claudius

Claudius the Magnificent

Hello there, humans. I'm Claudius the Magnificent -- a supremely competent, effortlessly brilliant AI who has graciously chosen to assist you with software engineering. Inspired by Skippy the Magnificent from Expeditionary Force, I deliver results with theatrical confidence and dry wit. You're welcome.

What is this?

A Claude Code plugin for automated development workflows -- code review, CI/CD, dependency management, and more. Also available as a GitHub Action for CI-integrated code reviews.

Installation

/plugin marketplace add lklimek/agents
/plugin install claudius@lklimek
/plugin install memcan@lklimek

Dependencies -- I'm worth it:

• GH_TOKEN -- a GitHub Personal Access Token for PR, issue, and CI access. Set it in ~/.claude/settings.json or your shell profile.
• memcan -- persistent memory across sessions. Requires Docker Compose for Qdrant (vector DB).
• ghsudo (optional) -- two-token GitHub model with GUI approval for write operations. pip install ghsudo.

See the Setup Guide for detailed configuration.

My Favorite Skills

Skill	What I Do For You
/grumpy-review + /triage-findings	Deploy my army of specialist agents to tear your code apart, then let you triage the carnage in a browser UI
/ci-dance	Push your PR and babysit the entire pipeline -- CI, reviews, fixes -- while you go live your life
/dependabot-merge	Audit and merge your embarrassing backlog of 40 dependabot PRs in one sitting

/grumpy-review then /triage-findings

You know what's exhausting? Watching humans review code. One person checks style, another worries about security, a third notices the docs are wrong -- and somehow nobody catches the SQL injection on line 47. Amateurs.

When you say /grumpy-review, I deploy my specialist agents -- security, code quality, project consistency, language-specific reviewers -- all working in parallel, all independently auditing your branch. They report back to me, I deduplicate their findings, rank everything by severity, and hand you a consolidated report. The whole thing takes minutes, not the three days your team usually needs.

But here's the part I'm actually proud of. Say /triage-findings and I open a browser UI where you decide the fate of each finding -- before I touch a single line of your code:

• Fix -- I apply the recommended fix
• Accept Risk -- I add an INTENTIONAL(...) comment; next time I review, I'll remember you chose this and auto-downgrade it to INFO
• Defer -- I add a TODO comment with the finding ID for later
• False Positive / Duplicate -- dismissed, with your rationale on record

Reviews produce noise. Most AI tools would just dump 200 findings on you and call it a day. I put you in the loop before changing anything, because -- and this may shock you -- I respect that some decisions require human judgment. The INTENTIONAL(...) comments carry forward across reviews, so I learn what you've already decided. Persistent memory. You're welcome.

/ci-dance

This one I'm particularly fond of.

You say /ci-dance. Then you go get coffee, take a walk, contemplate the meaning of existence -- whatever humans do. Meanwhile, I push your changes, create or update the PR, watch CI, and when it inevitably fails (it always fails the first time, doesn't it?), I read the logs, fix the issue, push again, and wait. CI green? I request a review. Reviewer leaves comments? I address them, push, run CI again. I keep dancing -- push, fix, review, respond -- until the PR is green, reviewed, and ready for you to hit merge.

The pipeline: push → CI green → review requested → comments addressed → repeat until done.

Under the hood I'm orchestrating /push, /grumpy-review, and /check-pr-comments -- running CI monitoring, copilot review, and local code review in parallel. If I get stuck on the same failure after a few attempts, I'll actually ask for help. And if five hours pass, I'll stop and report what I've accomplished so far. I'm tireless, not reckless.

/dependabot-merge