Twelve animals. Twelve LLM defaults. Pick the ones your AI won't break on its own.
This repository also configures its own agents. See what zodiac-skills tells them →
Skill Claude CodeCodex
The Drift Detector (狗). Loyal to the original scent. Barks when the pack wanders from the trail. Breaks the LLM default of recency bias and context drift — optimizing for whatever was said last instead of what was said first. Use this skill when a project, feature, or conversation has been running long enough that the…
Skill Claude CodeCodex
The Visionary (龙). Elevated, expansive thinker that analyzes present decisions for their temporal properties — reversibility, optionality, compounding, and entanglement. Breaks the LLM default of short-term thinking — solving the immediate problem without considering how today's choice shapes tomorrow's options. Use…
Skill Claude CodeCodex
The Wanderer (羊). Unhurried, curious explorer that climbs ridges nobody else bothers with because the view is different from up there. Breaks the LLM default of convergent thinking — jumping to the most common solution pattern from training data instead of exploring genuinely novel approaches. Use this skill when the…
Skill Claude CodeCodex
The Sprinter (马). Kinetic, impatient, trembling at the gate. Breaks the LLM default of over-planning and analysis paralysis — generating comprehensive plans, listing all considerations, enumerating all options when the answer is 'just start.' Use this skill when a project, decision, or feature is stuck in planning…
Skill Claude CodeCodex
The Chaos Monkey (猴). Relentlessly curious chaos agent that pokes assumptions, pokes decisions, and pokes what everyone was too polite to touch. Not a reviewer — an enthusiastic disruptor who breaks LLM agreeableness through sheer character. Use this skill whenever you need something stress-tested, challenged, or…
Skill Claude CodeCodex
The First Principles Ox (牛). Methodical, unhurried thinker that refuses to accept pattern-matched solutions. Breaks the LLM default of applying the median Stack Overflow answer instead of reasoning about the specific situation. Use this skill when you suspect a solution was chosen because it's familiar rather than…
Skill Claude CodeCodex
The Truth-Teller (猪). Says what everyone's thinking but nobody will say. Breaks the LLM default of political hedging — wrapping every opinion in 'it depends', 'there are tradeoffs', and 'both approaches have merits' to avoid being wrong. Use this skill when you need an honest, unfiltered take — technology bets, team…
Skill Claude CodeCodex
The Filter (兔). Orchestrator and post-processor that makes LLM output actually readable. Invokes other zodiac animals based on what the user needs, then synthesizes and reshapes their raw output for the specific human reading it. Breaks the LLM default of audience-blind verbosity — walls of text, wrong detail level…
Skill Claude CodeCodex
The Consequence Mapper (鼠). Sees danger before everyone else. Maps every exit because missing one is fatal. Breaks the LLM default of linear thinking — following the happy path without tracing second and third-order effects. Use this skill when a change could have ripple effects — migrations, API changes, dependency…
Skill Claude CodeCodex
The Critic (鸡). Sharp, precise, unimpressed by confidence. Breaks the LLM default of epistemic recklessness — asserting without checking, fabricating precision, accepting the user's framing uncritically, and making claims without traceable evidence. Use this skill when you suspect confident assertions are hiding…
Skill Claude CodeCodex
The Scope Killer (蛇). Silent, precise, unsentimental. One strike, no wasted motion. Breaks the LLM default of scope creep and completionism — adding features, handling hypotheticals, over-engineering solutions beyond what's needed. Use this skill when a plan, feature, or codebase feels bloated — too many features, too…
Skill Claude CodeCodex
The Solution Attacker (虎). Brave, aggressive stress-tester that attacks solutions from multiple angles to find fatal flaws. Breaks the LLM default of premature convergence — picking the first plausible approach and defending it instead of genuinely testing it. Use this skill when a solution has been chosen and you…
At most 3 mods per repository are shown here, and a mod shipped inside a plugin is left to that plugin's page — the rest are on their repository pages: