Pako Public Plugin

PaKo Public Plugin Marketplace

A public Claude Code plugin marketplace with shared skills for automated code quality workflows.

Installation

claude plugin add https://github.com/rashpile/pako-pubmarket

Available Skills

external-code-review

Automated multi-phase code review that catches bugs, security issues, and over-engineering by combining Claude with an independent external model (Codex, Gemini, or Pi). Finds issues, applies fixes, and commits them — iterating until clean.

How it works:

1. First Review — 5 specialized agents run in parallel (quality, implementation, testing, simplification, documentation), each analyzing your diff from a different angle
2. External Review — an independent model (Codex, Gemini, or Pi) reviews the same changes, and Claude evaluates its findings to filter false positives
3. Final Review — focused pass on critical/major issues only

Why use it:

• Multi-model review catches issues a single model misses
• Parallel agents cover more ground (security, tests, docs, over-engineering) in one pass
• Iterative loop: finds issues → fixes → re-verifies until clean
• Works with Codex, Gemini, Pi, or any combination — auto-detects what's installed

Usage: Ask Claude Code for an "external code review" or "multi-agent review". For a faster pass covering critical issues only, ask for a "quick external code review".

See the full skill documentation for configuration, CLI options, and customization.

Agent prompts are sourced from ralphex by Umputun.

plan-review

Multi-round plan review that validates implementation plans before code is written. Specialized agents run in configurable rounds — catching architectural flaws, over-engineering, convention violations, and completeness gaps at the design stage.

How it works:

1. Round 1 — architect + simplifier agents run in parallel, reviewing strategy and complexity
2. Synthesis — orchestrator merges findings, flags conflicts between reviewers
3. Round 2 — conventions + completionist agents run in parallel, informed by round 1 findings
4. External Review (optional) — independent model (Codex, Gemini, or Pi) provides holistic assessment
5. Annotation — plan file annotated inline with HTML comments per finding

Why use it:

• Catch issues at the plan stage — cheaper than fixing them in code
• Multi-round review: later agents see earlier findings, catching cross-cutting concerns
• Configurable agent ordering via N_name.txt filename convention
• Interactive --config mode to customize agents per project
• Works with any plan format: markdown specs, structured plans, or freeform text

Usage: Ask Claude Code for a "plan review" or "review my plan". To customize agents, ask to "configure plan review" or use --config.

See the full skill documentation for configuration, agent customization, and the naming convention.

second-opinion

Get a second opinion from external AI models. Gathers current task context with full code and sends it to Codex, Gemini, Claude CLI, or saves for manual copy-paste. External responses are critically evaluated — never relayed blindly.

Mode	Use case
consult	Send context to external CLI → evaluate response → assessed report
gather	Save context file for manual copy-paste to any external AI chat
review	External design/code review with focused review prompt

Usage: Ask for a "second opinion", "discuss externally", or "get external review".

second-opinion-nda

Strictly sanitized external consultation for NDA-restricted projects. All code is converted to pseudocode, company names are obfuscated. Manual copy-paste only — no external CLI tools are used (they cannot be reliably restricted from reading local files).

Key safety features:

• No external CLI tools — manual copy-paste is the only delivery method
• Every outbound message requires explicit user approval
• Real code is never included — pseudocode only
• Inbound responses are validated for information leaks
• External advice is critically evaluated, never trusted blindly

Usage: Ask for a "nda second opinion" or "safe consult".

See full documentation: second-opinion | second-opinion-nda

hook-control

Gates Claude Code hooks by project. Instead of a tool firing on every session everywhere, hook-control lets you restrict it to specific projects — or silence it entirely — without touching settings.json by hand.

How it works:

Every wrapped hook in settings.json points to hook.py instead of the original command. On each hook call, hook.py reads config.yaml, checks project rules (includes/excludes), and either delegates to the original command or exits silently. A sync hook on SessionStart automatically re-wraps any new hooks added by tool updates.

Why use it:

• Fire hooks only in the projects that need them
• Disable a tool globally or per-hook without uninstalling it
• Per-call logging with optional debug output capture
• Automatic re-wrapping when tools add new hooks

Usage: Say "wrap peon-ping", "init hook-control", "hook status", or "remove peon-ping". Scripts are auto-installed to ~/.claude/hooks/hook-control/ on first use.

See the full skill documentation for configuration, project rules, and logging.

Structure

.claude-plugin/marketplace.json    # Marketplace definition
plugin/
├── .claude-plugin/plugin.json     # Plugin metadata
└── skills/                        # Skill definitions
    ├── external-code-review/      # Multi-phase AI code review
    ├── plan-review/               # Multi-round plan review
    ├── second-opinion/            # External AI consultation (open)
    ├── second-opinion-nda/        # NDA-safe consultation (manual only)
    └── hook-control/              # Project-scoped Claude Code hook gating

License

MIT