north-star
Hermes Agent skill that turns an intention into a checkable finish line, with shell checks and Jev judging the fuzzy requirements.
Overview
north-star is a Hermes Agent skill aimed at long autonomous runs that either stop early and claim success or never stop. It interviews the owner until there is a written finish line, then produces three artifacts: the star (goal, audience, requirements and checks as JSON), a generated run prompt, and a gate that decides whether the work is actually finished.
The gate has two layers. Machine-checkable requirements run as plain shell commands. Requirements no script can decide, such as whether a stranger could complete checkout unaided, go to Jev, which returns a probability and calibrated confidence. Exit 0 means every requirement is met; exit 1 keeps the run going and names the weakest requirement. The tool is pure Python with no dependencies, and the deterministic half needs no key. It works with any agent that can run shell commands. It is not an official Nous Research or TypeSafe project.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this fuzzy requirement met given the evidence? | yes · no |
| Choice | Which requirement is still least proven?(inferred) | — |
How it works
- Interview the owner to write a star with requirements and checks.
- Generate the run prompt and start the agent.
- At the gate, run shell checks, then ask Jev about fuzzy requirements.
- Exit 0 if all pass; otherwise continue with the weakest requirement named.
Strengths
- Stops agents declaring done without evidence.
- Deterministic checks stay in code.
- No dependencies; works without Hermes.
Caveats
- Fuzzy requirements need a TypeSafe key.
- Hermes install does not copy scripts; clone the repo too.
Getting started
hermes skills install https://raw.githubusercontent.com/poponline63/north-star/main/SKILL.md, then git clone the repo and run python3 ~/north-star/scripts/north_star.py --help.At a glance
- Type
- Agent skill
- Status
- prototype
- Language
- Python
- Runtime
- Hermes Agent skill, CLI
- Has code
- yes
Use cases and integrations
How curated lists describe it
Hermes Agent skill whose north-star gate is judged by Jev (TypeSafe System One): turn an intention into a checkable finish line, generate the run prompt, and let Jev rank what is still unproven.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev judge?
Requirements that no script can check, returning a probability that each is met.
Does it work without a TypeSafe key?
The deterministic half does; the gate states plainly when Jev is not available.
Do I need Hermes?
No, any agent that can run shell commands can use the two Python scripts.
Related projects
Software-factory supervisor where Jev independently judges coding agents' completion, test sufficiency, progress and need for a human.
ProjectbetaPythonAsynchronous System-1 supervisory plugin for Hermes Agent with typed Jev decisions, ranking, verification and an opt-in tool gate.
ProjectbetaPythonPortable skill pack giving agents a conservative Jev second opinion on intent, completion, routing and bounded browser navigation.
ProjectbetaPythonClaude Code Stop hook that blocks an unverified done: one four-question Jev call only when files changed and no check passed.
PluginbetaTypeScriptClaude Code plugin where Claude plans phased work with checks and, at each stop, Jev judges whether the work is really done.
PluginbetaTypeScriptStop hook for coding agents where Jev checks each attempted stop against plain-language rules and sends the agent back if one is violated.
HookbetaPython