dsh-bridges

English | 中文

This project is implemented by DeepSeek Harness.

A DeepSeek Harness plugin that bridges projects already configured for Claude Code, CodeBuddy Code, opencode, or Codex into DeepSeek Harness, so existing skills, commands, memory, and hooks continue to work without any migration.

Quick start

# 1. install once into a DeepSeek Harness profile
dsh plugin --profile <name> add dsh-bridges

# 2. run DeepSeek Harness in a project already configured for another agent
cd my-claude-project        # has .claude/ assets
dsh --profile <name> "list the skills available in your catalog"
# → .claude skills and commands are registered as /name skills, CLAUDE.md is
#   injected, and the project's settings.json hooks run unchanged.

To install from a checkout of this repository, build it first: pnpm install && pnpm build && dsh plugin --profile <name> add .

Complete example projects for each agent tool are available in examples/ (claude-code, codebuddy-code, opencode, codex); open one as the session workspace to observe its skills, memory, and hooks being bridged.

All bridges are enabled by default and can be configured or disabled individually from any patch layer:

- id: bridges
  config:
    claudeCode:
      enabled: true     # master switch for this bridge
      skills: true      # .claude skills and commands
      memory: true      # CLAUDE.md memory
      hooks: true       # settings.json hooks

Full per-bridge configuration and behavior: docs/guides/

What it bridges

Assets already in your project	What DeepSeek Harness provides
.claude/ .codebuddy/ .opencode/ .agents/ skills and commands	model skill catalog + /name invocation
CLAUDE.md, CODEBUDDY.md, AGENTS.md chains and rules	session-start memory injection
settings.json, hooks.json, config.toml hooks	the same hooks at DeepSeek Harness lifecycles

Supported agents

Agent	Status	Skills / commands	Memory	Hooks
Claude Code	✅ phase 1	.claude/skills, .claude/commands (+ ~/.claude)	.claude/CLAUDE.md, ~/.claude/CLAUDE.md	settings.json hooks (SessionStart, UserPromptSubmit, Pre/PostToolUse(+Failure), Stop, SessionEnd)
CodeBuddy Code	✅ phase 2	.codebuddy/skills, .codebuddy/commands (+ ~/.codebuddy)	CODEBUDDY.md, ~/.codebuddy/CODEBUDDY.md, .codebuddy/rules/	settings.json hooks (SessionStart, UserPromptSubmit, Pre/PostToolUse(+Failure), Stop, SessionEnd)
opencode	✅ phase 3	.opencode/skills, .opencode/commands (+ ~/.config/opencode), command.* in opencode.json	AGENTS.md (+ CLAUDE.md fallback), instructions files	— (opencode has no hooks config; its plugin API is out of scope)
Codex	✅ phase 4	.agents/skills (cwd → repo root), ~/.agents/skills, /etc/codex/skills	~/.codex/AGENTS.md + per-directory AGENTS.md chain	hooks.json / config.toml hooks (SessionStart, SubagentStart, UserPromptSubmit, Pre/PostToolUse, Stop, SubagentStop, SessionEnd)

Resources

• Example projects, one per bridged agent tool: examples/
• Usage guide (per-bridge details, full config reference, what is not bridged yet): docs/guides/
• Bridge-target reference materials (official upstream specs): docs/reference/
• Contributor documentation (how to add a new agent tool, integration surface, pitfalls): docs/development/