This is one of those repos that makes you do a double-take. Ego Debugger is an AI agent skill (and Codex plugin) designed to help people craft relationship-safe replies — essentially treating interpersonal communication like a debuggable system, where your ego is the bug and the reply is the patch.

The premise is wonderfully nerdy. Instead of asking an LLM "rewrite this message to be nicer," the skill appears to instrument the reply pipeline: it inspects what you're about to send, identifies ego-driven failure modes (defensiveness, one-upmanship, passive-aggression, the urge to "win"), and suggests revisions that preserve your intent without lighting the conversation on fire. The fact that it ships as both an agent skill and a Codex plugin suggests the author wants it available inside whatever environment you happen to be writing in — Slack drafts, email, code review comments, GitHub PR replies.

Why it's interesting:

• Skills as a delivery format — most "communication coach" tools are standalone apps. Packaging this as an agent skill means it composes with whatever else your assistant is doing.
• Narrow, opinionated scope — "relationship-safe replies" is a sharper framing than generic "tone improvement," and narrow tools tend to actually work.
• The debugger metaphor — treating ego as a debuggable subsystem is a genuinely useful reframe for engineers who find emotional regulation easier to approach as a technical problem.

Who would benefit: engineers doing high-stakes code reviews, open source maintainers handling spicy issue threads, anyone who has ever hit "send" on a Slack message and immediately regretted it, and folks experimenting with the Claude/Codex skills ecosystem who want a concrete example of a non-coding skill.

At zero stars on day one, it's a fresh experiment — worth a look if only to study how the author scoped a behavioral-coaching tool into a skill manifest.