用 jev-axi 给 Grok Bot 加命令与内容审查层

Solomon Rojie @Solomonrojie原推发布于 2026.10.18

在 X 上打开原推

It is becoming f*cking dangerous: your Grok Bot can now have an AI bodyguard checking risky commands and catching hidden instructions before it acts → Ask Grok Bot to create a folder named JEV LAB on its cloud computer and use it for this workflow → Install the community Jev CLI: npm install -g jev-axi → Get a key from http://console.typesafe.ai and configure it securely with: jev-axi config set apiKey YOUR_KEY → Route the workflow’s shell commands through: jev-axi guard-exec --on-ask deny --on-error deny -- "YOUR_COMMAND" The wrapper checks the command before execution and blocks when it cannot approve it → If your Bot works in a Git repository, run: jev-axi setup git-hooks Added credentials get checked locally before a commit, while risky changes get flagged for review → Screen fetched text before using it: jev-axi guard < email.txt Suspicious instructions get flagged, and your workflow can stop when the check blocks the input → For coding tasks, check the work before calling it finished: jev-axi progress --job TASK.md --log test.log Get a completion assessment against the task and available evidence start with one workflow and test what passes, what gets blocked, and what needs your review give Grok Bot a second opinion on the commands, fetched content, and code changes your workflow routes through these checks Bonus: pipe a failing test run into Jev for a likely root cause: npm test 2>&1 | jev-axi triage

这条是引用着另一条写的：x.com/0xnineinch/status/2108604801411879139