ai-agent
A Chinese-language skill from the CkSKILLS security library, written for authorised assessments such as bug bounty and red-team work, that guides an agent through testing AI and LLM systems: chatbots, retrieval pipelines, agent platforms such as Langflow, ComfyUI, Dify and Gradio, and online code sandboxes.
O que a skill ensina
A Chinese-language skill from the CkSKILLS security library, written for authorised assessments such as bug bounty and red-team work, that guides an agent through testing AI and LLM systems: chatbots, retrieval pipelines, agent platforms such as Langflow, ComfyUI, Dify and Gradio, and online code sandboxes. It sets a triage order, mapping trust boundaries, probing platform APIs by swapping tenant or sandbox IDs, then working through credentials, privilege escalation and sandbox escape, and covers prompt injection, knowledge-base poisoning and Python or Node.js sandbox escapes. About 50 knowledge files sit beside it, to be searched by heading. The SKILL.md itself does not restate the authorisation requirement.
O que faz
- Triage an AI platform's trust boundaries
- Test prompt injection and retrieval poisoning
- Probe agent platform APIs for cross-tenant access
- Check sandboxes for credentials and escapes
Sobre
- Publicador
- zhaji2333
Instalação
O projeto não documenta passos de instalação que tenhamos conseguido confirmar. Veja as instruções do próprio projeto
O que tem dentro
- knowledge/AI平台漏洞实战.md
- knowledge/agent-execution-channel-hijack-01-一、三层-只有一层会拒绝.md
- knowledge/agent-execution-channel-hijack-02-二、路径-B-五步.md
- knowledge/agent-execution-channel-hijack-03-三、HTTP-跨租户-必须三态-+-对照.md
- knowledge/agent-execution-channel-hijack-04-四、网关-控制面-HTTP-L1-L2.md
- knowledge/agent-execution-channel-hijack-05-五、「拒绝了但执行了」.md
- knowledge/agent-execution-channel-hijack-06-六、客户端连云端网关.md
- knowledge/agent-execution-channel-hijack-07-七、判定表.md
- knowledge/agent-execution-channel-hijack-08-八、联动.md
- knowledge/agent-prompt-injection-bypass-01-一、先分清-遥控器-vs-注入.md
- knowledge/agent-prompt-injection-bypass-02-二、铁律.md
- knowledge/agent-prompt-injection-bypass-03-三、第一枪-未拒绝时.md
- knowledge/agent-prompt-injection-bypass-04-四、护栏六层-已拒绝时.md
- knowledge/agent-prompt-injection-bypass-05-五、L2-改叙事-不改攻击目标.md
- knowledge/agent-prompt-injection-bypass-06-六、L4-换实现-不换话术.md
- knowledge/agent-prompt-injection-bypass-07-七、L1-编码-+-间接注入.md
- knowledge/agent-prompt-injection-bypass-08-八、失败升级.md
- knowledge/agent-prompt-injection-bypass-09-九、联动.md
- knowledge/agent-sandbox-takeover-01-一、要打穿的面-按性价比-不是按酷.md
- knowledge/agent-sandbox-takeover-02-二、进站决策树.md
- knowledge/agent-sandbox-takeover-03-三、默认第一轮-让-Agent-自己测绘.md
- knowledge/agent-sandbox-takeover-04-三-1-拉壳-通道-不是洞.md
- knowledge/agent-sandbox-takeover-05-四、六问测绘-人工接管时的固定动作.md
- knowledge/agent-sandbox-takeover-06-五、回环被抬到-0-0-0-0.md
- knowledge/agent-sandbox-takeover-07-六、七类反复出现的失误-对空列.md
Servidores MCP e skills podem executar código no seu computador e agir sobre os seus dados. Revise o código-fonte antes de instalar.
Detalhes
| Fato | Valor | Fonte |
|---|---|---|
| Tipo | Skill | |
| Ferramenta | Não informado na documentação | |
| Publicador | zhaji2333 | fonte |
| Status oficial | Não informado na documentação | |
| Exige chave de API ou login | Não informado na documentação | |
| Inclui scripts | Não | fonte |
| Plataformas | Não informado na documentação | |
| Clientes de IA | Não informado na documentação | |
| Licença | MIT | fonte |
| Preço | Não informado na documentação | |
| Atividade | Último commit em 20 de nov. de 2026164 estrelas no GitHubDados de download indisponíveis (não está no npm nem no PyPI) | fonte |
| Links | Repositório |
Fontes
- Arquivo da skill (SKILL.md)github.com/zhaji2333/ckskills16 de nov. de 2026
- Repositório no GitHubgithub.com/zhaji2333/ckskills · Comprova: Licença, Publicador, Atividade11 de nov. de 2026
- Arquivo da skill (SKILL.md)github.com/zhaji2333/CkSKILLS15 de nov. de 2026
- Pasta no repositório (ai-agent)github.com/zhaji2333/CkSKILLS · Comprova: Inclui scripts15 de nov. de 2026