AgentForge: General-Purpose Agent Workflows and Engineering Skill Set

Introduction

When building agent applications, maintaining unified workflow standards, test design specifications, and code review processes often requires reinventing the wheel. AgentForge provides a curated set of integrated agent workflow and engineering skills aimed at solving these problems.

Overview

This is a generic agent workflow and engineering skill set that can be installed in Codex, Claude Code, and DeepSeek Harness. The plugin itself is JavaScript-based, has zero dependencies, and requires no additional MCP services. It provides standardized execution paths for different Agent tasks by reusing existing project testing and audit tools.

Core Features

The plugin includes eight core skills, covering the full process from research to implementation:

  1. Research and evidence: Handles code and log investigation, evidence-chain collection, and report generation.
  2. Test design: Designs test cases based on business rules and existing project facilities, assesses scenario gaps, and defines verification methods. This skill only delivers design; it does not write or execute tests.
  3. Bug debugging: Reproduces the issue and verifies the root cause. During execution, it presents the cause, evidence, and remediation plan, then performs repair and regression validation only after user consent.
  4. Architecture investigation: User-initiated architecture opportunity review, supports parallel evidence collection, and allows zero candidates.
  5. Implementation workflow: Supports task decomposition, serial or parallel implementation, and integration validation with risk-based auditing.
  6. Diff review: Reviews independently from three axes: Correctness (correctness), Standards (standards), and Spec (specification).
  7. Complexity audit: Evaluates complexity and coverage risk. When data is insufficient, it explicitly provides a fallback plan.
  8. Mutation testing: Invokes existing mutation testing tools to find surviving mutants and expose test gaps.

Installation and Activation

Install the plugin in DeepSeek Harness:

dsh plugin --profile web add github:Lixiuxiu559/AgentForge

After installation, the profile must be restarted for the plugin to take effect. Because the plugin is not published to npm, DSH has no version-number-based upgrade path.

Typical Usage

In conversation, directly invoke the corresponding skill:

  • Investigate an error call chain: Use research to investigate this error call chain, providing evidence and unconfirmed items.
  • Design test cases: Use test-design to design test cases for this feature, checking whether business boundaries, exceptions, and state changes are sufficiently covered.
  • Troubleshoot an exception: Use bug-debugging to troubleshoot this exception, first explaining the cause and fix plan, and waiting for my consent before making changes.
  • Implement a feature: Use implementation-workflow to implement this feature, allowing independent modules to run in parallel.
  • Review changes: Use diff-review to review the current changes, reporting the results for each of the three axes.
  • Architecture check: Use architecture-scout to check whether this module has boundaries worth abstracting.

Applicable Scenarios and Cautions

This plugin is suitable for projects that require structured workflows, unified testing standards, and rigorous review processes.

Pay special attention to the following points:
1. Isolated contexts: Independent contexts do not isolate the filesystem. Therefore, operations such as modifying the same file or sharing build artifacts still need mutual exclusion to avoid conflicts.
2. Git installation mechanism: Installation via git has no version gating; the lockfile pins a specific commit. The update operation tracks the latest commit on the main branch, effectively making the main branch the release channel.
3. Approval mechanism: The bug-debugging skill requires explicit user consent to the plan before repair. A simple “help me fix it” request does not automatically skip the approval step.

Future

AgentForge maintains skills uniformly under the skills/ directory, aiming to serve multiple platforms with a single codebase. For complete usage instructions and project structure, refer to the project homepage.

羽毛球分组比赛记分
小程序二维码

欢迎使用《羽毛球分组比赛记分》微信小程序

Xiaoye