Hunt a bug the way a senior debugger would: reproduce it on demand before touching any code, read what the error actually says, test hypotheses one at a time, fix the cause instead of the symptom, then prove the fix against the original failure. Exists because the default is the opposite — guess a fix, declare it done, and leave the real bug alive under a patch. MANDATORY TRIGGERS: 'bug hunter', 'run the bug-hunter', 'hunt this bug'. STRONG TRIGGERS (use when something is actually broken): 'it's still broken', 'you said it was fixed and it isn't', 'this keeps happening', 'why is this failing', 'debug this properly', 'find the real cause'. The rule that pays for everything: never fix what you haven't reproduced.
Turn 'I want to build X' into a staged, written plan that a cheaper model — or a future session that remembers nothing — can execute. Asks the questions that change the build first, offers real options with trade-offs, then slices the work into stages that each end with something you can see working, every step carrying its own verification. MANDATORY TRIGGERS: 'build planner', 'run the build-planner', 'plan this build'. STRONG TRIGGERS (use when a real project is on the table): 'plan out how to build this', 'break this down for me', 'what's the plan for X', 'I want to build X, where do I start', 'write a plan a smaller model can follow'. The point: the smartest model available thinks once, and everything after just executes.
Stress-test a business idea, plan, or big decision with the yes-man switched off. Steelmans the idea first so the critique isn't lazy, then attacks it properly: who actually pays, the one number everything hangs on, why you, what's the cheapest real-world test — and returns a straight verdict it won't soften when you push back. MANDATORY TRIGGERS: 'honest advisor', 'run the honest-advisor', 'stress test this idea'. STRONG TRIGGERS (only when a business idea or a money/time decision is on the table — never for code review or technical questions): 'what do you honestly think of this idea', 'poke holes in this', 'would this work as a business', 'should I build this', 'tell me why this fails', 'be brutal about this plan'. Built because AI's default answer to every idea is applause — and applause is the most expensive advice there is.
Start a new project so it doesn't fall apart in month two. Does the first hour properly: version control before the first mistake, secrets out of the code before the first key exists, a CLAUDE.md written at birth, and the thinnest possible version deployed live before any feature gets built. MANDATORY TRIGGERS: 'project setup', 'run project-setup', 'set this project up properly'. STRONG TRIGGERS (use at the start of something new): 'new project', 'let's start building X', 'set up the foundations', 'start this the right way'. Built for the month-two disasters: the lost work, the leaked key, the app that's never actually been deployed.
Check everything you're about to ship the way a security reviewer would — tuned to the holes vibe-coded apps actually get hacked through. Reads the real code (routes, auth checks, database rules, keys, uploads, AI features), builds a ranked list of what an attacker could genuinely do, explains each one in plain English, and offers to fix them worst-first. MANDATORY TRIGGERS: 'security sweep', 'run the security sweep', 'sweep this before I ship'. STRONG TRIGGERS (use when tied to this codebase): 'is this safe to ship', 'could this get hacked', 'check my app for security holes', 'am I leaking anything', 'review this before launch'. Run it before anything real goes live — and again whenever a feature touches money, logins, or user data.