lldois/dsh-jev

Semantic tool routing and typed System One decisions for DeepSeek Harness (DSH) powered by TypeSafe Jev.

Modeled after the Pi Jev extension (pi-jev), this plugin natively integrates TypeSafe Jev into DeepSeek Harness using Cordis service injection, DSH tool definitions, human slash commands, and pre-turn lifecycle hooks.

---

Features

• Typed Judgments (jev_evaluate tool): Run fast, calibrated System One decisions directly from any LLM turn in DSH using choice (categorical selection), noul (yes/no probability), and score (rubric scale).
• Skill Discovery (jev_find_skill tool): Semantically matches and suggests the most relevant specialized agent skills (SKILL.md) for any task without cluttering prompt context.
• Tool Discovery (jev_find_tools tool): Semantically evaluates and identifies relevant registered tools in DeepSeek Harness for a user task.
• Human Slash Commands (/jev):
  • /jev status — Displays TypeSafe Jev configuration, key origin, session metrics (requests, tokens, latency), and tool/skill counts.
  • /jev test [prompt] — Run live connectivity test or evaluate a prompt against Jev.
  • /jev skills [query] — Semantically search and rank available skills directly from the chat box.
  • /jev tools [query] — Semantically search and rank available tools directly from the chat box.
  • /jev auto [on|off] — Toggle automatic per-prompt skill suggestions.
  • /jev help — Display help.
• Auto Mode Hook (agent/pre-step): Opt-in background evaluation that semantically detects matching skills for incoming user prompts and injects guidance to invoke the relevant skill tool before executing actions.
• Post-Run Gate CLI (dsh-jev-gate / jev-gate): Standalone binary for CI/CD pipelines, subagents, or verification gates. Checks git diff (HEAD or cached), file, or stdin against natural language acceptance criteria using Jev probability (exits 0 on pass, 1 on reject, 2 on error).
• Graceful Fallback & Zero Crash: Fails open to local keyword heuristic shortlists when Jev is unconfigured or offline, ensuring uninterrupted agent execution.

Architectural Differences from Pi Jev

In keeping with DeepSeek Harness's design:

• Subagents: DSH possesses its own first-party native subagent service (@deepseek-ai/dsh-subagent with typert protocol and session trees); external RPC workflow scripts from pi-subagents are omitted in favor of DSH's native architecture.
• Model Selection: DSH manages model presets and reasoning effort via agent-default-model and settings.yaml; auto-model switching per prompt is omitted to prevent overriding user-configured preset pipelines.
• Compaction: DSH's session log projection is handled natively by @deepseek-ai/dsh-compaction.

---

Installation in DeepSeek Harness

1. Link or Install the Plugin

In your DSH profile directory (e.g. ~/.dsh/profiles/web):

# Install via GitHub or local link
pnpm add github:lldois/dsh-jev
# or link local workspace
pnpm add file:C:/Users/lldois/workspace/dsh-jev

2. Register in Profile Bundles

In ~/.dsh/profiles/web/package.json, add "dsh-jev" to dsh.profile.bundles:

{
  "dsh": {
    "profile": {
      "bundles": [
        "@deepseek-ai/dsh-base",
        "@deepseek-ai/dsh-web-app",
        "dsh-jev"
      ]
    }
  }
}

The bundled cordis.patch.yml will automatically insert id: jev into the Cordis plugin loader.

---

Setup & Credentials

Set your TypeSafe API key via any of the following methods:

1. Environment variable:

   export TYPESAFE_API_KEY=your_key_here
   

2. DSH secret file: Write the key to ~/.dsh/secrets/typesafe_api_key.
3. Cross-agent secret file: Write to ~/.pi/agent/secrets/typesafe_api_key.
4. Workspace .env: Located in workspace/chat/typesafe-eval/.env.
5. DSH Settings: Configured in ~/.dsh/settings.yaml under the jev: namespace.

Verify your setup by running:

/jev status

---

Tools

1. jev_evaluate

Used by the model to get structured decisions, classifications, triage, and scoring.

{
  "state": "The user reported an unexpected 500 error on /api/checkout",
  "questions": {
    "is_incident": {
      "type": "noul",
      "instructions": "Is this message reporting a production incident?"
    },
    "category": {
      "type": "choice",
      "instructions": "Which service does this error belong to?",
      "criteria": {
        "checkout": "Payments and checkout flow",
        "auth": "User authentication and tokens",
        "search": "Catalog search service"
      }
    },
    "severity": {
      "type": "score",
      "instructions": "Rate severity of the issue",
      "criteria": ["minor", "major", "critical"]
    }
  }
}

2. jev_find_skill

Used by the agent or user to semantically find specialized workflows and instructions for complex tasks.

{
  "query": "build accessible modal component in React",
  "threshold": 0.65
}

3. jev_find_tools

Used by the agent to find the most relevant tools registered in the current session.

{
  "query": "query database table schema and row count"
}

---

CLI Gate Checker (dsh-jev-gate / jev-gate)

Check git changes or command output against natural language criteria:

# Check git diff against acceptance criteria
dsh-jev-gate -c "All exports have TypeScript type annotations" -d

# Check piped test/lint output
npm test 2>&1 | dsh-jev-gate -c "Zero test failures and no unhandled rejections"

# JSON output with custom threshold
dsh-jev-gate -c "Documentation updated" -f ./README.md -p 0.85 --json

Exits with 0 on pass, 1 on reject, 2 on error (or 0 with --fail-open).

---

Running Tests

node --test test/*.test.js

---

License

MIT © lldois