Build fewer agents. Capture better methods.
A practical seven-part test and worked starter package for deciding when a repetitive task should become a reusable skill.
No signup. One top-level folder. Includes SKILL.md, a template, and three release tests.
The operating model: let the agent hold the stable role and let each skill hold one bounded, testable method. Do not turn a vague procedure into a skill until its trigger, inputs, steps, output, quality checks, evidence, and stop rule are explicit.
Before a task becomes a skill, require seven things
Recognizable trigger
When should the method activate?
Defined inputs
What must be available before work begins?
Repeatable steps
What sequence should it follow?
Required output
What exactly must it produce?
Quality checks
What must it verify before returning the result?
Verifiable evidence
How can a person check the answer?
Stop rule
When must it ask, refuse, or escalate?
A worked weekly-brief example
The package shows how to turn supplied project information into a concise leadership brief while preserving evidence, supported owners and dates, and explicit escalation behavior.
Download the starter packageweekly-work-brief/
SKILL.md
references/test-cases.md
templates/brief-template.md
Need help turning repetitive work into a controlled AI workflow?
Start with the business task, evidence, boundaries, and stop rule before choosing the implementation.
Book an AI Time Back Audit