Community · E2 · artifact verified

Gate agent commits with plain-English rules

tenet makes agents fix rule violations before you ever see the diff: rules live in a YAML file in plain language, and Jev answers each one with a calibrated probability that becomes a pass-or-fail cutoff.

01 · Role in the system

What Jev does here

You write each tenet once - what a comment is for, that a failure is raised rather than hidden - and every commit is judged against the file. Jev answers each rule with a calibrated probability, so a tenet is a cutoff rather than a review comment to skim. The README reports a staged change judged in 1.1 seconds for $0.0022 and a 2,062-line diff in 1.5 seconds and $0.0036, against 50.3 seconds and $0.0535 for one small-model LLM call over the same diff and 121.4 seconds and $0.1943 for a frontier model. Install is a single binary from Homebrew, npm, or an install script, with pre-commit and GitHub Actions integrations so every commit or pull request is linted automatically. The project flags itself as pre-release: format, flags, and output are not stable.

02 · Control boundary

Where Jev sits

Plain-language rules in config; per-rule calibrated probabilities from Jev become deterministic pass-or-fail cutoffs enforced at commit time by a local binary.

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

  • Speed and cost comparisons against LLM reviewers are author-reported from the author's own benchmark tables.
  • Pre-release software: the tenet format and CLI surface can change between versions.
  • Rules judge what the diff says against written intent; they do not test behavior, and the README positions the gate as fixing style-rule violations, not replacing review.

04 · Attribution

Public sources

This is a Community record: the project was published by a third-party community author.

  • zoidshCommunity · github · public · checked 2026-09-23