hermes-jev-skills
View on GitHubJev-powered model routing, memory, compaction, skill selection, computer and browser use for Hermes agents (also Claude Code and Codex)
Agent-agnostic SKILL.md pack that offloads small agent decisions (model routing, memory filtering, compaction, skill pick, triage, mail sorting, computer/browser action picks) to TypeSafe's Jev decision model. Ships a stdlib-only Python CLI, installer for Hermes/Claude Code/Codex, and a routing dashboard.
Use Cases
Route each turn to the cheapest sufficient modelLoad the right skill out of hundreds per turnFilter retrieved memory passages for relevance and hidden instructionsSelect which transcript turns to keep during compactionGenerate agent handoffs with higher recallDecide next GUI action for computer useDecide next page action for browser useTriage incoming messages by urgency and typeSort an inbox into reply/updates/promo/spam lanesScreen prompts and documents for injectionCut frontier-model token spend on non-writing decisionsRun routing in shadow mode to log decisions before switching
Built With
- Language
- Python
- Frameworks
- Hermes · Claude Code · Codex · Jev (TypeSafe)
Tags
agent skills · model routing · LLM router · agent memory · context compaction · computer use · browser use · skill selection · prompt injection screening · cost optimization · handoffs · triage · Hermes · Claude Code · Codex · SKILL.md