skill-creator

To create AI skills, you define the skill intent, triggering conditions, and output formats in a SKILL.md file with YAML metadata. You then optimize the skill by benchmarking its performance against baseline runs and refining the activation descriptions to improve triggering accuracy.

Benchmarking for prompt engineering involves creating test prompts, comparing skill-assisted runs against baselines, grading assertions, and aggregating performance metrics. This process uses Python-based evaluation workflows to analyze performance variance and review qualitative outputs.

Trigger optimization improves skill activation accuracy by generating realistic trigger and non-trigger queries to measure how well a description activates a skill. You refine the activation description to improve triggering accuracy without overfitting to specific query patterns.

Yes, you can use Python-based evaluation and reporting workflows to test behavior with baseline comparisons. The skill structure supports bundling optional scripts and assets to grade assertions, aggregate metrics, and analyze performance variance across iterations.

Structuring a new skill requires a SKILL.md file containing YAML metadata, along with optional scripts, references, and assets. This structure supports progressive disclosure by defining skill intent, instructions, bundled resources, and operational entry points.

You should avoid overfitting description optimization when your trigger queries are too narrow to generalize across realistic user inputs. If your activation description becomes overly tuned to specific test prompts, it may fail to trigger appropriately for varied agent workflows.