Community · E2 · artifact verified
Plug a fail-closed decision layer into your agent
A pluggable decision layer for the ego agent: System One (Jev) by default, swappable to local or other OpenAI-compatible backends, fail-closed guardrails - and the README's whole argument is that it is measured: 16 suites, 429 checks, rerun in full.
01 · Role in the system
What Jev does here
When the agent must choose, the layer asks Jev typed questions over the systemone protocol and executes only on a calibrated answer; if the backend is slow, wrong-shaped or absent, the guardrails fail closed - the action does not run on an unmeasured guess. The backend is a config choice (hosted Jev, local models, any OpenAI-compatible endpoint), and the repo proves the measurement claim rather than asserting it: benchmark harnesses that post to the live endpoint, baseline engines, route-trace and offline end-to-end suites, with the full 16-suite, 429-check rerun recorded from 2026-09-26. Distributed as an agent skill, so installation is one skills.sh command.
02 · Control boundary
Where Jev sits
Jev as the default decider behind a swappable decision interface: typed answers gate execution, guardrails fail closed when no calibrated answer arrives, and 16 benchmark suites with 429 checks keep the layer accountable.
Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.
03 · Known limits
What this evidence does not prove
- Built for the ego agent family; other agents need adapter work.
- The 429-check rerun is author-executed; no independent replication.
- 5 stars; fail-closed coverage is only as good as the guardrail tests.
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- jiangkoumo ↗Community · github · public · checked 2026-10-03