Six engineering skills for Claude Code and DSH: research, architecture scouting, implementation orchestration, three-axis diff review, complexity audit and mutation testing. Skills delegate through the host’s generic subagent tools when available.
Install
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:Lixiuxiu559/AgentForge
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
Engineering workflows for your coding agent
AgentForge is an engineering skills plugin for coding agents. Build a Project Wiki to preserve code knowledge, then use practical workflows to implement and validate changes. The skills cover understanding a repository, building features, debugging, reviewing changes and assessing tests.
Install it in Codex, Claude Code or DeepSeek Harness and work in your existing repository. You set the goal; the agent uses the skills to organize investigation, implementation and validation, then reports results and unresolved issues.
Get started · Project Wiki · All skills · Installation and updates
What you can do
Project knowledge that developers and agents can navigate
When joining an unfamiliar repository, investigating complex modules or preparing for future development, use wiki to generate a Project Wiki from the actual code. It explains architecture, module responsibilities, key flows and development practices, with source references for key conclusions.
- A code guide for developers: navigate from the project overview to architecture, modules and flows, then follow source links to the implementation.
- A knowledge entry point for agents: use the catalog and page summaries to choose what to read, and check the code before making key decisions.
- Knowledge maintained alongside code: Markdown lives in the repository's
docs/wiki/, ready to commit, review and update when the code changes.
Use wiki to generate an English Project Wiki for this repository, covering
architecture, core modules, key flows and a development guide.
For an existing Wiki:
Use wiki to update the pages affected by these code changes,
verify the source references, and run check when done.
check validates the catalog, pages, language and links. Content accuracy still requires checking the code. Explore AgentForge's own Project Wiki (Chinese).
Deliver features with clear validation
With a clear goal, the agent assesses the impact, breaks down the work, implements changes and runs relevant checks. Independent tasks can use the host's sub-agent capability to run in parallel. The agent then integrates the work and explains what changed, how it was validated and what remains.
Use implementation-workflow to add CSV export to this project. Follow the existing
permission and filtering rules, add necessary tests, and report changes and validation results.
Base fixes on causes and evidence
Start with a specific symptom, reproduce it, trace the code paths and validate the root cause. bug-debugging presents evidence and a repair proposal, waits for your approval, then implements the fix and checks the original scenario.
Use bug-debugging to investigate this intermittent request timeout.
First explain the cause, evidence and proposed fix.
Review the important questions before committing
diff-review checks correctness, repository standards and requirements separately. When you need more evidence, audit complexity and coverage risks, or use the project's existing mutation-testing tools to assess whether tests catch logic errors.
Use diff-review to review the current changes. Report correctness,
standards and requirements findings separately.
How the skills can work together

This illustrates capabilities; it is not a host recording. Choose the skills your task needs. Running the entire sequence is optional. View the still image.
Use pm to clarify requirements, or prototype and architecture-design to explore pages and technical approaches in readable HTML. If the requirements are already clear, you can go straight to implementation.
Get started
Codex
codex plugin marketplace add Lixiuxiu559/AgentForge
codex plugin add agentforge@agentforge
Start a new task after installation and try one of the examples above.
Claude Code
claude plugin marketplace add Lixiuxiu559/AgentForge
claude plugin install agentforge@agentforge
Restart Claude Code after installation. You can also select a skill explicitly with commands such as /agentforge:implementation-workflow.
DeepSeek Harness
In the desktop app, open Plugins → Add plugin in the sidebar, enter github:Lixiuxiu559/AgentForge, then install and enable it.
Web / CLI:
dsh plugin --profile web add github:Lixiuxiu559/AgentForge
Restart the corresponding profile after installation. See Installation and updates (Chinese) for updates, local installation, pinned versions and troubleshooting.
Choose from fourteen skills
| What you want to do | Skill | What you get |
|---|---|---|
| Build and maintain project knowledge | wiki |
A Project Wiki for developers and agents: navigation, architecture and module guides, source references and updates based on code changes |
| Pressure-test a plan or decision | grilling |
Check facts, ask one question at a time with a recommendation, and let you decide and confirm agreement |
| Implement a feature | implementation-workflow |
Task breakdown, code changes, integration checks and a delivery report |
| Diagnose and fix a bug | bug-debugging |
Reproduction and root-cause evidence, a proposal for approval, then an approved fix and regression checks |
| Review changes before committing | diff-review |
Separate findings for correctness, standards and requirements |
| Investigate code or technical questions | research |
Evidence-based conclusions from multi-step investigation and cross-checking |
| Turn an idea into a reviewable design | pm · prototype · architecture-design |
A PRD, page prototype or technical design, delivered as HTML by default |
| Assess test protection | test-design · mutation-testing |
Gaps in business scenarios and assertions, or an audit using existing mutation-testing tools |
| Assess change risk and refactoring opportunities | complexity-audit · architecture-scout |
Complexity and coverage risks, or evidence-based architecture candidates; architecture investigation requires an explicit user request |
| Prepare a separate development directory | worktree |
Create or safely remove a Git worktree while preserving the branch |
Use grilling to pressure-test this plan. Ask one question at a time, give a
recommendation for each, and wait for my answer before continuing.
grilling delivers agreed decisions. Use pm for a PRD and request execution separately; it does not start another workflow automatically.
Work in your existing development environment
Skills are maintained in Markdown and share one source of methods and references. The bundled JavaScript has zero dependencies, and no MCP service is required. Your host provides the model, tools and execution permissions. Scripts use Node.js; Git operations require Git.
Parallel execution depends on the host's generic sub-agent capability; otherwise, work proceeds sequentially. Audits reuse the project's existing tools and report limitations when tools or data are missing. Review and audit skills report findings; fixing them is a separate task.
Contribute
node tools/doctor.mjs
npm test
Structural checks validate plugin manifests, versions and skill resources. Tests cover skill loading, invocation policies, Wiki tools and release scripts. Model behavior in a real host still requires direct validation.
Project Wiki · Architecture · Releasing · Changelog · Artwork sources — supporting documentation is primarily in Chinese.
License
MIT · Copyright (c) 2026 lixiuxiu
Links
More in this category
loopx-project/loopx#dsh-loopx-plugin★ 6537
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
Q00/ouroboros#integrations/dsh-plugin★ 6308
Config-only bundle that mounts Ouroboros through the DSH MCP client, exposing 36 interview, Seed, execution, evaluation, and evolution workflow tools in DSH.
chuspeeism/dashi-taskboard#deepseek-harness★ 3434
Embeds the active installed Codex Taskboard runtime in the DeepSeek Harness sidebar, using its launcher runtime descriptor instead of a fixed port.
NanmiCoder/dsh-agent-teams★ 2647
AgentTeams multi-agent teams.
EthanYoQ/AI-Novel-Writer#dsh-ai-novel-writer★ 2265
Installs a dedicated AI novel-writing preset and workbench: revisioned local project assets, a compact side drawer, and native approval-gated single-file changes.
Miaotofu01/Study-Mate★ 1307
Your AI learning companion: helping you plan your studies, understand concepts, and build projects to master a subject.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.