Hermes Jev Skills
kerpopule/hermes-jev-skills
Skills that give Hermes agents a cheap decision model for routing, skill selection and web screening
Hermes Jev Skills is a set of SKILL.md skills that send small agent decisions to Jev, TypeSafe's decision model, instead of a frontier model. The decisions include model routing, skill selection, memory retrieval and web screening for Hermes agents, and the same skills also work in Claude Code and Codex.
What Hermes Jev Skills does
Jev is TypeSafe's decision model, and the README stresses that it never writes text. You give it a state and typed questions, such as pick one, score this or answer yes or no, and it replies with a calibrated confidence. This repository wires that into an agent's routine so the expensive model is left to do the writing. The README gives rough timings, for example about 0.4 seconds per model-routing turn and 377 skills scanned in about 2.8 seconds.
Eleven skills ship as plain SKILL.md files, including model routing, search, memory, web screening, choosing turns to keep, skill selection, triage, mailbox sorting, computer use and browser use. For web screening, the README reports that Jev caught 70 of 79 planted attacks in 80 real web results, against 11 for Hermes's own pattern scan. A jev dashboard shows each decision as it happens.
Key features
- Model routing that picks a good-enough model for each turn in about 0.4 seconds
- Skill selection across 377 installed skills in about 2.8 seconds
- Web screening that withholds instruction-bearing parts of web_search and web_extract results
- Memory skill that ranks retrieved passages and screens them for hidden instructions
- Triage and mailbox sorting for incoming messages
- Computer-use and browser-use skills that choose the next action from a table of actions you judged safe
When to use it
- Cutting frontier-model spend by routing simple turns to a cheaper model
- Screening fetched web pages for hidden instructions before the agent reads them
- Sorting an agent's inbox into lanes such as needs reply, updates and spam
Who it is for: Hermes Agent operators who want a cheap decision layer for routing, screening and skill selection.
How it fits with Hermes Agent
Built for Hermes agents, including a Hermes-specific web screening skill. Because the skills are plain SKILL.md files, the same folder also works in Claude Code and Codex.
Note: The skills depend on Jev, TypeSafe's decision model, although the README says the memory injection screen runs locally even when Jev is down.
FAQ
What is Hermes Jev Skills?
Hermes Jev Skills is a collection of SKILL.md skills that hand small agent decisions to Jev, a decision model from TypeSafe. Jev answers typed questions with a confidence score and never writes text.
Does Hermes Jev Skills work with Hermes Agent?
Yes, it is built for Hermes agents and includes a web screening skill specific to Hermes. The skills are plain SKILL.md files, so they also work in Claude Code, Codex or anything that reads a skill file.
Is Hermes Jev Skills free and open source?
Yes, the repository is released under the MIT license. The skills call Jev, which is a separate TypeSafe product.
Similar skills for Hermes Agent
All skillsAgent skill that turns six top X creators' methods into a tweet writing and growth playbook
Spielewoy AutopromptCoding-agent workflow that reviews, fixes and rechecks its own work across 11 agents including Hermes
ZeroPointRepo YouTube SkillsYouTube transcript, search, channel and playlist skills for agents including Hermes Agent and OpenClaw
LearnPrompt LubanSkill-polishing workshop that audits an agent skill and improves it through five gated steps
alchaincyf Steve Jobs SkillAgent skill that answers in the style of Steve Jobs using six mental models and eight decision heuristics
momozi1996 Awesome AI Persona SkillsCollection of persona and writing-style skills with multi-agent teams, documented in Chinese
Related guides: What is Hermes Agent? · SOUL.md for Hermes Agent: what it is and how to write one