You Can Start Building LLM Skills Before You Know the Whole Shape
6 hours ago
- #LLM skills
- #automation
- #iterative development
- Begin automation efforts with a single useful behavior rather than designing a complex system upfront.
- Define automation clearly by naming specific functions (e.g., scanning, fixing, reviewing) to avoid ambiguity and focus on actionable tasks.
- Create a skill around a narrow scope, such as fixing one violation in one file, to enable immediate testing and iteration based on real use.
- Iteratively refine the skill by observing its performance on actual problems, adjusting descriptions, scope, and outputs to close gaps.
- Allow the skill to grow organically through accretion, adding features only when demanded by practical needs, not pre-designed diagrams.
- Incorporate human checkpoints (e.g., review, approval) as intentional design elements to maintain judgment and reliability.
- Start building immediately with a minimal viable skill, letting its shape emerge from real-world application rather than upfront planning.