginhub it-bens/claude-code-pluginsUses power tools
Uses Bash, Write, or Edit tools
Share bugs, ideas, or general feedback.
Give Claude Code a second opinion using OpenAI Codex - automatic plan review via hooks
Debug issues systematically with root cause analysis and execution tracing
Debugging specialist for errors, test failures, and unexpected behavior. Use proactively when encountering any issues.
Use when the user asks to run Codex CLI (codex exec, codex resume) or references OpenAI Codex for code analysis, refactoring, or automated editing. Uses GPT-5.2 by default for state-of-the-art software engineering.
Capture mistakes and improvement opportunities with automatic 5 whys analysis
Delegate coding, planning, and debugging to Cursor CLI via ACP. Part of cc-multi-cli-plugin. Requires the `multi` plugin.
A personal collection of skills, agents, MCP servers, hooks, and LSPs to enhance Claude Code with capabilities across software development and beyond.
Add this marketplace to Claude Code:
/plugin marketplace add it-bens/claude-code-plugins
Then browse and install plugins:
/plugin
| Plugin | Description | Category |
|---|---|---|
| prompt-engineering | Expert prompt engineering for Claude 4, GLM 4.7, and Gemini 3 (including Deep Research) with production-ready templates | Productivity |
| python-plan-optimizer | Analyze Python code in planning documents for design principles and improvement opportunities | Development |
| native-tools-enforcer | Enforces use of Claude Code native tools instead of bash equivalents via PreToolUse hook (smart piped grep filtering) | Guardrails |
| explore-with-sonnet-enforcer | Enforces Sonnet model for Explore subagent to prevent lossy Haiku summaries | Guardrails |
| explore-with-opus-enforcer | Enforces Opus model for Explore subagent for maximum exploration accuracy | Guardrails |
| plan-with-sonnet-enforcer | Enforces Sonnet model for Plan subagent to ensure thorough architectural reasoning | Guardrails |
| plan-with-opus-enforcer | Enforces Opus model for Plan subagent for maximum reasoning depth | Guardrails |
| tandoor-recipe-converter | Convert recipe images, PDFs, text, or URLs to Tandoor-compatible JSON with German output | Productivity |
| codex-debugger | Automatic debugging escalation to OpenAI Codex (GPT-5) when stuck after three failed attempts | Development |
| commit-message-generator | Generate and validate conventional commit messages with automatic type/scope detection | Development |
| behavior-diagnostics | Debug AI tooling behavior with root cause analysis and honest self-diagnosis across sessions | Development |
claude-code-plugins/
├── .claude-plugin/
│ └── marketplace.json # Marketplace registry
├── .github/
│ └── scripts/
│ └── setup-bats.sh # BATS test framework setup
├── plugins/
│ └── <plugin-name>/ # Each plugin follows this pattern:
│ ├── .claude-plugin/
│ │ └── plugin.json # Plugin manifest
│ ├── skills/ # Optional: Skills
│ ├── agents/ # Optional: Agents
│ ├── hooks/ # Optional: Hooks
│ └── README.md # Plugin documentation
├── plugin-tests/ # BATS tests for hook scripts
├── build-skill-for-web.sh # Creates Claude Web-compatible ZIPs
├── LICENSE
└── README.md
marketplace.json)The source field in each plugin entry is resolved relative to the marketplace.json file location (not the pluginRoot):
{
"metadata": {
"pluginRoot": "./plugins" // Informational only, not used for resolution
},
"plugins": [
{
"name": "my-plugin",
"source": "./plugins/my-plugin", // Relative to marketplace.json
"category": "development"
}
]
}
Key rule: source must start with ./ and include the full path from .claude-plugin/ to the plugin directory.
plugin.json)Each plugin's plugin.json uses paths relative to its own location. Component discovery (skills, agents, hooks) is automatic from the plugin root.
Skills in this marketplace use the allowed-tools frontmatter field to restrict tool access in Claude Code. However, this field is only supported in Claude Code and must be removed for Claude Web compatibility.
.claude-web MarkerNot all skills are suitable for Claude Web. To enable a skill for web builds, add an empty .claude-web file to the skill directory:
touch ./plugins/<plugin>/skills/<skill>/.claude-web
The GitHub workflow automatically builds ZIP archives only for skills with this marker file.
The build-skill-for-web.sh script creates Claude Web-compatible ZIP archives by stripping Claude Code-specific fields:
./build-skill-for-web.sh ./plugins/<plugin>/skills/<skill>
# Output: <skill-name>.zip
According to Claude Code Skills Documentation:
allowed-toolsis only supported for Skills in Claude Code.
Skills uploaded to Claude Web will ignore this field. The build script ensures clean ZIPs without unsupported fields.