用 Jev 评估多路对话行为规范的 Loom TUI 工具 carla

kenneth @local0ptimist原推发布于 2026.09.27

在 X 上打开原推

having some fun with this - i built a loom TUI called carla (character lab) that follows @parafactual's methodology taking you from initial seed docs > continuations > an anthology > a chat simulator this is a screenshot of the latter where i can run multiple conversations at once with /loom [count] after each turn, Jev evaluates against a behavior spec i can check for looping, spiraling, and harmful language or whatever i can define into a spec. when it detects something beyond a threshold, the grid pulses orange so i know to check on it

这条是引用着另一条写的：x.com/parafactual/status/2103319912035057924