jev-agent-skills
Portable skill pack giving agents a conservative Jev second opinion on intent, completion, routing and bounded browser navigation.
Overview
Jev Agent Skills (jev-judgment) from SISO is a portable skill pack that gives an agent an optional, conservative second opinion. It checks intent, completion evidence, progress, route or skill selection, bounded Camofox browser navigation, and relevance of shadow context. It explicitly does not run the task, grant permissions, spawn workers or install hooks; the agent loads SKILL.md when it chooses to use it.
It uses only the Python 3.10+ standard library and calls Jev through OpenRouter with the user's own key. Every check has a dry-run mode that validates packets without paid calls or browser actions, and live invocations are explicit after reviewing the packet and the data boundary. Scripts cover intent, completion, progress, route and select checks, single-link browser judgments and a bounded fill, click and link navigation loop. Version 1.3.0 is the first public release, MIT licensed, promoted from the SISO Skills Hub.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is there sufficient evidence that this task is complete?(inferred) | yes · no |
| Choice | Which route or skill fits this request?(inferred) | — |
| Choice | Which link or control should the bounded browser loop use next?(inferred) | — |
How it works
- Install one copy in the agent's skill directory
- Prepare a packet for the check
- Run with --dry-run to validate
- Run live via OpenRouter and use the answer as a second opinion
Strengths
- Dry-run validation before any paid call
- Conservative: never acts or grants permissions itself
- Standard library only
Caveats
- Requires an OpenRouter key
- Agent must choose to load the skill; no automatic activation
Getting started
`git clone --branch v1.3.0 --depth 1 https://github.com/sisodias/jev-agent-skills.git ~/.codex/skills/jev-judgment`, then `python3 harness.py intent examples/intent.json --dry-run`.At a glance
- Type
- Project
- Status
- beta
- Language
- Python
- Runtime
- agent skill
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A conservative second opinion for your agent: intent, completion, routing and…
wh000wh000/awesome-jev-live
Frequently asked questions
What does the pack check?
Intent, completion evidence, progress, routing, browser navigation steps and context relevance.
Will it take actions?
No. It only offers a second opinion and never grants permissions or runs tasks.
Can I test without paying?
Yes, every check supports --dry-run.
Related projects
Claude Code skills that wait for AI pull-request reviewers and judge whether their reports actually permit a merge, never inferring a pass.
ProjectbetaLinear-backed work ledger for coding agents where a Jev-compatible model verifies done reports, blockers and plan changes against the task.
ProjectprototypeJavaScriptIntent compiler skill that turns vague coding requests into a typed IntentSpec, with Jev or LAYA as optional decision backends.
ProjectbetaMCP server giving coding agents fast typed decisions (ship gates, risk checks, file triage, page checks) as accept, review or escalate.
MCP serverbetaTypeScriptTyped judgment layer for Claude Code, Codex and Hermes Agent: deterministic facts in code, language judgments in Jev, with receipts.
ProjectbetaPythonHermes Agent skill that turns an intention into a checkable finish line, with shell checks and Jev judging the fuzzy requirements.
Agent skillprototypePython