Jev For All — AI skill for Claude Code

opencode-system-one

An OpenCode V2 plugin that pairs the System One model Jev (TypeSafe, reached through OpenRouter's alpha Decisions API) with a classic coding agent. Jev makes the fast, structured decisions — which skill to load, which tool subset this step needs — and the classic agent does the work. The agent starts building instead of deliberating: a smaller tool catalog, no tool-choice reasoning, and more reliable skill loads.

Install and configure

{
  "plugins": [
    {
      "package": "opencode-system-one",
      "options": {
        "apiKey": "sk-or-...",        // or set OPENROUTER_API_KEY
        "model": "~typesafe/jev-latest",
        "timeoutMs": 1000,
        "debug": false,
        "agents": ["build"],
        "serverURL": "https://openrouter.ai",
        "skills": { "enabled": true, "rerank": "auto", "gateThreshold": 0.3,
                    "rerankAbove": 40, "rerankBelowP": 0.5, "shortlist": 3,
                    "fitsThreshold": 0.3, "minConfidence": 0.3 },
        "tools":  { "enabled": true, "maxTools": 12, "minToolProbability": 0.05,
                    "needsToolThreshold": 0.3, "minConfidence": 0.3,
                    "alwaysVisible": ["read","write","edit","bash","grep","glob"],
                    "stateBudget": 6000 }
      }
    }
  ]
}

The API key comes from `options.apiKey` or the `OPENROUTER_API_KEY` environment variable. Zero config works once the env var is set. Main options:

• model — Jev model slug (default ~typesafe/jev-latest).
• timeoutMs — per-request timeout (default 1000 ms).
• skills.* / tools.* — thresholds and caps for the two hooks; skills.enabled and tools.enabled toggle them.
• agents — allowlist for tool routing (default: all agents).
• debug — log routing decisions.
• serverURL — override the API host (test/proxy seam; default https://openrouter.ai).

Data egress

On every model dispatch the `context` hook sends the **tail of the conversation** to OpenRouter's alph