dsh-plugin-adapter

1. 安装并启动 DeepSeek Harness：npx @deepseek-ai/dsh web
2. 在终端执行上面的安装命令（CLI 会解析插件并核验来源）
3. 用 dsh plugins list 确认已安装，必要时重启 Harness 生效

插件以当前 dsh 进程的权限运行，安装时可能执行代码。请先通读仓库源码与许可证，确认无破坏性命令与越权访问；本站只做索引，不对第三方插件安全性作担保。

事实信息来自公开插件目录快照（2026-10-01），介绍文案由本站再加工。

以下为插件仓库 README 全文（原始内容，由公开目录抓取整理）。

# dsh-plugin-adapter

**Free OpenCode Zen models, natively inside DSH (DeepSeek Harness).**

No API key. No registration. No extra process.

[![DeepSeek Harness](https://img.shields.io/badge/DeepSeek%20Harness-plugin-blue)](https://github.com/1624318455/dsh-plugin-adapter)

  [图片: license]

  [图片: Awesome]

  [图片: node]

  [图片: tests]

  [图片: stars]

  [图片: last commit]

English | [简体中文](README.zh-CN.md)

---

`dsh-plugin-adapter` registers a native DSH `LlmAdapter` that streams directly from
[OpenCode Zen](https://opencode.ai/zen)'s **anonymous free lane** — the same
models OpenCode's own CLI uses without an account, served to your DSH model
picker as a regular provider called `opencode2dsh`.

Requests leave your machine looking exactly like traffic from the OpenCode
CLI (same user agent, same correlation headers), and the model catalog stays
fresh through a three-tier fallback chain. There is nothing to log into and
nothing to host.

## Features

- **Zero credential, zero setup** — the anonymous lane needs no key; install, restart, chat
- **Native adapter, no sidecar** — one npm package, no child process, no binary, no local port (the legacy Go sidecar is not part of the published package; see `legacy/`)
- **CLI-identical disguise** — requests carry the OpenCode CLI user agent and its session/request/project header set, derived per conversation
- **Gateway-compat tracking** — Responses-only models auto-route to `/responses`, and a synced live session keeps the free lane open while Zen only serves known sessions
- **Live catalog with a fallback chain** — live upstream list ∩ free-by-metadata, falling back to offline cache and a verified static list
- **Self-healing** — fast startup retries, periodic refresh, and a written health snapshot for diagnostics
- **Proper error surfaces** — upstream failures (rate limit, auth, timeout, transport) arrive in DSH as classified finish reasons, and retries stay owned by DSH

## Requirements

DSH (DeepSeek Harness) with a web profile; Node.js ≥ 20 (already present if
DSH runs); outbound HTTPS to `opencode.ai` and `models.dev`.

## Install

**From the plugin market** (recommended, once listed): in DSH open
**Settings → Plugin Market**, search `dsh-plugin-adapter`, one-click install.

**From GitHub**:

```sh
dsh plugin --profile web add github:1624318455/dsh-plugin-adapter
```

**From npm**:

```sh
dsh plugin --profile web add @memef1f1y/dsh-plugin-adapter
```

**From source** (build the tarball yourself):

```sh
git clone https://github.com/1624318455/dsh-plugin-adapter.git
cd dsh-plugin-adapter/packages/plugin
pnpm install && pnpm pack
dsh plugin --profile web add ./memef1f1y-dsh-plugin-adapter-.tgz
```

**Verify**: restart `dsh web`, open the model picker, and pick a model from
the **opencode2dsh** group.

## Configuration

Defaults work out of the box. Override via the profile's `cordis.patch.yml`:

```yaml
- id: opencode2dsh
  name: '@memef1f1y/dsh-plugin-adapter'
  config:
    mode: adapter        # adapter (default) | sidecar
    providerId: opencode2dsh
    refreshSeconds: 300  # catalog refresh cadence
```

| Option | Default | Description |
| --- | --- | --- |
| `mode` | `adapter` | `adapter`: native LlmAdapter streaming straight from Zen. `sidecar`: legacy local-agent mode, not bundled — build the agent from `legacy/agent` and pass `agentPath`. |
| `providerId` | `opencode2dsh` | Provider name shown in DSH. |
| `refreshSeconds` | `300` | Live catalog refresh interval. Pricing metadata refreshes every 24 h. |
| `gatewaySession` | — | A live CLI session id sent as `x-opencode-session` (Zen only serves known sessions). Takes precedence over `gatewaySessionFile`. |
| `gatewaySessionFile` | — | File holding the session id, re-read every turn so an external helper can rotate it without a restart. |
| `agentPath` | auto-resolved | Sidecar only: path to the agent binary. |
| `agentArgs` | — | Sidecar only: extra CLI args for the agent. |
| `restartDelayMs` / `restartMaxDelayMs` / `maxConsecutiveCrashes` | `1000` / `60000` / `5` | Sidecar only: restart backoff and circuit breaker. |

## How it works

```
DSH session
   │  harness chunks (block-start / text-delta / usage / finish …)
   ▼
ZenAdapter (registered LlmAdapter)
   │  pi-ai openai-completions stream (chat models)
   │  pi-ai openai-responses stream (Responses-only models)
   ▼
https://opencode.ai/zen/v1        ← Authorization: Bearer public
   with CLI-identical headers:
     user-agent: opencode/… (CLI-identical, runtime values)
     x-opencode-client, x-opencode-session, x-session-affinity,
     X-Session-Id, x-opencode-request, x-opencode-project
```

- **Session correlation** — by default session/project ids are SHA-256 derived from the
  conversation's first user turn (stable per conversation, non-reversible),
  and each request gets a fresh random id, mirroring the CLI. When
  `gatewaySession`/`gatewaySessionFile` is set, the synced live session id is
  sent instead (Zen only serves sessions it has seen from a genuine CLI flow).
- **Catalog fallback chain** — S1: live `GET /v1/models`; S2: models.dev
  pricing metadata decides "free"; S3: a compile-time verified static list.
  A disk cache (~7-day TTL) covers upstream outages.
- **Resilience** — the adapter registers immediately at startup; if the first
  catalog fetch races your network (VPN/TUN reconnects, DNS), the plugin
  retries on a short cadence (~1 min) before settling into the periodic
  refresh. Responses models get a wider 300 s body-idle watchdog for bursty
  reasoning; chat models keep the 120 s default.
- **Sidecar mode** (`mode: sidecar`, legacy) — spawns a local Go agent (a
  single-tenant port of [opencode2api](https://github.com/jasonxu114514/opencode2api))
  on `127.0.0.1:`, token-authenticated, and registers a standard
  `llm-pi-ai` route. **Not part of the published package**; build it from
  `legacy/agent` (`go build ./cmd/agent`) and point `agentPath` at the binary.

## Edge cases handled

- **Responses-only models** (`muse-spark-*`): chat returns a bare 500 while
  `/responses` returns 200 — auto-routed, no config needed.
- **Unknown sessions**: unknown `x-opencode-session` ids get
  `403 FreeTierError` — sync a live CLI session via `gatewaySessionFile`.
  (Note: non-streaming probes always 403 — diagnose with `stream:true`.)

## Settings persistence

Adapter config lives in the profile's `cordis.patch.yml` (static per
install). The synced session file is plain text (one line) re-read every
turn. The IP-pool card (settings UI) owns the live routing section; catalog
health persists to `~/.opencode2dsh/adapter-status.json`.

## Health & troubleshooting

The plugin writes a health snapshot after every refresh round:

```
~/.opencode2dsh/adapter-status.json
```

```json
{
  "status": "ready",
  "total": 64,
  "exposed": 9,
  "lastError": "",
  "writtenAt": "2026-08-29T07:01:54.915Z"
}
```

| Symptom | Likely cause & fix |
| --- | --- |
| Boot screen shows `Failed to load plugins … list slot "settings.plugin.item" requires options.id` | Your DSH is too old (≤ 0.1.0-rc.6): upgrade DSH to ≥ 0.1.0-rc.7 (latest recommended). Model routing is unaffected. |
| Only 3 models | Startup fetch raced your network; retries land within ~1 min. Check `adapter-status.json` for `lastError`. |
| `lastError: "fetch failed"` persisting | Outbound HTTPS to `opencode.ai` blocked; check proxy/VPN rules. |
| Rate-limit errors in chat | The anonymous lane is quota-per-IP; switch network node or wait. |
| `500` on `muse-spark-*` via chat | Responses-only model; routed to `/responses` automatically. |
| `403 FreeTierError: free tier can only be used from within OpenCode` | Zen only serves gateway-known sessions: sync a live CLI session via `gatewaySessionFile` (re-run the sync helper when it recurs). |
| `stream body idle timeout` on reasoning models | Bursty chain-of-thought tripped the watchdog; Responses models use 300 s. If it persists, the exit node may be killing long SSE — switch nodes. |
| Connection error to `127.0.0.1:*` | A stale sidecar route shadows the adapter; plugin ≥ 0.2.1 removes it at startup. |
| Install fails with `ERR_PNPM_IGNORED_BUILDS` | A transitive dependency of `pi-ai` (`@google/genai`, `protobufjs`) has build scripts that are not needed at runtime. Approve-or-decline them via the plugin market, or set both to `false` under `allowBuilds:` in the profile's `pnpm-workspace.yaml`. |

## FAQ

- **Do I need an API key?** No. The anonymous lane's key is the literal
  string `public`; quotas are per exit IP.
- **Why did a working model suddenly 403/500?** Zen moves models between
  APIs and tightens fingerprinting without notice — update the plugin first,
  then check the table above.
- **muse-spark is slow?** It reasons with high effort before answering;
  the first tokens can take ~30 s. That is the model, not the plugin.

## Development

```sh
git clone https://github.com/1624318455/dsh-plugin-adapter.git
cd dsh-plugin-adapter/packages/plugin
pnpm install
pnpm typecheck && pnpm test
pnpm build                    # bundle to lib/
```

The legacy Go sidecar lives in `legacy/agent` (`go test ./...`). Architecture
notes and the porting record live in `docs/`.

Releasing: `pnpm pack` in `packages/plugin` (prepack builds and syncs docs).

Test status: full suite 162 passed; 3 environment-sensitive groups
(watchdog timing, live-subscription, airport fixture) fail without a
built bundle/network/slack and pre-date this fork.

## Known limits

- Free models may train on your data during the free period (Zen policy) —
  use privacy models or local runners for sensitive code.
- Responses models need a live synced session; a stale file yields 403
  until refreshed.

## Acknowledgments

- [opencode2dsh](https://github.com/FishBottle7/opencode2dsh) by
  [FishBottle7](https://github.com/FishBottle7) — the adapter, catalog and
  IP-pool design originate there; this project maintains that core with
  gateway-compatibility fixes.
- [**opencode2api**](https://github.com/jasonxu114514/opencode2api) by
  [@jasonxu114514](https://github.com/jasonxu114514) — the legacy Go sidecar
  in `legacy/agent` is a port of its anonymous-lane implementation.
- [OpenCode](https://opencode.ai) — for running the free anonymous Zen lane.
- [@earendil-works/pi-ai](https://www.npmjs.com/package/@earendil-works/pi-ai) — the wire layer used by adapter mode.
- [DeepSeek Harness](https://www.npmjs.com/package/@deepseek-ai/dsh) and the
  [dsh-market](https://github.com/dsh-market/dsh-market) community.

## Friends

**[LinuxDo](https://linux.do)** — 新的理想型社区 / a new ideal community

## License

[MIT](./LICENSE) © FishBottle7, © 1624318455