10 open-source repos that wrap a Jev decision layer around Claude Code

Chen @ch3nweiiiOriginally posted 2026.10.08

Open original on X

Opus 5.5 + Jev make Claude Code look f…cking illegal 10 GitHub repos that wrap a decision layer around every session 01 toolgate ▸ https://github.com/RiskAverseTech/toolgate → a tool-call firewall: Jev answers 7 yes/no risk questions per call, code maps them to allow, ask or deny and fails closed 02 Switchboard ▸ https://github.com/ruban-24/switchboard → Jev sizes up the task, code picks the model and reasoning effort, then pins the pair so the prompt cache stays warm 03 spending-effort-with-jev ▸ https://github.com/Yaxin9Luo/spending-effort-with-jev → a UserPromptSubmit hook that asks Jev which /effort level the prompt actually needs 04 winnow ▸ https://github.com/GhalebDweikat/winnow → judges every tool result before it's allowed into Claude Code's context 05 fast-jev-compaction ▸ https://github.com/tamaratran/fast-jev-compaction → scores old tool calls for deletion and keeps everything it retains word for word 06 Canny ▸ https://github.com/qkal/Canny → an evidence ledger that challenges every "done" your coding agent can't back up 07 Médula ▸ https://github.com/JoaquinRuiz/medula → several Claude Code agents on one repo: before each write Jev scores the chance of a collision, then it goes, waits or escalates 08 jevwire ▸ https://github.com/Brainwires/jevwire → MCP tools plus advisory or restrictive Claude Code hooks, all built on typed Jev decisions 09 skillranker ▸ https://github.com/Dicklesworthstone/skillranker → a Rust CLI that ranks your skills against the live session and abstains when nothing fits 10 wakegate ▸ https://github.com/shitianfang/wakegate → Jev decides if a timer or event is worth waking the LLM, and skips it only when it's sure Copy any of these Jev builds, then read the full Jev setup below ↓

This post quotes another post: x.com/hanakoxbt/status/2101709924828934222