Hermes Jev Skills
Thirteen skills hand an agent's small decisions to the cheap Jev model: model, skill, memory passages and next step.
- 1,338 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- Requests and fragments of working context go to the external TypeSafe service under your key.
- The skills can choose browser and GUI actions, so limit the action table to safe ones.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/hermes-jev-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Requests and fragments of working context go to the external TypeSafe service under your key.
- The skills can choose browser and GUI actions, so limit the action table to safe ones.
This is third-party code. Review the repository files before installing.
What it does
Hermes Jev Skills connects an agent to Jev, TypeSafe's decision model. It writes no text: it is given a state and a typed question such as pick one or yes or no, and returns a choice with a confidence score in about 0.4 seconds for a fraction of a cent. The expensive model is left for the actual writing. The repository holds 13 skills as plain `SKILL.md` files. They route models per turn, choose which search results to open, pick memory passages and find hidden instructions among them, select the needed skill out of hundreds installed, sort mail into lanes, rate message urgency and choose the next action in a GUI or browser. A web-result screen withholds fragments carrying instructions aimed at the agent before it reads them. `jev dashboard` opens a panel of decisions. The installer finds Hermes, Claude Code and Codex on the machine and installs for each one it finds. The author gives their own measurements and notes where the gain is not proven: for example, a handoff built from Jev's digest did worse than one built from the plain transcript, so the whole dialogue is used by default.
Who it is for. People who run agents on a stream of small decisions and burn expensive tokens on choosing a model, a skill or a source.
Good fit when
- An agent holds hundreds of skills and choosing the right one each turn costs time and tokens.
- You want simple turns on a cheap model and hard ones on a strong one.
- Search returns many pages and only the useful ones should be opened.
- An agent reads third-party web pages and you want embedded instructions cut before reading.
- A mailbox needs sorting into lanes with an urgency estimate.
Not a fit when
- Data must not go to an external cloud service: decisions come from the TypeSafe model through a key.
- You want one ready feature with no setup: the skills suit an agent that has something to route.
- The job is long writing rather than choosing: Jev writes no text.
Example request
Install Hermes Jev Skills, set up the key and turn on per-turn model routing, then show the decisions panel.Limitations
A TypeSafe key is required (`jev setup-key`) with a `jev doctor` check. The platform is paid, and the price and speed figures come from the author's own measurements. Python 3.9 or newer is required and no external packages are needed. Some skills, such as Blender and video, are limited to local checks, and the author has not confirmed an overall gain for video review. The web-result screen was tested on a subset of attacks and does not guarantee full protection against instruction injection.
How to disable. Run `python3 ~/hermes-jev-skills/install.py --uninstall` and remove the `jev` key from storage by hand. The `--check` flag shows the plan without changing anything.
FAQ
What is Jev?
TypeSafe's decision model: it answers multiple-choice style questions and gives a confidence score but writes no text.
Does it work without Hermes?
Yes. The skills are plain `SKILL.md` files, and the installer also finds Claude Code and Codex.
How much does it cost?
By the author's measurements a decision costs around a hundredth of a cent, for example 0.00006 dollars per message for triage. Check current prices with TypeSafe.
What happens if Jev is unavailable?
Some checks run locally, such as the injection screen for memory, and the other skills fall back to the agent's normal behavior.
How do I check the install without changes?
Run `python3 install.py --check`, which shows what would be done.
Discussion
No questions or reviews yet.
Related
SkillsFive skills for the Jev model: sorting records, source checks, rubric scoring and choosing an agent's next action.
The Jev classifier as an agent tool: labels items, answers yes or no and ranks candidates with probabilities within milliseconds.
A local router merges models, subagents, skills and MCP tools into one candidate set and picks who handles the next step.
Jev-Style runs small decision models on your machine: a probability per option, six skills, MCP and a Claude Code guard hook
A Claude Code skill that hands bulk sorting of files, logs and tickets to the fast Jev model to cut Claude token use
A plugin for DeepSeek Harness and iPolloWork: structured choice of tools, skills and agents and scoring of outputs
README in short
The authors present Jev as a fast, cheap second brain for an agent's small decisions, with a table of 13 skills, measurements and caveats. MIT license.