User repository on GitHub
OpenCode v2 plugin that uses Jev (TypeSafe AI) to dynamically select relevant skills and tools per turn, reducing context window usage.
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.
CommunityJev Agents & MCP
Jevselector
OpenCode v2 plugin that asks Jev each turn which skills and tools matter for the request, then trims the advertised set for that request only.
- Stars
- 3
- Language
- TypeScript
- License
- MIT
- Published by
- Community
- Author
- universam1
- Last verified
- Oct 2026
About this project
An OpenCode v2 plugin that uses Jev (TypeSafe AI's "System One" decision model) to dynamically select which skills and tools are exposed to the model on each turn — trimming the system prompt and tool set to only what is relevant to the current request. OpenCode advertises every discovered skill's description in the system prompt on every turn, and exposes the full tool set. With 30 skills installed, the block alone can be over half of the system prompt — spent describing skills the current request will never use.
Quoted from the project's README. Read the full README.
Works forJev for Classification
Used inJev for Software & DevTools
- Languages
- TypeScript
Highlights
- Registers a per-turn hook sending one batched Jev noul relevance question per skill and tool over shared state.
- Rewrites only the outgoing skills block and tool set, caching decisions per agent, request, and candidate set.
- Fails open on error or timeout, never unregisters anything globally, and supports allowlists for pins.
- Exposes threshold, model, endpoint, timeout, skill and tool filters, and verbose logging options.
Watch out
Savings depend on skill descriptions dominating the prompt and on Jev judging relevance well; workflow-critical skills may need pinning. Copy reviewed against the repository README on 2026-10-04, not from running it.
Questions about this entry
Short answers drawn from the linked project or page.
The repo measures roughly half the system prompt removed on typical turns with dozens of skills installed, compounding every turn while the Jev decision itself is cached across continuations.
It fails open and exposes everything on any Jev error or timeout, and a hidden skill can still be loaded by exact id; pin orchestration skills with alwaysKeepSkills in multi-agent setups.
Set TYPESAFE_API_KEY, install it as a local plugin through a relative path shim until it is published, and tune keepThreshold around 0.5 with optional verbose logging.
When Code Mode collapses MCP tools behind one execute tool they are not separately addressable for filtering, so only skills and top-level tools can be trimmed there.
OpenCode v2, tested on v2.0.12, using the @opencode-ai/plugin v2 promise API, plus a TypeSafe AI key in TYPESAFE_API_KEY. It runs on the OpenCode plugin runtime with Node 20+ or Bun, and the project is MIT licensed.
A session.context hook fires once per model turn and sends one batched Jev noul relevance question per skill and per tool over shared state. The decision is cached per agent, request and candidate set, so continuations and delegated sub-agents such as orchestrator to librarian each get their own rating without re-calling Jev.
Against the oh-my-opencode-slim orchestrator the system prompt fell from about 10,201 to about 7,597 tokens, a 25.5% cut, with advertised skills dropping from 33 to 5 and tools from 19 to 3. The percentage is lower than for a lightweight agent because the large orchestration playbook stays, but the absolute saving of roughly 2,600 tokens per turn compounds.
More Jev resources like this
5GitHub stars
Bun app that feeds live Kraken BTC, ETH, and SOL order-book features into Jev for typed BUY, SELL, or HOLD calls and runs them through deterministic paper execution with an SSE dashboard.
Jev Agents & MCP#community#github
34.9kGitHub stars
Stencil Labs' coding agent embeds TypeSafe Jev for typed checks, routing, and workflow decisions, with measured Jev tokenizer probes in pi-natives.
Jev Agents & MCP#community#typescript#coding-agent
Community
can1357/oh-my-piTypeScript
can1357/oh-my-piTypeScript1.5kGitHub stars
Open-source platform that puts ACP-compatible agents such as Claude Code, Codex, and Pi inside the team's Slack, Telegram, Discord, and GitHub workflows, with reusable Jev decisions for routing, response gating, and reviewer selection.
Jev Agents & MCP#community#jevbest
488GitHub stars
A coding agent built on pi that hands its non-code decisions — what enters the context, whether a command is safe, whether the work is done — to a small judge model at 35 decision points per turn.
Jev Agents & MCP#community#jevbest
Community
qybaihe/muTypeScript
qybaihe/muTypeScript210GitHub stars
An agent plugin for Pi, Claude Code, Codex CLI and Grok Build that asks Jev two classification questions to decide whether a session sits at a safe compaction boundary, then hints or compacts for you.
Jev Agents & MCP#community#jevbest
572GitHub stars
A CLI plus agent hooks that ask Jev one question per rule in your AGENTS.md on every edit or turn and return a probability, so a violating edit gets repaired inside the same turn. Claude Code, Codex, OpenCode, Pi.
Jev Agents & MCP#community#jevbest
Community
coldteadotai/abideTypeScript
coldteadotai/abideTypeScriptFrom the Jev community
Posts from builders shipping with Jev right now.