dsh-jev-grep
DeepSeek Harness jev_search tool that runs ripgrep, scores every hit line with Jev, and returns only the top rows above a cut.
Overview
dsh-jev-grep is an adjudication layer for search floods in DeepSeek Harness. It does not replace grep; it replaces the step after a search returns hundreds or thousands of lines, when the agent either reads them all into context or guesses with grep -v. The jev_search tool runs ripgrep to find candidates, sends hit lines to Jev in concurrent batches for noul or choice judgments, ranks them by probability and returns about 30 lines above a cut, regardless of pool size.
When grep or glob results exceed a threshold (120 lines by default), a short nudge suggests using jev_search. If the endpoint is unreachable, times out or has no key, it returns the raw grep lines clearly marked as not adjudicated, never silently. It works with any systemone-shaped endpoint, including OpenCode Zen's free Jev tier, TypeSafe's API or OpenRouter. It has zero npm dependencies and runs on Windows, macOS and Linux. Documentation is in Chinese.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Is this matching line relevant to what the agent is looking for? | yes · no |
How it works
- Run ripgrep for candidate lines
- Batch lines to Jev for per-line relevance
- Rank and cut at a probability threshold (0.7 default)
- Return the top rows to the agent
Strengths
- Output size independent of hit count
- Transparent degradation when Jev is unavailable
- Free tier option via OpenCode Zen
Caveats
- Line-level judgments lack wider file context
- Chinese documentation
Getting started
Install as a DSH plugin and set url, model and key in ~/.dsh/dsh-jev-grep/settings.json.At a glance
- Type
- Plugin
- Status
- beta
- Language
- JavaScript
- Runtime
- DeepSeek Harness plugin
- Has code
- yes
Use cases and integrations
How curated lists describe it
Search-flood adjudication plugin for DeepSeek Harness: ripgrep + TypeSafe Jev…
wh000wh000/awesome-jev-live
Frequently asked questions
Does it replace grep?
No, it adjudicates large grep results so the agent reads only the most relevant lines.
Which endpoints work?
Any systemone-shaped endpoint, including TypeSafe, OpenRouter and OpenCode Zen.
What if Jev is down?
It returns raw grep lines marked as not adjudicated.
Related projects
Retrieval layer for Codex that finds evidence locally, optionally ranks sanitized excerpts with Jev, and returns bounded, cited evidence.
ProjectbetaGo-based Claude Code plugin where Jev reads code so the agent doesn't: semantic find, bounded questions and large-read takeovers.
PluginbetaGoPipe grep or file lists into Jev to get every candidate scored, sorted and cut at a calibrated threshold, so agents read only what matters.
ProjectbetaTypeScriptMCP server that returns only the JS/TS code a task needs by following the TypeScript compiler graph and asking Jev about relevance.
MCP serverbetaTypeScriptCodex search plugin that runs ripgrep and keeps only passages Jev judges relevant to the question.
PluginprototypeTypeScript- winnow★ 110
Calibrated context sieve for Claude Code: Jev judges each block of large tool output and hides confident-no blocks behind recallable stubs.
Projectbeta