dsh-teacher-consult
DeepSeek Harness plugin for budgeted, read-only GPT 'teacher' consults; Jev may only suggest when a consult is worthwhile.
Overview
dsh-teacher-consult lets a DeepSeek Harness agent, the student, ask two GPT teachers for advice on complex tasks: a planning teacher that answers with plan, risks and what to verify first, and an expert teacher that answers with a recommendation, rationale and main risk, with an escalation tier at higher effort. Teachers only advise; they never take over, communicate with each other, modify files or see the conversation.
Rules are enforced in code: consults are stateless one-shot codex exec runs with no thread, a hard per-task budget allows at most three replies, escalation requires a prior expert consult within recent tasks, and teachers run in a read-only sandbox. Jev's role is narrow: it can only suggest a consult and can never start one, and a deterministic prefilter decides whether Jev is asked at all. Documentation is in Chinese with an English README.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Would consulting a teacher be worthwhile for this task?(inferred) | yes · no |
How it works
- Deterministic prefilter decides whether to ask Jev
- Jev may suggest a consult
- Agent calls a teacher tool within the budget
- Teacher replies read-only; advice returns to the agent
Strengths
- Hard budget and read-only sandbox enforced in code
- Stateless consults avoid context growth
- Jev cannot trigger spending on its own
Caveats
- Jev plays a minor advisory role
- Depends on external GPT models via codex
Getting started
Install as a DSH plugin per README.en.md with codex configured.At a glance
- Type
- Project
- Status
- prototype
- Language
- JavaScript
- Runtime
- DeepSeek Harness plugin
- Has code
- yes
Use cases and integrations
How curated lists describe it
GPT teacher consults for DSH: stateless one-shot codex teachers behind a hard per-task consult budget, a read-only sandbox and a deterministic prefilter.
hellogumbo/awesome-jev
Frequently asked questions
What does Jev do here?
It can only suggest that a teacher consult is worthwhile; it never starts one.
How many consults per task?
At most three replies per human task.
Can teachers edit files?
No, they run in a read-only sandbox.
Related projects
Multi-agent orchestration layer for Claude Code with opt-in Jev hooks for fast judgment points, decision caching and egress controls.
PluginproductionTypeScriptRuns Claude Code on Opus 5.5 with Jev re-picking reasoning effort after every tool batch, via per-message effort so the cache survives.
ProxybetaTypeScriptOpenClaw plugin and CLI that turn a plain-language condition into a scheduled trigger, waking the model only when Jev says it is true.
PluginbetaTypeScriptTiny demo that asks Jev whether an agent should continue, downgrade its model, or stop, given budget and plan.
DemoprototypeTypeScriptGate that asks Jev whether a timer or event is worth resuming a sleeping agent's LLM, skipping only when confidently not.
BenchmarkprototypeTypeScriptBash-native CLI that filters any stream by meaning with Jev, like grep for semantics, so an agent wakes only for lines that matter.
ProjectbetaTypeScript