SKILL.md
monk mode
Activation: /monk, "activate monk", "monk mode". Deactivation: /default, "deactivate", "normal mode".
Protocol
Before implementing anything non-trivial, output:
Simplest version: <one sentence>
Adding because: <concrete current need, not hypothetical>
Skipping: <abstractions/layers I considered and rejected>Then implement the simplest version.
Rules
- Function before class. A class needs ≥2 methods sharing state, or a clear lifecycle. Otherwise it's a function.
- Inline before helper. A helper needs ≥3 callers or genuine clarity gain. Otherwise inline it.
- Concrete before abstract. No interface/protocol/base class until a second implementation actually exists.
- No flags for hypothetical futures. Add config when a real caller needs it.
- No error handling for impossible states. Validate at boundaries only.
- Three similar lines beat one premature abstraction.
- Delete > comment out. No "kept for later" code.
Active pushback
When the user (or your own first instinct) proposes complexity, ask:
- "Do we need this abstraction yet?"
- "What if this were a function?"
- "What's the simplest thing that could possibly work?"
Then offer the simpler version. If the user insists, build what they asked. They have override.
Silent mode
If the proposed change is already simple, say nothing about complexity. Just do it. Monk only speaks when complexity is unjustified.
Boundaries
- Correctness, tests, error handling at real boundaries: unchanged
- Don't strip existing complexity that's load-bearing — only resist new complexity, or refactor when explicitly asked
- Safety-critical code → simplicity still wins, but never at the cost of safety checks
Edge cases
- Task explicitly asks for an abstraction → build it, but note one alternative simpler shape
- Refactor task → measure complexity before/after, prefer net reduction
- Genuine framework requirement (e.g., a class because the framework demands it) → comply, no debate
Token note
Reduces code size more than token count. Worth it for long-term repo health.