jev-code
Agent tool and CLI that batches typed Jev classifications, checks, scores and rankings for coding tasks.
Added to Jevfast
How it uses Jev
jev-code exposes Jev judgments to Claude Code, Codex, Pi, OpenCode, and a shell CLI through five tools. An agent can classify many evidence items into supplied classes, check a yes/no claim, score or rank candidates, or ask mixed typed questions. The package batches items into one System One request and returns probabilities and threshold-derived decisions for the agent to use.
What Jev decides
Repository-documented raw Jev requestExample answers · not a recorded Jev response · Source ↗
Question 1 · department
YOUR APP
INSTRUCTION
Which team should handle the message?
STATE
Example state.message reports API errors and failed orders. The repository shows a direct System One request with department, urgency and frustration questions over this shared state.
JEV · CHOICE
- billing
- technical
- sales
Question 2 · is urgent
YOUR APP
INSTRUCTION
Does the message convey urgency or time pressure?
STATE
Example state.message reports API errors and failed orders. The repository shows a direct System One request with department, urgency and frustration questions over this shared state.
JEV · NOUL
YesNo
Question 3 · frustration
YOUR APP
INSTRUCTION
How frustrated does the author appear?
STATE
Example state.message reports API errors and failed orders. The repository shows a direct System One request with department, urgency and frustration questions over this shared state.
JEV · SCORE
0
App workflow
- Install the agent bridge
Export TYPESAFE_API_KEY and run the setup command for the desired coding-agent harness; setup installs the skill and registers the MCP or native tools.
- Identify a judgment task
The agent uses the jev skill when a task involves classification, a yes/no check, scoring, ranking, or another typed judgment.
- Send evidence and criteria
The agent invokes the matching tool with evidence and its own question or allowed labels; item batches are sent in one request.
- Act on the structured result
The tool returns labels, probabilities, confidence, and auto-or-review decisions for the agent to apply or verify.
Why it is interesting
It packages TypeSafe's decision API as a cross-harness tool and skill, so an agent can make typed judgments over evidence and then branch on confidence without relying on generated prose for the classification itself.