DEMO / RECORDING
Overview & Result
MIKE highlights Cobus Greyling’s unofficial Jev showcase and skills library. Its route skill maps Jev decisions to workflow or model lanes, its guardrail skill screens inputs and outputs, and its fan-out skill asks independent questions over one state in a single request. Application code controls the actions.
How Jev fits in the loop
- jev-route uses Choice and confidence to select a handling lane or model tier; code maps the result to the actual model or action.
- jev-guardrail asks independent hazard questions and a harm score, then code applies pass, review or block policies.
- jev-fanout sends multiple independent Choice, Score and Noul questions over the same state in one request. It is not arbitrary parallel task execution; the questions cannot see one another’s answers.
How to reproduce
- Open the repository linked by MIKE in his reply. The implementation is by Cobus Greyling, not MIKE.
- Read skills/jev-route/SKILL.md, skills/jev-guardrail/SKILL.md and skills/jev-fanout/SKILL.md before adapting a workflow.
- Follow the repository’s setup. Offline fixtures are available; live judgments require your TypeSafe credentials. Validate thresholds on your own examples.
Why this build matters
Provides a practical starting point for common decision patterns without inventing a new integration for every workflow.
Limitations
- An unofficial community resource, not a TypeSafe-endorsed package.
- Repository contents inspected; skills have not been independently installed or benchmarked by JevForAgents.
- Routing and guardrail thresholds are repository defaults, not guarantees. Application code remains responsible for side effects.