技能 invoking-agents
📦
invoking-agents
内容修订 r1 高风险 ⚙️ 外部命令🌐 网络访问📁 文件系统访问
Orchestrate Issues Through Review and QA
Multi-agent issue delivery can lose state, mix roles, and skip safeguards. This skill coordinates implementation, review, QA, escalation, and pull request creation through agent-flow.
支持: Claude Codex Code(CC)
安装前请停下并请求确认。
请审查安装计划，并在更改文件前获得用户明确同意。
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 "invoking-agents" from https://skillstore.io/skills/drix10-invoking-agents.md and its manifest at https://skillstore.io/api/skills/drix10-invoking-agents/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.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“invoking-agents”。 Implement issue #42 and open a pull request.
预期结果:
Issue #42 completed. Implementation, review, and QA passed. Pull request #108 was opened from the issue branch.
正在使用“invoking-agents”。 Run issue #57 through the pipeline.
预期结果:
Issue #57 needs a human decision because the acceptance criteria conflict. The decision brief includes the exact ambiguity and attempted resolution.
正在使用“invoking-agents”。 Process issues #60 and #61 in parallel.
预期结果:
- Issue #60 started in its own worktree.
- Issue #61 was serialized because both classifications include the same file.
安全审计
高风险v1 • 2026/9/27 打开版本化报告
The 81 static alerts are false positives caused by Markdown formatting, safe documentation references, or fixed workflow checks. Two semantic risks remain: unconfined implementer processes and unsafe shell interpolation of an untrusted issue title.
1
已扫描文件
200
分析行数
0
复核项
0
已忽略误报
已确认安全问题 (2)
高
Unconfined Implementer Processes
Claude and Codex implementers run without enforced worktree confinement while processing untrusted issue content. Environment variables and prompts are advisory, so a compromised agent could modify files outside the worktree.
The skill explicitly states that Claude and Codex lack worktree confinement and recommends an external container or sandbox. Their launch commands do not enforce that boundary.
高
Untrusted Issue Title Interpolated Into Shell Command
The pull request command places the untrusted issue title inside a double-quoted shell argument. Shell substitutions within a title could execute unless the value is escaped or passed without shell parsing.
The skill declares issue text untrusted, then shows the issue title substituted into a shell command. Double quotes do not suppress shell command substitution.
风险因素
⚙️ 外部命令 (50)
SKILL.md:12 SKILL.md:13 SKILL.md:14 SKILL.md:17 SKILL.md:24 SKILL.md:25 SKILL.md:27 SKILL.md:29 SKILL.md:31-35 SKILL.md:35-37 SKILL.md:37-41 SKILL.md:41 SKILL.md:42 SKILL.md:43-49 SKILL.md:49-54 SKILL.md:54-58 SKILL.md:58-63 SKILL.md:63-67 SKILL.md:67-72 SKILL.md:72-74 SKILL.md:74 SKILL.md:76 SKILL.md:78 SKILL.md:82 SKILL.md:84 SKILL.md:85 SKILL.md:89-91 SKILL.md:91-93 SKILL.md:93 SKILL.md:97 SKILL.md:99-103 SKILL.md:103-105 SKILL.md:105 SKILL.md:107-111 SKILL.md:111-113 SKILL.md:113 SKILL.md:115-117 SKILL.md:117-121 SKILL.md:121-123 SKILL.md:123 SKILL.md:125 SKILL.md:131-132 SKILL.md:132 SKILL.md:133-134 SKILL.md:134-137 SKILL.md:137 SKILL.md:141-147 SKILL.md:147-149 SKILL.md:149-151 SKILL.md:151-152
🌐 网络访问 (1)
📁 文件系统访问 (6)
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/drix10-invoking-agents/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/drix10-invoking-agents?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/drix10-invoking-agents?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/drix10-invoking-agents/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/drix10-invoking-agents.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
drix10. (2026). invoking-agents security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/drix10-invoking-agents/audits/1BibTeX 引用
@techreport{drix10-drix10-invoking-agents-2026,
author = {drix10},
title = {invoking-agents security audit report (audit version 1)},
institution = {Skillstore},
year = {2026},
number = {1},
url = {https://skillstore.io/skills/drix10-invoking-agents/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: "invoking-agents security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
- name: "drix10"
date-released: "2026-09-27"
url: "https://skillstore.io/skills/drix10-invoking-agents/audits/1"
identifiers:
- type: other
value: "skillstore:drix10-invoking-agents: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
83
规范符合性
你能构建什么
Deliver a GitHub Issue
Run one tracked issue through implementation, independent review, QA, and pull request creation.
Process a Local Work Item
Use acceptance criteria from local issue files when a GitHub issue is unavailable.
Enforce Review Separation
Keep implementation, review, and QA in separate processes with file-based evidence and bounded review rounds.
试试这些提示
Run One Issue
Implement issue #{issue_number} through the full agent-flow pipeline and open a pull request.Use Local Acceptance Criteria
Process the work item in {issue_file} as issue {issue_number}, then implement, review, test, and prepare a pull request.Resume an Escalated Issue
Resume issue #{issue_number} after this decision: {decision}. Preserve existing artifacts and continue from the recorded state.Coordinate Parallel Issues
Classify issues {issue_numbers}, identify overlapping files, run independent worktrees in parallel, and serialize any conflicting changes.最佳实践
- Provide testable acceptance criteria before starting the pipeline.
- Use enforced containers or sandboxes for every agent that can edit or execute commands.
- Review escalation briefs and critical draft pull requests before continuing.
避免
- Do not combine implementer, reviewer, and QA responsibilities in one agent context.
- Do not bypass protected-path checks, review-round limits, or required human approval.
- Do not interpolate issue text directly into shell commands.
常见问题
Which issue sources are supported?
The skill accepts inline criteria, GitHub issues through gh, and local ISSUES.md or .agent-issues.json files.
Does it implement changes itself?
No. It coordinates separate implementer, reviewer, and QA processes and exchanges artifacts between them.
When does it request human input?
It escalates missing specifications, architecture decisions, protected-path changes, critical changes, repeated failures, and unavailable GitHub operations.
Can it create pull requests?
Yes. It pushes the issue branch and uses gh after implementation, review, and QA complete.
Does it support parallel issues?
Yes. Independent issues can use separate worktrees, but issues touching the same files should run sequentially.
Is worktree confinement automatic?
Only some harnesses enforce it. Claude and Codex implementers require an external container or sandbox for a hard boundary.