Jev-cu
Let Jev choose text-grounded computer-use actions while Codex reads and executes under local safety policy.
What Jev decides
Which visible text candidate and action to use next, whether the goal is complete, and whether the action is sensitive.
Jev-cu is an installable Codex skill and decision loop that sends UI text candidates rather than screenshots, asks Jev to select elements and actions and assess completion and risk, then applies local confirmation rules.
- Jev role
- Choose the next text-grounded UI action and estimate completion and operational risk.
- Input
- The English task goal, application and accessibility text, candidate elements and actions, and recent loop state.
- Output
- Choice results for element and action plus Noul-style completion and risk probabilities.
- Decision frequency
- Once per step in the Codex computer-use loop.
- Fallback
- Default to dry-run, require confirmation for sensitive action classes, restrict applications through a local allowlist, and stop rather than bypass login or verification boundaries.
How it works
Overview
Jev-cu connects TypeSafe Jev to the Codex desktop computer-use runtime. Codex reads the interface and performs actions; Jev selects from text-derived elements and actions and judges completion and risk. Screenshot pixels are not sent to Jev.
The loop defaults to dry-run and has an explicit application allowlist. Deletion, sending, payment, authorization, uploads, verification codes, installation, and system settings stop for human confirmation. The repository includes offline unit tests plus a credentialed P0 evaluation over accessibility snapshots; this listing does not claim that the live evaluation was reproduced by this site.
- Language
- JavaScript
- License
- Not declared
- Author
- Sac-Y
Sources
Related projects
typesafe-computer-use
A native macOS computer-use loop built from local OCR, accessibility data, and bounded Jev actions.
What Jev decides
Which computer action, visible item, site, or off-screen accessibility control to use next.
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.
jev-model-router
Route Claude Code model tiers and reasoning effort with one parallel Jev classification.
What Jev decides
Which fast, balanced, or deep tier fits the work, how much reasoning it needs, and whether production risk requires escalation.