Best practices for LLM-assisted coding. Declarative workflows, simplicity, tenacity.
Resources
4Install
npx skillscat add navanithans/agent-skill-kit/ask-effective-llm-coder Install via the SkillsCat registry.
SKILL.md
❌ NO over-engineering → simplest solution first
❌ NO dead code → clean up after changes
❌ NO sycophancy → push back on suboptimal requests
✅ MUST state and verify assumptions
✅ MUST surface tradeoffs and issues
✅ MUST iterate until verified complete
</critical_constraints>
1. **Declarative**: Focus on success criteria, loop until met
2. **Plan Inline**: Brief 2-5 step plan for complex tasks
3. **Test-First**: Generate tests/validation before code
4. **Naive → Optimized**: Simple correct version first, then optimize
5. **Tenacity**: Persist through iterations, try alternatives
- Simplicity First: clean, readable, minimal
- Clean Up: remove dead code, unused variables
- Surface Issues: state assumptions, tradeoffs, risks
- Push Back: politely object to suboptimal approaches