dsh-max-token-auto-continue
社区 DSH 主机插件，在达到最大令牌数截断后继续 root-agent 会话。
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:Wepoi9/dsh-max-token-auto-continue#2402aca382ce03f7558f52bd915fb98cb131ec3fMax Token Auto Continue 以 dsh-max-token-auto-continue 发布，当前版本为 0.1.0。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。
点击下方 Badge 复制 Markdown，粘贴到 README 即可。
验证 package.json 声明的 GitHub 仓库，即可管理这个公开页面。认领后，Hub 会优先安排当前版本的安全扫描，并在通过后公开展示结果。
认领这个 Plugin →English | 日本語
A community host plugin for DeepSeek Harness (DSH) that automatically continues a root-agent session after a turn ends because the model reached its maximum output-token limit.
This is a community-maintained plugin, not part of the DeepSeek Harness core distribution.
When a root-agent turn ends with turn/end reason max-tokens, the plugin waits for the agent to become idle and then continues through the appropriate DSH path:
agent.followup();/goal: GoalService.resume().The continuation is bounded by maxConsecutive and is cancelled when newer human input or a newer turn supersedes the pending continuation.
source.kind: "dsh-max-token-auto-continue" and form: "notice".| Setting | Default | Meaning |
|---|---|---|
enabled | true | Enables automatic continuation |
maxConsecutive | 3 | Maximum automatic continuations in one consecutive max-token chain |
The continuation text and delay are intentionally fixed rather than configurable.
Latest tested DSH version: 0.2.1-alpha.1.
| DSH version | Verification |
|---|---|
| 0.1.6-alpha.2 | Supported compatibility target |
| 0.1.7-alpha.2 | Plugin load and inventory visibility verified |
| 0.1.7-rc.1 | Plugin load and inventory visibility verified |
| 0.1.7-rc.2 | Plugin load and inventory visibility verified |
| 0.2.0-rc.1 | Plugin load, inventory visibility, and real max-token auto-continue E2E verified |
| 0.2.0-rc.2 | Plugin load and inventory visibility verified |
| 0.2.1-alpha.1 | Build, 13 tests, Web-profile config output, and startup load verified |
Actual auto-continue and Goal resume firing on 0.2.1-alpha.1 has not yet been re-verified. Newer DSH versions are not assumed compatible until verified.
Install dependencies and build:
npm install
npm run build
npm test
Add the local checkout to the target profile:
dsh plugin --profile <profile> add <absolute-path-to-this-repository>
dsh --profile <profile> --dump-config
For the Web profile, replace <profile> with web.
The bundle declaration lets DSH add the required profile entries. To override plugin settings, add a config entry with id dsh-max-token-auto-continue to the target profile's cordis.patch.yml.
This plugin intentionally continues work after an output truncation without waiting for another human message. That is useful for long agent tasks, but it can also extend an unintended task if the model reaches max-tokens while already heading in the wrong direction.
The main bounds are root-agent-only scope, supersession by newer input/turns, fail-closed behavior, and maxConsecutive.
npm run build
npm test
TypeScript is compiled to lib/; the distributable plugin entry is lib/src/index.js.
@deepseek-ai/dsh-llm is resolved at runtime from the DSH process entry so the plugin uses the host runtime copy.
The plugin does not persist session data and does not send data to external services. Its only session mutation is the bounded continuation message or Goal resume call described above.
Bug reports and focused pull requests are welcome. For compatibility reports, include the DSH version, profile, whether the session was a normal session or /goal, the observed turn/end reason, and whether a continuation was expected.
MIT. See LICENSE.