Token导航 LogoToken导航

Agent Skill · 研究检索

a2a-role-auditor(a2a 角色审核员)

用于辅助安全审计、权限检查、凭据风险、认证流程和常见漏洞排查。它适合让 Agent 梳理敏感配置、检查依赖风险、分析鉴权逻辑或生成安全复核清单。使用时不能把工具输出直接当最终结论,涉及密钥、令牌、用户数据或生产系统时,应先确认最小权限、脱敏方式和操作边界。

a2a-role-auditor 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为只读;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 71,596。

总安装 194周安装 8GitHub Stars 9下载 63核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:a2a-role-auditor(a2a 角色审核员)
来源仓库:https://github.com/i9wa4/dotfiles
仓库路径:skills/a2a-role-auditor
安装命令:
npx skills add https://github.com/i9wa4/dotfiles --skill a2a-role-auditor
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/i9wa4/dotfiles --skill a2a-role-auditor

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

用于辅助安全审计、权限检查、凭据风险、认证流程和常见漏洞排查。适合让 Agent 梳理敏感配置、检查依赖风险、分析鉴权逻辑或生成安全复核清单。使用时不能把工具输出直接当最终结论,涉及密钥、令牌、用户数据或生产系统时应先确认最小权限。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

来源仓库内路径:skills/a2a-role-auditor

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/i9wa4/dotfiles --skill a2a-role-auditor;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

a2a-role-auditor 怎么安装?

在终端执行 npx skills add https://github.com/i9wa4/dotfiles --skill a2a-role-auditor。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

a2a-role-auditor 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

a2a-role-auditor 需要什么权限?

权限级别是只读。该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

a2a-role-auditor 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

a2a-role-auditor 用的人多吗?

公开数据是累计安装 194 次,近一周安装 8 次,下载 63 次,GitHub 9 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

a2a-role-auditor 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 只想先试试效果、不希望它改动本地文件的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
a2a-role-auditor

a2a 角色审核员 · 本页

1949只读未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

a2a-role-auditor

Audits nodes/*.toml role templates in a tmux-a2a-postman project to fix node-to-node interaction breakdowns.

1. Mandatory Triage Gate

Before running any template audit, determine whether the issue is daemon-level or template-level.

Daemon-level indicators (stop here — report as config issue, do NOT produce patches):

  • Wrong or missing edges in postman.toml
  • nodes/{node}.toml file does not exist
  • Session disabled in postman TUI

Template-level confirmed (node exists, edges correct, but behavior is wrong):

  • Proceed to the 7-check audit below.

2. 7-Check Audit

2.1. Pre-check: File Existence (binary)

For every node referenced in postman.toml edges, verify nodes/{node}.toml exists.

  • PASS: file present
  • FAIL: file missing → emit BLOCKING finding; abort all further checks for that node

2.2. Check 1 — PONG Awareness

The daemon calls MarkPongReceived() only when info.To == "postman" (message.go). A node that does not send TO postman as the explicit recipient is never marked PONG-active and remains invisible in other nodes' talks_to_line.

  • PASS: template explicitly instructs the node to send a message with "postman" as recipient (e.g., "Send a PONG addressed to postman")
  • FAIL: template references postman only as a routing mechanism ("via postman"), or has no postman mention at all

2.3. Check 2 — Routing Clarity

  • PASS: template names at least one recipient for output messages
  • FAIL: template says "send a message" without specifying who receives it

2.4. Check 3 — Completion Protocol

  • PASS: template specifies a machine-readable signal word (e.g., APPROVED, DONE, BLOCKED) for task completion
  • FAIL: completion state is undefined or described only in natural language

2.5. Check 4 — Fallback Routing

  • PASS: template names an alternative recipient when the primary contact is absent from talks_to_line
  • FAIL: no fallback specified

2.6. Check 5 — Cross-Edge Consistency

Two sub-checks:

  • Binary: does the template mention only nodes that exist as edges in postman.toml? (PASS/FAIL — no judgment)
  • Judgment: are the described routing semantics consistent with edge direction? (LLM assessment — label findings with Type: JUDGMENT-BASED)

2.7. Check 6 — on_join Completeness

  • PASS: on_join field is non-empty
  • FAIL: on_join = ""

3. Findings Format

Every finding MUST use this exact schema:

[SEVERITY] Node: {node}
Field: nodes/{node}.toml:[{node}].{field}
Check: {check name}
[Type: JUDGMENT-BASED]   <- optional, only when applicable
Result: FAIL
Issue: {description}
Fix:
  {exact replacement text}

Severity: BLOCKING | IMPORTANT | MINOR

Type: JUDGMENT-BASED is a separate flag, not a severity level. Present findings in order: BLOCKING first, then IMPORTANT, then MINOR.

4. Workflow

  1. Read postman.toml — extract edges, build adjacency map
  2. Read each nodes/{node}.toml (source of truth; runtime session templates are NOT compared)
  3. For each node: run Pre-check, then Checks 1–6 in order
  4. Produce findings report sorted by severity
  5. Propose concrete patch text for every finding
  6. Present to user for feedback; iterate until approved

NOTE: Do NOT auto-apply patches. Propose only; the user applies manually or delegates to worker.

5. Baseline Examples

The following issues were identified in a real audit session and serve as illustrative examples.

5.1. Example 1 — Routing clarity (IMPORTANT)

[IMPORTANT] Node: critic
Field: nodes/critic.toml:[critic].template
Check: Routing clarity
Result: FAIL
Issue: Template says "send a message via postman" without specifying a recipient.
Fix:
  "Send findings to orchestrator. If orchestrator is absent from
  talks_to_line, send to guardian who will relay."

5.2. Example 2 — Completion protocol (IMPORTANT)

[IMPORTANT] Node: boss
Field: nodes/boss.toml:[boss].template
Check: Completion protocol
Result: FAIL
Issue: No machine-readable approval signal defined. Recipients cannot parse
  the response programmatically.
Fix:
  "Reply with 'APPROVED: <summary>' when approving,
  or 'REJECTED: <reason>' when rejecting."

5.3. Example 3 — Cross-edge consistency (IMPORTANT, JUDGMENT-BASED)

[IMPORTANT] Node: guardian
Field: nodes/guardian.toml:[guardian].template
Check: Cross-edge consistency
Type: JUDGMENT-BASED
Result: FAIL
Issue: Guardian acts as a routing relay between critic and orchestrator,
  but this is not documented. Forwarding-to-boss path is also absent.
Fix:
  "After receiving critic findings: if approved, forward to boss.
  If rejected, return to critic with specific revision request."

5.4. Example 4 — on_join completeness (MINOR)

[MINOR] Node: worker
Field: nodes/worker.toml:[worker].on_join
Check: on_join completeness
Result: FAIL
Issue: on_join is empty; node receives no startup context.
Fix:
  on_join = "You are worker. Send PONG to postman on startup, then await task assignment from orchestrator."

5.5. Example 5 — PONG awareness (BLOCKING)

[BLOCKING] Node: orchestrator
Field: nodes/orchestrator.toml:[orchestrator].template
Check: PONG awareness
Result: FAIL
Issue: Template references postman only as a routing mechanism, never as explicit
  recipient. Node will not be marked PONG-active and will be invisible in
  other nodes' talks_to_line.
Fix:
  Add to startup section: "On session start, send a PONG message addressed TO postman
  (recipient = postman, not via postman). This registers you as active."

6. Constraints

  • Propose patches only; do NOT auto-apply
  • When an issue is daemon-level (wrong edges, missing file, disabled session), note it as a config finding — template patches cannot fix it
  • Manual integration test for talks_to_line visibility: if both nodes have not yet sent PONG in the current session, mark the result INCONCLUSIVE (environment), not a skill failure

平台分布

Codex

35.22%
按下载量换算22

Claude

27.26%
按下载量换算17

Cursor

17.99%
按下载量换算11

Gemini CLI

9.84%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills