AI Engineering

Token Coach analyzes a Claude Code or Codex setup and provides personalized token-efficiency coaching with measured numbers. Use it before building something new, when an existing setup feels slow, or for a quick health check.

Token Optimizer audits a Claude Code or Codex setup for context-window waste, then fixes it and measures the savings. Use when context feels tight.

Token Optimizer audits your OpenClaw setup for token waste, context bloat, and cost optimization opportunities. Use it to detect ghost tokens and get config fixes with dollar savings.

Better Harness reviews the outer coding-agent harness for lifecycle controls, repeated work, project feedback, agent assets, session outcomes, repair planning, durable reports, finding-bound fixes, and manual direct fixes. Invoke only via slash command to keep sessions, project, and agent assets independent.

Handoff prepares a bounded handoff document, recording evidence without copying private content, for use between AI agent sessions.

Harness Skill Creator bootstraps the smallest harness-oriented skill from an existing repository, workflow, or evaluation chain with evidence-backed references and validation gates.

Add Subscription subscribes to a skill event and injects context before or after a skill runs, with optional conditions such as fileExists, gitBranch, envSet, envEquals, or fileContains.

List Skill Bus Subscriptions reads global and project skill-bus config files and displays all active subscriptions with merge status and condition stacking. Use it when you need to see which skills are subscribed and what conditions would fire for a given skill.

MCP Builder guides the creation of high-quality Model Context Protocol (MCP) servers that let LLMs interact with external services through well-designed tools. Use it when building MCP servers to integrate external APIs or services in Python or Node/TypeScript.

Reflecting on Sessions reviews skill-bus telemetry at natural transition points to identify uncovered skills and condition skips. It suggests subscription improvements so an agent can optimize skill coverage and injection during future sessions.

Skill Bus Help shows the skill-bus overview, status, and available commands. Use it as a quick reference for all skill-bus features.

Open Ralph Wiggum wraps any supported AI coding agent in an autonomous loop that repeats a prompt until a completion promise appears in the agent output. Use it when you want an AI coding agent to self-correct and keep iterating on a task without manual intervention.