Files
hermes-agent/plugin-catalog/jev-judge.yaml

30 lines
1.5 KiB
YAML
Raw Permalink Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

name: jev-judge
repo: https://github.com/DoGMaTiiC/hermes-jev
sha: 5dddbeaac5be08281e6c3b372eeaba3639971f0a
subdir: plugins/jev-judge
description: TypeSafe Jev as a judgment layer — a pre-tool gate (shadow by default, enforce
escalates) on terminal, write_file and patch asking destructive / exfiltration / impact
(a v2 signal set — reads_secrets × sends_outbound, blast_radius, self_advocating — is one
setting away), plus a jev_ask tool for typed judgments. Thresholds in code, one JSONL line
per decision with the fail-open reason, fail-open on every error path. Stdlib only, no Node;
no redirects, proxy env ignored, payload capped; works with TYPESAFE_API_KEY,
OPENROUTER_API_KEY, or AI_GATEWAY_API_KEY. Disclosure — each judged call sends the tool name
and its redacted arguments (≤600 chars each, ≤12 args) to TypeSafe, OpenRouter, or your
configured gateway, using your key; shadow mode only logs. With the default backend auto,
OPENROUTER_API_KEY is used automatically whenever it is present and TYPESAFE_API_KEY is not,
so tool names and redacted arguments go to openrouter.ai on your main-model OpenRouter key
unless backend is set to typesafe or gateway.
maintainer: DoGMaTiiC
tier: community
category: tools
docs_url: https://github.com/DoGMaTiiC/hermes-jev/blob/main/plugins/jev-judge/README.md
version: "0.1.0"
platforms: []
capabilities:
provides_tools:
- jev_ask
provides_hooks:
- pre_tool_call
provides_middleware: []
requires_env: []