Introduction

One of the core design philosophies of DeepSeek Harness (DSH) is “everything is a plugin,” extending agent behavior by combining native capabilities with external plugins. When handling full-stack engineering tasks, common pain points include a lack of structured execution flow, insufficient boundary constraints for sensitive operations, and completion states lacking supporting evidence. adithya-hmt/fullstack-expert is designed to address these issues. It is not a smarter model, but a rigorous engineering loop and safety layer built around DSH’s existing capabilities.

Plugin Positioning

Name: adithya-hmt/fullstack-expert
Core value: A Cordis-native (Cordis-framework-based), evidence-driven full-stack engineering specification.
Positioning: It provides DeepSeek Harness coding agents with a workflow and safety mechanism, formalizing repository inspection, vertical-slice planning, sensitive operation boundaries, and verification evidence recording into a process.

Core Features

The plugin provides a complete set of engineering practices, mainly including the following aspects:

  • Scoped methodology prompts: Defines a complete closed loop from code understanding, current-state inspection, architecture design, planning, implementation, testing, review, to final verification.
  • Key operations:
    • fullstack_plan: Used for substantive work, performing a bounded repository inspection and creating a vertical-slice plan.
    • fullstack_check: Used for structured inspection, command verification, or Supabase checks, while recording audit facts that are not verification evidence.
  • Safety and governance:
    • Sensitive operation boundaries: Clearly distinguishes sensitive operations.
    • Permission classification: Categorizes operations.
    • Approval-aware pre-execution gate: Checks approval status before executing sensitive operations and refuses execution if approval cannot be obtained.
    • Complexity governance decisions: Makes decisions based on requirements, reuse, dependencies, and safety controls.
  • Evidence and verification:
    • Verification evidence recording: Completion requires evidence recording through the native Harness.
    • Read-only Supabase project evidence detection: Detects Supabase project status in read-only mode.
    • Session and tool-result observation: Observes via Harness session/event and tools/result.

Installation and Enablement

Before installing, ensure that the DeepSeek Harness CLI is installed. You can install the plugin as a profile set using the following command:

dsh plugin --profile web add git+https://github.com/adithya-hmt/fullstack-expert.git#v0.1.1

After installation, you can confirm that the configuration is active with dsh --profile web --dump-config, then start Harness. If manual configuration is required, ensure that the host orchestration includes the native Harness rows (such as dsh-fs, dsh-shell, and dsh-tools), and overlay the plugin’s provided cordis.patch.yml into the configuration.

Usage

When using this plugin, follow the specific operational flow to ensure engineering standards are implemented:

  1. Prepare the orchestration environment: Start an orchestration environment that includes the plugin and the native Harness filesystem, tools, and skills rows.
  2. Submit the task: Request feature development, bug fixes, code review, refactoring, or application building from the agent.
  3. Execute planning: For substantive work, the agent should invoke fullstack_plan to inspect the repository and create a vertical-slice plan.
  4. Structured checks: Use fullstack_check to perform read-only checks on structure, commands, or Supabase, and record audit facts.
  5. Completion verification: A task is considered complete only when verification evidence recording through the native Harness passes and there are no relevant failures.

Applicable Scenarios and Considerations

This plugin is suitable for full-stack development scenarios that require strict engineering standards and audit trails. When using it, keep the following points in mind:

  • Capability boundaries: The plugin itself does not provide a browser driver, Shell executor, filesystem backend, Dev Server registry, Supabase SDK, database client, deployment adapter, or model provider. It reuses the host Harness’s capabilities. Browser verification is currently limited to guidance and evidence recording until a stable browser tooling composition is available.
  • Supabase operations: Supabase operations are currently limited to detection and recommendations. Sensitive operations (such as migrations, logins, and pushes) must be explicitly performed by the user.
  • Model configuration: The plugin assumes that GPT-5.6 Max is already configured in the Harness model routing and that configuration files do not contain credentials.
  • License: The plugin is licensed under the MIT License.

Conclusion

Fullstack Expert helps Harness agents move from “guesswork” coding to “evidence-driven” engineering practices by defining clear operation nodes (fullstack_plan, fullstack_check) and strict verification evidence requirements. For teams pursuing code quality and operational safety, this is a plugin worth adding to their workflow.