Hermes Jev Skills
A Jev toolkit for agent model routing, memory filtering, compaction, skill selection, triage, and bounded computer/browser decisions.
Added to Jevfast
How it uses Jev
The toolkit integrates Jev into agent routing, retrieval filtering, compaction, skill selection, triage, mailbox sorting, search, and GUI/browser actions. Each feature supplies bounded state and candidate choices; Jev returns typed decisions while surrounding code enforces fallbacks and safety constraints.
What Jev decides
Documented CLI requestExample answers · not a recorded Jev response · Source ↗
Question 1 · blocked
YOUR APP
INSTRUCTION
The work cannot continue until a person steps in
STATE
The deploy failed twice on the same migration step.
JEV · NOUL
YesNo
Question 2 · next
YOUR APP
INSTRUCTION
What should happen next?
STATE
The deploy failed twice on the same migration step.
JEV · CHOICE
- retry
- escalate
Question 3 · severity
YOUR APP
INSTRUCTION
How serious is this?
STATE
The deploy failed twice on the same migration step.
JEV · SCORE
0
App workflow
- Prepare bounded context
Feature-specific code redacts or screens user turns, retrieved passages, messages, or action descriptions and constructs the candidate set.
- Ask Jev
The selected feature sends typed questions for routing, ranking, classification, or the next action; handoff selection is disabled by default.
- Apply guardrails
Confidence and risk checks constrain whether a result is used; missing keys, timeouts, malformed replies, or low confidence fall back to the original behavior.
- Observe or act
Shadow mode logs a proposed route without switching; enabled features apply the bounded result, while dashboards expose decisions and operational metrics.
Why it is interesting
It applies the same select-from-candidates pattern across several agent bottlenecks and includes shadow mode, fail-open behavior, and local screening so model decisions remain bounded and inspectable.