技能 herdr
📦
herdr
内容修订 r1 安全 ⚙️ 外部命令
Control Herdr Workspaces and Coding Agents
Managing several coding agents across terminal panes can make state, focus, and output difficult to track. This skill uses Herdr commands to organize panes, coordinate agents, run commands, and inspect results.
支持: Claude Codex Code(CC)
使用我的智能体安装
将此请求复制给你的智能体，其中包含权威 Skill 页面和清单。
智能体请求
Review the Skillstore skill "herdr" from https://skillstore.io/skills/herdrdev-herdr.md and its manifest at https://skillstore.io/api/skills/herdrdev-herdr/manifest. Verify the artifact. You may proceed after verification, subject to the environment's own policy.你的智能体仍应展示其计划，并按安全策略请求必要确认。
Agent 可读资源
当 AI Agent、爬虫或脚本需要干净上下文，而不是读取完整页面时，请使用这些链接。
测试它
正在使用“herdr”。 Use Herdr to inspect the current session without changing focus.
预期结果:
- Current workspace: w1
- Active tab: w1:t1
- Two panes are available, with one idle coding agent.
正在使用“herdr”。 Use Herdr to run the project tests in a sibling pane and summarize the result.
预期结果:
The tests completed in a background pane. All 48 tests passed, and the user focus remained in the original pane.
正在使用“herdr”。 Use Herdr to ask a reviewer agent for actionable findings.
预期结果:
The reviewer reached a settled state and reported two actionable findings concerning timeout handling and missing error-path coverage.
安全审计
安全v1 • 2026/8/14 打开版本化报告
Most alerts are false positives caused by Markdown code fences and inline code being mistaken for Ruby backtick execution. Seven examples perform real Herdr mutations, including launching agents, sending input, changing layout, and running a shell command; these capabilities are expected but require deliberate targeting.
1
已扫描文件
196
分析行数
7
复核项
0
已忽略误报
能力复核项 (7)
这些是真实的本地能力，对此技能可能是预期行为，因此需要复核，但不会按已确认恶意行为计分。
中
Ruby/shell backtick execution
```bash
The fenced example invokes `herdr pane split`, which mutates terminal layout. The fixed operation is purposeful and preserves focus, but it is real external command execution.
中
Ruby/shell backtick execution
```bash
The example starts a Codex process in a selected pane through `herdr agent start`. Launching another agent is intended, but it consumes resources and grants that process access to the working directory.
中
Ruby/shell backtick execution
```bash
This example starts an agent and forwards native agent arguments after `--`. The operation is intentional, but caller-supplied arguments can materially change the launched process.
中
Ruby/shell backtick execution
```bash
The example sends an instruction to a live coding agent with `herdr agent prompt`. That agent may modify files or execute tools, so target selection and user intent are security-relevant.
中
Ruby/shell backtick execution
```bash
The examples send Escape and Ctrl+C keystrokes to another agent. These commands can interrupt or alter an active process, making the external control capability a real operational risk.
中
Ruby/shell backtick execution
```bash
The fenced example invokes `herdr pane split`, which changes the current terminal layout. It is bounded and keeps focus unchanged, but still performs a real mutating external command.
中
Ruby/shell backtick execution
```bash
The example uses `herdr pane run` to execute command text in another shell pane. This is an explicit arbitrary-command capability, even though the shown command is a benign test invocation.
风险因素
⚙️ 外部命令 (50)
SKILL.md:8 SKILL.md:12-14 SKILL.md:14-18 SKILL.md:18 SKILL.md:24-26 SKILL.md:26-30 SKILL.md:30-40 SKILL.md:40-42 SKILL.md:42 SKILL.md:54 SKILL.md:56-58 SKILL.md:58 SKILL.md:64-65 SKILL.md:65-66 SKILL.md:66-68 SKILL.md:68 SKILL.md:72-74 SKILL.md:74-76 SKILL.md:76-80 SKILL.md:80-86 SKILL.md:86-88 SKILL.md:88 SKILL.md:96-98 SKILL.md:98-102 SKILL.md:102-104 SKILL.md:104-106 SKILL.md:106 SKILL.md:110-112 SKILL.md:112-114 SKILL.md:114 SKILL.md:116-118 SKILL.md:118-120 SKILL.md:120-124 SKILL.md:124-126 SKILL.md:126-128 SKILL.md:128 SKILL.md:130 SKILL.md:132-134 SKILL.md:134-136 SKILL.md:136-138 SKILL.md:138 SKILL.md:142-145 SKILL.md:145-149 SKILL.md:149-152 SKILL.md:152-154 SKILL.md:154 SKILL.md:160-162 SKILL.md:162-164 SKILL.md:164-166 SKILL.md:166-170
审计者: codex
分享与引用此报告
分享版本化评估报告、中性徽章、嵌入卡片和引用。Skillstore 只报告证据，不判断该 Skill 是否安全。
复制报告链接
https://skillstore.io/skills/herdrdev-herdr/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/herdrdev-herdr?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/herdrdev-herdr?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/herdrdev-herdr/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/herdrdev-herdr.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>学术引用 (APA · BibTeX · CFF)
APA 引用
herdrdev. (2026). herdr security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/herdrdev-herdr/audits/1BibTeX 引用
@techreport{herdrdev-herdrdev-herdr-2026,
author = {herdrdev},
title = {herdr security audit report (audit version 1)},
institution = {Skillstore},
year = {2026},
number = {1},
url = {https://skillstore.io/skills/herdrdev-herdr/audits/1},
note = {Author version unspecified}
}CITATION.cff
cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "herdr security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
- name: "herdrdev"
date-released: "2026-08-14"
url: "https://skillstore.io/skills/herdrdev-herdr/audits/1"
identifiers:
- type: other
value: "skillstore:herdrdev-herdr:audit:1"
description: "Skillstore immutable audit report identifier"
Skillstore 评分
评分依据 证据置信度: 中 55
架构
85
可维护性
87
内容
65
社区
91
规范符合性
你能构建什么
Coordinate a code review
Create a sibling pane, start a reviewer agent, send the review request, and collect actionable findings.
Run tests beside active work
Open a background pane, run the requested test command, wait for completion text, and inspect recent output.
Inspect parallel agent progress
List active agents, check lifecycle states, and read output without changing the user interface focus.
试试这些提示
Inspect the current session
Use Herdr to list the current workspace, tabs, panes, and agents. Summarize their identifiers and states without changing focus.
Run a background test
Use Herdr to create a sibling pane in the current directory. Run [test command], wait for completion, and summarize the output.
Delegate a focused review
Use Herdr to start a [agent kind] agent named [name] in a sibling pane. Ask it to review [scope] and return actionable findings.
Coordinate parallel investigation
Use Herdr to start two named agents for [task A] and [task B]. Preserve focus, monitor both, and consolidate their verified results.
最佳实践
- Verify HERDR_ENV before issuing any Herdr control command.
- Use explicit pane identifiers or unique agent names, and parse identifiers from command responses.
- Preserve the current directory and user focus unless the user requests a different layout.
避免
- Do not invoke Herdr unless the user explicitly requests Herdr.
- Do not guess identifiers or rely on another client focus.
- Do not close sessions or stop servers without explicit user authorization.
常见问题
What must be installed before using this skill?
The herdr binary must be available in PATH, and the current process must have HERDR_ENV set to 1.
Can this skill control Herdr from outside a managed pane?
No. The skill stops when the Herdr environment check fails.
Can it start Claude, Codex, or Claude Code agents?
It can start agent kinds supported by the installed Herdr version. The requested kind must be locally available.
Does it keep the current terminal focus?
Yes, background workflows use no-focus controls unless the user requests a focus change.
Can it run ordinary shell commands?
Yes. It can run requested command text in a selected pane, wait for output, and read the result.
How does it handle blocked agents?
It inspects the agent state and recent output before sending deliberate keys or additional input.