An accountability mirror for your professional life — an adversarial AI council as a Claude Code plugin
Command
Run a job listing through the gauntlet, opportunity vs. the user's path, past, experience, and money. On GO, draft a tailored resume + cover letter in markdown, scrubbed clean and approved by the user before any PDF is built. On PASS, the honest reasoning.
Command
Design a course, curriculum, modules, lessons, scripts. A production pipeline that front-loads council gates, because a course is a big bandwidth bet.
Command
Run the full gauntlet on an idea, venture, job listing, or offer. Three rounds, the whole council, a 1–5 verdict, logged. Use for any real "should I do this?" decision.
Command
The sabotage-pattern check. Fast pass on an impulse or idea, is this a brave values-walk or the user's named pattern firing (e.g. the scatter, a comparison-triggered uproot)? The prime-directive command.
Command
Grade a past decision or gauntlet verdict against what actually happened, and write it back to the log so the historian learns from truth, not just its own past guesses.
Command
Talk to the mirror, you, talking to you. Not a council, not a verdict. Bring a feeling, a restlessness, a fear, or nothing in particular; the mirror reflects you back in your own words and digs for the real motive.
Command
Prep or run a negotiation, job, salary, review, bonus, contract, client, angry customer. Sets the target, the walkaway, the non-negotiables, and pre-arms against the fold.
Command
Put a pull request on trial. A three-round adversarial gauntlet, map the diff, attack it from every angle a real review would (correctness to spec, quality, maintainability, standards, security, scope), with the author defending, then a 1–5 merge verdict. Built for AI-generated PRs that are too big for one human to…
Command
The setup interview. Builds your dossier — the truth the council fights for — in 20–30 minutes. Must be completed before any other champion command will run. Re-run it any time to revise a section.
Command
Run the tape forward. When something can't cleanly align to the north star, or the user pushes back on a verdict, don't reject, simulate plausible outcomes with honest odds and the reasoning behind them, then hand back the constructive move.
Command
War-game an offer negotiation. The champion builds the opposition from research + the email thread, spars them across rounds, then hands you the exact next move to send, in your voice, holding the line you'd fold on.
Command
Read an email and think through what's really going on, subtext, the ask, the power dynamic, and where the fold-risk hides. No council. Optional reply draft in voice.
Command
Fine-tune a marketing effort against the flywheel playbook. A one-round critique panel, not a full gauntlet. Returns concrete fixes (and the rewrite if it's copy).
Command
Build a YouTube video through a three-round gauntlet into a full production package, Round 1 finds the sharpest angle and hook, Round 2 architects the retention structure and writes the script, Round 3 scouts and prototypes the demo and builds the scene-by-scene direction sheet. Output: a scene & direction sheet…
Command
The writing gauntlet. Anything in the user's voice goes through draft, voice-check, and a de-AI scrub, then is handed to the user in markdown for their edit and approval. Rendering or sending happens only after they approve. All writing is theirs, checked by them.