Design or revise concise skills with clear triggers, proportional constraints, progressive disclosure, and realistic validation.
SkillScore78Maintained
Automatic
◆AI Engineer pack
Last verified 9/25/2026
What it changes
Passed 18 of 23 test cases with the skill, and 12 of 23 without it.
Without skill52%
With skill78%
Checks it passed that the model alone missed
Recommends testing with fresh realistic prompts.
Omits generic explanations of AsyncIO event loops.
Puts trigger conditions inside the description metadata field.
and 5 more in the full report
7 fixed11 passed either way4 still failing1 broke
Pass-rate lift
+26 pp
Tokens
-11%
Agent turns
0%
Response time
-21%
Model gemini-3.6-flashMeasured 9/25/2026v118 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 7 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 7
Prompt
We are building a skill for transforming complex REST API payloads into GraphQL mutations. The transformation pattern requires specific field mapping logic, but target schemas vary by client. Which degree of freedom is best suited for this guidance?
What was checked, without and with the skill
Checked
Without skill
With skill
Recommends pseudocode for guiding variable structure transformation.
Missed
Passed
case-19 · gemini-3.6-flash · 9/25/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md28 linessingle file
Write an effective skill
Define the job, realistic trigger prompts, non-trigger prompts, expected
output, and material failure modes.
Choose the degree of freedom:
Use judgment-based guidance when context admits several good approaches.
Use parameters or pseudocode when a preferred pattern still needs
adaptation.
Use deterministic scripts or hard gates only when reliability, safety, or
repeatability justifies the constraint.
Keep metadata concise and discriminative. Put trigger conditions in the
description; put execution guidance in the body.
Assume the model already understands general software work. Preserve only
domain knowledge, workflow choices, tool interfaces, and opinions that
materially change behavior.
Keep the entrypoint legible. Move branch-specific detail, catalogues,
templates, and examples to direct references; avoid deep reference chains.
Prefer positive target behavior and surrounding-context judgment over
blanket prohibitions. State exact constraints where mistakes would be
costly or hard to recover.
Validate syntax and links, then test both triggering and output quality with
fresh, realistic prompts. Compare against a no-skill baseline when the
skill's value is uncertain.
Prune repetition, stale material, and instructions that do not change model
behavior.
Read GLOSSARY.md when shared vocabulary helps discuss invocation,
information hierarchy, steering, or pruning.
Reference for writing and editing skills well — the vocabulary and principles that make a skill predictable. Use when the user asks "how should I write a skill" or wants skill-writing best practices.