dsh-jev

dsh-jev is a DeepSeek Harness bundle that registers the jev_ask tool. A DSH agent on a profile that loads it can send typed noul, choice, and score questions to TypeSafe Jev.

Jev is TypeSafe's System One model. It returns structured answers and does not generate chat or code.

This package is version 0.1.0. It targets DSH 0.1.5-rc.2 from the npm next dist-tag, Node 22.19 or later on the 22.x line or Node 24 and later, and a TypeSafe API key.

Install from GitHub

Pin a commit so a later push cannot change what you run.

dsh plugin --profile web add github:noetion/dsh-jev#<commit>

Replace <commit> with a SHA from main.

The first add fails until pnpm allows this package's prepare script. Copy the exact key from the pnpm error into the profile's pnpm-workspace.yaml under allowBuilds. It is often dsh-jev, and some pnpm versions print a longer dsh-jev@github:... form. Use the string pnpm printed.

allowBuilds:
  dsh-jev: true

Then run the same dsh plugin add command again. Treat that allowlist as permission to run this package's build on your machine at install time.

Set TYPESAFE_API_KEY in the process environment, or store it in DSH credentials under the same name. Restart dsh web if it is running.

Install from a local checkout

pnpm install
dsh plugin --profile web add /absolute/path/to/dsh-jev

pnpm install runs prepare, which writes dist/.

Call jev_ask

Ask the agent to call jev_ask, or invoke the bundled jev skill. The session LLM decides when to call the tool. The plugin does not force it.

One call can mix question types over the same state.

{
  "state": "Help! My payouts have been failing for 3 days.",
  "questions": {
    "is_urgent": {
      "type": "noul",
      "instructions": "Does this convey urgency?"
    },
    "department": {
      "type": "choice",
      "instructions": "Which team should handle this?",
      "criteria": {
        "billing": "Payments, invoicing, refunds",
        "technical": "Bugs, outages, integrations",
        "sales": "Pricing, upgrades, new accounts"
      }
    }
  }
}

The tool result is the TypeSafe JSON. A noul answer carries a probability in noul. A choice or score answer carries probabilities and confidence.

Every call posts state to https://api.typesafe.ai/v1/systemone. Do not put secrets in state.

Configure

Optional cordis.yml fields on the jev row:

• model, default jev-latest
• endpoint, default https://api.typesafe.ai/v1/systemone
• apiKeyEnv, default TYPESAFE_API_KEY

The plugin reads the key on every call. It does not put the secret in cordis.yml.

Names

• GitHub repository: noetion/dsh-jev
• Package name: dsh-jev, installed from GitHub rather than npm
• Plugin id: jev
• Tool: jev_ask
• Bundled skill: jev

Verify

From a checkout:

pnpm install
pnpm run check

pnpm run check runs typecheck, test, build, and adopt-check in that order. The tests use a mock HTTP transport, so they need no API key.

Live TypeSafe proof, with a key in TYPESAFE_API_KEY or in the file named by TYPESAFE_KEY_FILE:

pnpm run live

Install-and-load proof against a throwaway DSH_HOME. Set DSH_BIN to use a dsh binary you already have instead of fetching one:

pnpm run verify-boot

This one installs the checkout over git+file:// and checks that the build wrote dist/. It also prints the allowBuilds key that install asks for, which is not the key a GitHub install asks for:

pnpm run verify-git-install

@deepseek-ai/dsh-tools is pinned to 0.1.5-rc.2. npm latest is a stale 0.0.1-rc.1. Do not replace the pin with an untagged install.

Why this is not an LLM adapter

An LLM adapter must stream text and tool-call chunks. Jev returns typed answers and cannot stream, so it cannot be one. The plugin registers a tool, and a bundled skill when the profile has ctx.skills.

License and support

MIT. See LICENSE.

Open a GitHub issue for bugs and questions. There is no response-time promise. Pull requests are welcome if they include pnpm run check.

To report a security vulnerability, use GitHub's private advisory form on this repository instead of a public issue.