jevbooks

← All projects

dsh-jev-guard

7starsseeker/dsh-jev-guard

Pre-execution safety valve for DeepSeek Harness shell commands.

Verification & Guardrails100%Runner-up: Classification & Routing

What it is

A DSH plugin that intercepts bash/pwsh tool calls before they run, judging whether a command will irreversibly delete or overwrite real data. Built for DSH users who want a hard interception rather than a model nudge.

How it uses Jev

After offline L0 static rules and a read-only pre-screen, Jev answers one yes/no question: will this command irreversibly delete or overwrite the user's real data? Two thresholds split the probability into allow, revise, or block; escalate handles hard-rule 'always ask' cases and spent retry budgets.

Primitives:choice

Technique worth stealing

Anchor dangerous-command rules at command position only, so mentions in arguments are not false positives and buried commands are not missed.

Try it

dsh plugin --profile web add github:7starsseeker/dsh-jev-guard, then node bin/guard.mjs selftest

View on GitHub

judged by Jevjev-1.13.0

Evidence

Each line is one question put to Jev about the README. ≥ 0.60 reads as yes, ≤ 0.40 as no; in between Jev is not making a call.

  • Jev-centricyes0.82
  • Shows a System One patternno0.34
  • Handles uncertaintyno0.25
  • Measuredno0.10
  • Runnableyes0.90
  • Worth recommendingno0.30
  • Model replicano0.07
  • Problem scopescore on a 0–2 scale1.13
  • About Jevyes0.97

Signals by Jev jev-1.13.0, card written by DeepSeek V4.1 Flash from the README on 20 Sept 2026.