delegate-grok

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Delegate to Grok

委托任务给Grok

Delegate through Herdr with one explicit mode:

```
review
```
: Keep Grok read-only. Use for plans, diffs, bugs, research, architecture, tests, adversarial analysis, or a second opinion.
```
implement
```
: Make Grok the sole writer. Stop editing until Grok finishes, then inspect and verify its work.

Default to

review

when authorization is ambiguous. Never infer permission to edit from phrases such as "ask Grok", "take a look", "analyze", or "check this". Use Grok 4.6 with

xhigh

reasoning by default.

通过Herdr以两种明确模式进行委托：

```
review
```
：设为Grok只读模式。适用于方案、差异对比、Bug、研究、架构、测试、对抗性分析或二次意见场景。
```
implement
```
：让Grok作为唯一编写者。在Grok完成前停止编辑，之后检查并验证其工作成果。

默认在授权不明确时使用

review

模式。绝不能从“ask Grok”、“看看这个”、“分析一下”或“检查这个”这类表述中推断出编辑权限。默认使用具备

xhigh

推理能力的Grok 4.6。

Prepare the handoff

准备交接工作

Create a self-contained task containing:

the objective and relevant project context;
the exact scope and files when known;
current evidence, errors, or competing hypotheses;
constraints and explicit non-goals;
required verification and return format.

Do not forward the entire conversation. Preserve only information needed for the delegated task. For opinion or research requests, distinguish sourced facts from inference and ask Grok to identify uncertainty.

Before

implement

, inspect the current workspace state and preserve unrelated or pre-existing changes. Tell Grok not to commit, check in, shelve, push, switch branches, or alter workspace configuration unless explicitly requested.

创建一个独立完整的任务，包含以下内容：

目标和相关项目背景；
已知的精确范围和文件；
当前的证据、错误或相互矛盾的假设；
约束条件和明确的非目标；
要求的验证方式和返回格式。

不要转发整个对话，只保留完成委托任务所需的信息。对于意见或研究请求，区分已查证的事实与推断内容，并要求Grok指出不确定性。

在

implement

模式前，检查当前工作区状态并保留无关或已存在的更改。告知Grok除非明确要求，否则不要执行commit、check in、shelve、push、切换分支或修改工作区配置的操作。

Run the delegation

执行委托操作

Resolve this skill's directory and invoke its script. Pass the task with

--task

--task-file

, or stdin.

bash

bash <skill-dir>/scripts/delegate-grok.sh review \
  --target-title "Grok 4.6 xhigh Reviewer" \
  --task "Review the current change for correctness, regressions, and missing tests."

bash

bash <skill-dir>/scripts/delegate-grok.sh implement \
  --task "Implement the agreed fix within the stated scope and run the relevant verification."

The script must:

Honor
```
--target
```
first, then
```
--target-title
```
. If a requested target is missing or busy, stop instead of falling back.
Without an explicit target, prefer a settled pane titled
```
Grok 4.6 xhigh Reviewer
```
, then reuse another idle Grok agent in the current Herdr workspace. Split only when no suitable Grok pane exists.
Start new panes with model
```
grok-4.6
```
and reasoning effort
```
xhigh
```
, submit the task, wait for settlement, and return visible output.
Leave the Grok pane open for follow-up work.

Do not manually duplicate these steps unless the script reports a concrete failure.

Default to one complete Grok turn per hypothesis. If the response lacks the required handoff sections, report an incomplete handoff; do not automatically send an open-ended continuation. On timeout, send Escape to cancel the active turn, collect its last output, and stop.

解析此skill的目录并调用其脚本。通过

--task

、

--task-file

或标准输入传递任务。

bash

bash <skill-dir>/scripts/delegate-grok.sh review \
  --target-title "Grok 4.6 xhigh Reviewer" \
  --task "Review the current change for correctness, regressions, and missing tests."

bash

bash <skill-dir>/scripts/delegate-grok.sh implement \
  --task "Implement the agreed fix within the stated scope and run the relevant verification."

脚本必须满足以下要求：

优先遵循
```
--target
```
参数，其次是
```
--target-title
```
。如果请求的目标不存在或忙碌，应停止操作，不要自动 fallback。
若无明确目标，优先选择标题为
```
Grok 4.6 xhigh Reviewer
```
的已就绪面板，然后复用当前Herdr工作区中其他空闲的Grok agent。仅当没有合适的Grok面板时才新建面板。
新建面板时使用模型
```
grok-4.6
```
和推理级别
```
xhigh
```
，提交任务，等待任务完成，然后返回可见输出。
保持Grok面板处于打开状态，以便后续工作。

除非脚本报告明确故障，否则不要手动重复这些步骤。

默认每个假设对应一次完整的Grok交互。如果响应缺少所需的交接部分，需报告交接不完整；不要自动发送开放式的后续请求。若超时，发送Escape键取消当前交互，收集最后输出后停止。

Handle the result

处理结果

For

review

, report Grok's findings alongside your own assessment. Do not implement findings unless the user separately authorizes changes.

For

implement

Resume ownership only after Grok settles.
Inspect the actual changed files and version-control state.
Run proportionate verification independently.
Report what Grok changed, what you verified, and any remaining risk.

If Grok is

blocked

, show the blocking output. Never auto-approve permissions, destructive actions, commits, check-ins, shelvesets, pushes, or branch changes.

If no Herdr environment is available, stop and explain that the skill must run from an agent pane inside Herdr.

对于

review

模式，将Grok的发现与你自己的评估一起报告。除非用户单独授权更改，否则不要实施Grok的发现。

对于

implement

模式：

仅在Grok完成任务后恢复所有权。
检查实际修改的文件和版本控制状态。
独立执行相应的验证操作。
报告Grok修改了什么、你验证了什么以及任何剩余风险。

如果Grok处于

blocked

状态，显示阻塞输出。绝不要自动批准权限、破坏性操作、commit、check in、搁置集、push或分支更改。

如果没有可用的Herdr环境，停止操作并说明此skill必须在Herdr内的agent面板中运行。