@zhang-guo-wen/dsh-claude-compat
Claude Code 综合兼容性：技能/规则发现 + /btw 命令（主机端）+ 设置界面（客户端）
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:zhang-guo-wen/dsh-claude-compat#ae35ac793b2466871657e2717ac48cc55af510e2Claude Compat 以 @zhang-guo-wen/dsh-claude-compat 发布，当前版本为 0.1.3-alpha.1。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
English | 中文
A standalone plugin for DeepSeek Harness (DSH) that makes the harness work with your existing Claude Code / Codex setup, and adds a Harness 兼容 settings page for managing MCP servers and prompt rules from the Web UI instead of hand-editing YAML.
It does not bundle @deepseek-ai/*; those resolve from the host harness at runtime.
Lists each MCP server with its scope, description, and running status, and lets you add, edit, enable, or disable one. Changes reach the running host with no restart. Server names are blurred here; the scope, status, and controls are not.
Paste Claude-compatible connection JSON; it is parsed and validated on save, so a typo is reported instead of stored.
A system-level prompt box plus master switches for Claude-rule and Codex-rule injection.
启动中 / 停止中 while the MCP child process
starts or stops, so the list never blocks.mcp_list, mcp_load, and
mcp_unload to start a server for the calling session only; the loaded tools never leak into another session.
Three loading modes trade tool-list stability against binding quality — see below.<root>/.claude/skills/** into the session skill catalog, folds
.claude/CLAUDE.md and ~/.claude/CLAUDE.md into the first request, and folds .claude/rules/** — including
paths:-scoped rules that activate once you read a matching file..codex/AGENTS.md and ~/.codex/AGENTS.md./btw. Ask a side question in a forked, continuable child subagent.A row's enable switch and the loading mode answer different questions: the switch says whether a server may be used at all, the mode says when an allowed server enters context.
| Mode | Behavior |
|---|---|
Load all (eager) | Allowed servers mount at session start; their tools are always in the request |
Dynamic insert (dynamic, default) | Allowed servers stay unmounted; mcp_load mounts one into the calling session, so its tools join the request — best tool binding, but the tool list changes once per load |
Lazy (lazy) | Allowed servers stay unmounted; mcp_load connects over the MCP SDK without registering anything and returns the tool schemas, and the model calls them through the fixed mcp_call proxy — the tool list never changes, so the request-cache prefix is never invalidated |
Set it in 设置 → Harness 兼容 → MCP 管理 → MCP loading. The choice is stored in the user's
context-injection settings namespace and swaps the tool set from the next request on, in every session.
Measured on one preset with four MCP servers: dynamic sends 29 tools on the first request (built-ins plus
mcp_list/mcp_load/mcp_unload), eager sends 378, 348 of them MCP tools.
The built lib/ is committed, so the repository installs and runs directly — no build step on your machine.
# over HTTPS
npx @deepseek-ai/dsh plugin --profile web add git+https://github.com/zhang-guo-wen/dsh-claude-compat.git
# or over SSH
npx @deepseek-ai/dsh plugin --profile web add git+ssh://git@github.com/zhang-guo-wen/dsh-claude-compat.git
Pin a release tag so a later work-in-progress commit on the default branch is not picked up:
npx @deepseek-ai/dsh plugin --profile web add "git+ssh://git@github.com/zhang-guo-wen/dsh-claude-compat.git#v0.1.3-alpha.1"
To develop against a local checkout, install the directory. pnpm creates a symlink, so a rebuilt lib/ reaches
the host on the next start with no reinstall:
npx @deepseek-ai/dsh plugin --profile web add /absolute/path/to/dsh-claude-compat
Then restart the host and open the settings page:
npx @deepseek-ai/dsh web
You do not edit the profile manifest by hand: dsh plugin add adds both the dependency and the bundle entry.
Remove it, dependency and layer together, with dsh plugin --profile web remove @zhang-guo-wen/dsh-claude-compat.
Every field has a working default; the table is for overriding one. Fields marked with a settings-page control can be changed there instead of in the composition.
| Field | Default | Meaning |
|---|---|---|
providerName | claude-code | Unique provider name registered on ctx.skills |
claudeHome | $CLAUDE_HOME or ~/.claude | Claude Code home, scanned for skills and CLAUDE.md |
codexHome | $CODEX_HOME or ~/.codex | Codex home, scanned for AGENTS.md |
projectRootMarkers | ['.git'] | Directory entries that identify the project root |
includeProjectRoot / includeGlobalRoot | true | Scan the project / user .claude/skills root |
includeProjectRule / includeGlobalRule | true | Load the project / global CLAUDE.md rule |
includeProjectRules / includeGlobalRules | true | Fold the project / user .claude/rules/** tree |
maxRuleSourceBytes | 1048576 | Maximum UTF-8 bytes read from one rule file |
maxRuleRenderBytes | 262144 | Maximum UTF-8 bytes rendered in one rules batch |
claude / codex | true | Rule-injection master toggles (settings page) |
mcpLoading | dynamic | MCP loading mode (settings page) |
maxQuestionBytes | 4096 | Maximum UTF-8 bytes in the /btw side question |
provider | fork | The ctx.subagents fork provider name used by /btw |
- name: '@zhang-guo-wen/dsh-claude-compat'
config:
mcpLoading: lazy
.claude/skills is discovered when the catalog is listed; an add, rename, or delete is picked
up on the next discovery..claude/CLAUDE.md, ~/.claude/CLAUDE.md, and .codex/AGENTS.md are read at the
first request; later edits are not re-read mid-session.read only — write and edit do not activate them.npx -y … / uvx … usually 1–3s). The UI stays
responsive; installing the server as a direct executable shortens it.AGENTS.md owns the build, the Cordis/Typert plugin contract, and the pitfalls.
npm run build # host (tsdown) + client (rolldown ModuleLoader handoff)
npm run typecheck
Apache License 2.0 — see LICENSE. This product includes MIT-licensed portions derived from DeepSeek Harness; see NOTICE. Not affiliated with or endorsed by Claude Code, Codex, or their owners.