jev-dsh-decision

What it solves

Beyond planning and execution, an agent loop usually lacks a judgement layer: which tool, Skill or sub-agent should own a task, whether a result needs review, and how good it is. Jev DSH Decision Engine wires TypeSafe AI’s Jev decision model into the harness and returns a choice, probability, confidence and usage in one call — no free-form reasoning, no automatic execution, and no takeover of the host agent’s planning or execution.

Highlights

• Recommends among tools, Skills and agents that are actually available and permitted right now.
• Three structured question types: choice, score and noul (probability of “yes”); up to 64 questions per call, so complementary Skills can be batched.
• Ships jev_status (local config state), jev_check_connection (real connection test) and jev_evaluate (structured evaluation), plus a jev-decision Skill.
• Requests always go to https://api.typesafe.ai/v1/systemone with model jev-latest; limits are 256 KiB request, 1 MiB response and a 20 s total network budget, with at most two retries on 429/529. No background learning, no global request interception.
• Does not switch models or route all traffic; upstream error bodies are redacted.

Who it’s for

• DeepSeek Harness users who want a native dsh.bundle plugin that runs standalone, without iPolloWork.
• iPolloWork GUI users: install once in Work and share it across the OpenCode, DeepSeek Harness and Codex Harness engines it supports.
• Teams that want a clearly bounded decision and output-quality step in the loop without handing control of the main agent to a plugin.

Install and setup

Native DSH path:

git clone https://github.com/Devin-AXIS/jev-dsh-decision.git
dsh plugin add /absolute/path/jev-dsh-decision

Or install the packaged .tgz: dsh plugin add /absolute/path/dsh-jev-decision-model-0.1.2.tgz. Then set your own TYPESAFE_API_KEY in DSH credential management or the launch environment (never in source or chat), restart the session, and ask the agent to “use Jev, check config first, then test the connection”. The plugin reuses DSH’s credentials, tools and skills services, so custom configurations must keep them.

iPolloWork path: import the signed package via Extensions → Import plugin package (publisher: iPolloWork), enter your TypeSafe API Key in the plugin details, and enable “Jev DSH Decision Engine” in the conversation.

Caveats

• You must supply your own TypeSafe API Key; check_connection sends one real, low-usage request, and “configured” locally does not mean the key is verified.
• A recommendation is not execution authorization — external actions still follow the host harness’s permission and confirmation flow.
• The repository declares no open-source license; OpenCode and Codex Harness support is currently via iPolloWork only, with no standalone native packages here.
• Maintained and published by iPolloWork; Jev belongs to TypeSafe AI, and the README explicitly states there is no official endorsement by TypeSafe AI, DeepSeek or other harness projects.