Introduction¶

DeepSeek Harness (DSH) adopts a plugin-based architecture, emphasizing modularity and flexibility. In real-world development, when managing AI skills, developers often face challenges such as inconsistent scoring criteria, difficult cost estimation, and configuration dependency drift. The Cromus plugin addresses these issues by providing DSH with an automated solution for scoring, validation, and cost estimation.

Plugin Overview¶

This plugin is maintained by cromus-ai and is released under the MIT license. It is a DeepSeek Harness bundle that allows users to quantitatively score, validate, and estimate the cost of AI skills directly within a local agent environment, without switching to external tools.

Core Capabilities¶

After installation, 11 tools appear under the mcp__cromus__ namespace in the model tool registry, covering the following functions:

mcp__cromus__score_skill: Returns a 0–1000 Croms score and a four-dimension breakdown.
mcp__cromus__simulate_cost: Estimates per-run and monthly costs for a specific workflow (supports five routing modes).
mcp__cromus__check_mcp_drift: Compares the MCP dependencies declared by a skill against live servers and flags changes.
mcp__cromus__check_model_drift: Detects retired, renamed, or superseded models declared in SKILL.md.
mcp__cromus__validate_skill: Checks SKILL.md files against the Open specification and lists violations.
mcp__cromus__validate_mcp_deps: Checks the completeness and valid configuration of the skill’s mcp_dependencies block.
mcp__cromus__validate_ethos: Checks the ETHOS.md governance file against the Open specification.
mcp__cromus__validate_memory: Checks the MEMORY.md handoff file against the Open specification.
mcp__cromus__validate_agents: Validates AGENTS.md files and returns a readiness score.
mcp__cromus__reality_check_summary: Returns the most recent Reality Check records for a skill (read-only).
mcp__cromus__check_usage_waste: Analyzes exported AI API usage data to identify waste patterns (cache utilization, model sprawl, idle burn).

Installation and Enablement¶

Install the plugin and configure an API key to enable its features:

Install the plugin:

    dsh plugins add @cromus-ai/dsh-plugin

Set the API key:

    export CROMUS_API_KEY=your_key_here

After the plugin is installed, the 11 tools appear in the model tool registry. The system requires Node.js version >= 18.

Typical Usage¶

Once installed, the tools above can be invoked directly within the DSH context. For example, you can invoke the scoring function when handling skill files, or manually configure the connection through cordis.patch.yml (optional):

- insert:
    - id: cromus-mcp
      name: '@deepseek-ai/dsh-mcp-client'
      config:
        serverName: cromus
        transport: streamable-http
        url: https://mcp.cromus.ai/mcp
        headers:
          Authorization: !!js '`Bearer ${process.env.CROMUS_API_KEY}`'

Use Cases and Considerations¶

This plugin is suitable for developers or teams that need strict governance of AI skills. When using it, note the following:
* The CROMUS_API_KEY environment variable must be set.
* The plugin runs with the permissions of the current DSH process. It is recommended to review the source code and license before installation.
* The skill directory has no official relationship with DeepSeek / Huanfang.

Conclusion¶

The Cromus plugin provides DeepSeek Harness with a complete toolchain from scoring to governance. For more technical details and integration options, see the GitHub repository.

Cromus DSH Plugin: AI Skill Scoring, Validation, and Cost Estimation