token-reporter

Design documents, requirements analysis, architecture decisions - creates specifications for orchestrator

Run linters, formatters, and type-checkers without installing them using uvx, bunx, npx, and other no-install runners. Includes smart executor script and expert agent.

High-accuracy skill activation (88%+) with AI-analyzed keywords, weighted scoring, synonym expansion, and three-tier confidence routing.

General-purpose programmer agent for code implementation, testing, and debugging across multiple languages (Python, JS/TS, Rust, Go, .NET, C/C++, Swift).

Cross-platform clipboard operations for Claude Code agents. Native macOS/Windows/WSL support plus Linux scripts with Wayland, X11, and history tracking.

Comprehensive validation suite for Claude Code plugins, marketplaces, hooks, skills, and MCP servers. Includes 17 Python validators, 2 expert agents, and 190+ validation rules.

Universal terminal-menu system for Claude Code plugins. Any agent invokes scripts/menu_write.py via Bash (~25ms) to render a JSON menu spec to a tempfile; a bundled Stop hook emits it as systemMessage at main-session turn end — the menu appears in the user terminal exactly when they can type a reply. No subagent fork, no extended-thinking overhead, no token cost to the orchestrator.

Task distribution, agent coordination, progress monitoring - executes architect's plans via subagents

Exports current session segment (since last compaction) with system-reminder stripping -- main conversation, subagent transcripts, sidechains, and debug logs in structured markdown.

Automatically reviews and fixes Claude's code changes using a separate Claude instance in a git worktree after commits. Two-phase pipeline: code correctness (Phase 1) then functionality verification (Phase 2).