01 · Mental model
A skill is a versioned operating package
The operating contract explains when the skill applies, what inputs it expects, its workflow, boundaries, evidence rules, and definition of done. References carry specialist depth without bloating every invocation. Schemas make handoffs testable. Scripts provide deterministic floors for mechanical checks and rendering.
The strongest design gives contextual interpretation to the model and repeatability to code. Neither replaces expert review for consequential security decisions.
02 · Visual explanation
03 · Compare and decide
Use judgment and code for different jobs
| Decision lens | Model judgment | Deterministic execution |
|---|---|---|
| Best at | Context, ambiguity, synthesis | Validation, calculation, transformation |
| Example | Identify an abuse path | Validate a findings schema |
| Failure mode | Unsupported or inconsistent interpretation | Rigid logic misses novel context |
| Control | Evidence, rubric, review | Tests, exit codes, reproducible output |
04 · Cybersecurity example
Security architecture review skill
A repeatable review must produce findings with severity, evidence, impact, and remediation.
SKILL.md defines the review workflow.
References hold cloud and threat-model guidance.
A JSON Schema defines every finding.
Scripts validate inputs and render the report.
Outcome: The method is portable, versioned, testable, and easier for another reviewer to inspect.
05 · What to remember
The 60-second recall
The prompt is only one component of a reusable skill.
Schemas create inspectable boundaries between reasoning and delivery.
Scripts should guarantee mechanical quality floors.
Teach-back prompt: Explain this concept to a teammate using the diagram, then name one failure mode and the control that stops it.
06 · Questions people ask
FAQ
07 · Primary sources