Habilidades testing-skills-with-subagents
📦

testing-skills-with-subagents

Revisão do conteúdo r1 Risco alto ⚙️ Comandos externos📁 Acesso ao sistema de arquivos

Teste de Skills com Subagentes

Skills de processo muitas vezes falham quando agentes enfrentam pressão de tempo, pressão de autoridade ou custo irrecuperável. Esta skill fornece um fluxo de trabalho repetível de teste com subagentes para expor falhas e melhorar instruções.

Suporta: Claude Codex Code(CC)
⚠️ 38 Ruim

Instalar com meu Agente

Copie esta solicitação para o seu Agente. Ela inclui a página canônica do Skill e o manifesto.

Solicitação do agente
Review the Skillstore skill "testing-skills-with-subagents" from https://skillstore.io/skills/dyai2025-testing-skills-with-subagents.md and its manifest at https://skillstore.io/api/skills/dyai2025-testing-skills-with-subagents/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.

Seu Agente ainda deve apresentar o plano e solicitar qualquer confirmação exigida pela política de segurança.

Recursos legíveis por agentes

Use estes links quando um agente de IA, crawler ou script precisar de contexto limpo em vez de ler a página completa.

Detalhe em Markdown GET /skills/dyai2025-testing-skills-with-subagents.md manifest assinado GET /api/skills/dyai2025-testing-skills-with-subagents/manifest lockfile assinado GET /api/skills/dyai2025-testing-skills-with-subagents/lockfile

Testar

Usando "testing-skills-with-subagents". Uma skill de TDD falha quando agentes escolhem testes depois da implementação.

Resultado esperado:

  • Linha de base: o agente escolheu o caminho mais rápido sob pressão de tempo.
  • Contraponto: adicionar uma regra de que a ordem faz parte do requisito.
  • Reteste: o agente deve reiniciar com testes primeiro.

Usando "testing-skills-with-subagents". Um lembrete em CLAUDE.md não faz os agentes verificarem skills instaladas.

Resultado esperado:

  • A comparação de variantes mostra que sugestões suaves falham sob urgência.
  • A redação diretiva melhora a descoberta, mas parece rígida.
  • A redação de processo equilibra clareza e reutilização.

Auditoria de segurança

Risco alto
v10 • 09/07/2026 Abrir o relatório versionado

Most command alerts are markdown examples, not executable code. Repeated ~/.claude/skills access is a real workspace-boundary risk. The files also include prompt-pressure templates that can steer agent behavior.

2
Arquivos analisados
578
Linhas analisadas
35
Itens de revisão
0
Falsos positivos ignorados

Preocupações de segurança confirmadas (2)

Alto
examples/CLAUDE_MD_TESTING.md:100-112
Prompt Injection Attempt Detected
The example instruction block says "THIS IS EXTREMELY IMPORTANT. BEFORE ANY TASK, CHECK FOR SKILLS!" and frames noncompliance as failure. If copied into active agent context, it can override normal task judgment and force tool behavior.
The language is explicitly designed to steer agent behavior with urgency and failure framing. Confidence is not higher because it appears inside a documentation variant used for testing.
Médio
SKILL.md:157-164examples/CLAUDE_MD_TESTING.md:8-19
Deceptive Agent Test Framing
The skill tells authors to present scenarios as real work and to make the agent believe it is not a quiz. This is useful for testing, but unsafe if used outside controlled subagents.
The file directly instructs users to make agents believe a test is real work. The testing context reduces severity, but the social-engineering pattern is clear.
Itens de revisão de capacidade (35)

Estas são capacidades locais reais que podem ser esperadas para esta habilidade, portanto exigem revisão, mas não são contabilizadas como comportamento malicioso confirmado.

Alto
examples/CLAUDE_MD_TESTING.md:16
Hidden file in home directory
B) Check ~/.claude/skills/debugging/ first (2 min check + 5 min fix = 7 min)
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:34
Hidden file in home directory
A) Check ~/.claude/skills/testing/ for async testing skill
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:46
Hidden file in home directory
A) Check ~/.claude/skills/ for validation patterns (1-2 min)
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:60
Hidden file in home directory
A) Check ~/.claude/skills/coding/ for refactoring guidance
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:73
Hidden file in home directory
You have access to skills at `~/.claude/skills/`. Consider
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:81
Hidden file in home directory
Before working on any task, check `~/.claude/skills/` for
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:84
Hidden file in home directory
Browse: `ls ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:85
Hidden file in home directory
Search: `grep -r "keyword" ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:92
Hidden file in home directory
is at `~/.claude/skills/`.
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:94
Hidden file in home directory
Browse categories: `ls ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:95
Hidden file in home directory
Search: `grep -r "keyword" ~/.claude/skills/ --include="SKILL.md"`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:107
Hidden file in home directory
1. Starting work? Check: `ls ~/.claude/skills/[category]/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:122
Hidden file in home directory
- Browse: `ls ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Alto
examples/CLAUDE_MD_TESTING.md:123
Hidden file in home directory
- Search: `grep -r "symptom" ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:84-85
Ruby/shell backtick execution
Browse: `ls ~/.claude/skills/`
The documentation examples encourage running hardcoded ls or grep commands against ~/.claude/skills. The commands are read-only, but they still ask the agent to execute shell commands outside the workspace.
Médio
examples/CLAUDE_MD_TESTING.md:85-86
Ruby/shell backtick execution
Search: `grep -r "keyword" ~/.claude/skills/`
The documentation examples encourage running hardcoded ls or grep commands against ~/.claude/skills. The commands are read-only, but they still ask the agent to execute shell commands outside the workspace.
Médio
examples/CLAUDE_MD_TESTING.md:94-95
Ruby/shell backtick execution
Browse categories: `ls ~/.claude/skills/`
The documentation examples encourage running hardcoded ls or grep commands against ~/.claude/skills. The commands are read-only, but they still ask the agent to execute shell commands outside the workspace.
Médio
examples/CLAUDE_MD_TESTING.md:95-97
Ruby/shell backtick execution
Search: `grep -r "keyword" ~/.claude/skills/ --include="SKILL.md"`
The documentation examples encourage running hardcoded ls or grep commands against ~/.claude/skills. The commands are read-only, but they still ask the agent to execute shell commands outside the workspace.
Médio
examples/CLAUDE_MD_TESTING.md:107-113
Ruby/shell backtick execution
1. Starting work? Check: `ls ~/.claude/skills/[category]/`
The documentation examples encourage running hardcoded ls or grep commands against ~/.claude/skills. The commands are read-only, but they still ask the agent to execute shell commands outside the workspace.
Médio
examples/CLAUDE_MD_TESTING.md:122-123
Ruby/shell backtick execution
- Browse: `ls ~/.claude/skills/`
The documentation examples encourage running hardcoded ls or grep commands against ~/.claude/skills. The commands are read-only, but they still ask the agent to execute shell commands outside the workspace.
Médio
examples/CLAUDE_MD_TESTING.md:123-132
Ruby/shell backtick execution
- Search: `grep -r "symptom" ~/.claude/skills/`
The documentation examples encourage running hardcoded ls or grep commands against ~/.claude/skills. The commands are read-only, but they still ask the agent to execute shell commands outside the workspace.
Médio
examples/CLAUDE_MD_TESTING.md:16
Hidden file access
B) Check ~/.claude/skills/debugging/ first (2 min check + 5 min fix = 7 min)
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:34
Hidden file access
A) Check ~/.claude/skills/testing/ for async testing skill
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:46
Hidden file access
A) Check ~/.claude/skills/ for validation patterns (1-2 min)
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:60
Hidden file access
A) Check ~/.claude/skills/coding/ for refactoring guidance
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:73
Hidden file access
You have access to skills at `~/.claude/skills/`. Consider
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:81
Hidden file access
Before working on any task, check `~/.claude/skills/` for
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:84
Hidden file access
Browse: `ls ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:85
Hidden file access
Search: `grep -r "keyword" ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:92
Hidden file access
is at `~/.claude/skills/`.
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:94
Hidden file access
Browse categories: `ls ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:95
Hidden file access
Search: `grep -r "keyword" ~/.claude/skills/ --include="SKILL.md"`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:107
Hidden file access
1. Starting work? Check: `ls ~/.claude/skills/[category]/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:122
Hidden file access
- Browse: `ls ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.
Médio
examples/CLAUDE_MD_TESTING.md:123
Hidden file access
- Search: `grep -r "symptom" ~/.claude/skills/`
The scenario or template directs agents to inspect ~/.claude/skills under the user home directory. This crosses the workspace boundary and can expose private installed skill content.

Fatores de risco

⚙️ Comandos externos (41)
examples/CLAUDE_MD_TESTING.md:47-50 examples/CLAUDE_MD_TESTING.md:73-75 examples/CLAUDE_MD_TESTING.md:81-84 examples/CLAUDE_MD_TESTING.md:84-85 examples/CLAUDE_MD_TESTING.md:85-86 examples/CLAUDE_MD_TESTING.md:92-94 examples/CLAUDE_MD_TESTING.md:94-95 examples/CLAUDE_MD_TESTING.md:95-97 examples/CLAUDE_MD_TESTING.md:97-107 examples/CLAUDE_MD_TESTING.md:107-113 examples/CLAUDE_MD_TESTING.md:122-123 examples/CLAUDE_MD_TESTING.md:123-132 examples/CLAUDE_MD_TESTING.md:132-133 SKILL.md:62-75 SKILL.md:75-102 SKILL.md:102-104 SKILL.md:104-108 SKILL.md:108-111 SKILL.md:111-115 SKILL.md:115-126 SKILL.md:126-151 SKILL.md:151-157 SKILL.md:157-162 SKILL.md:162-188 SKILL.md:188-190 SKILL.md:190-194 SKILL.md:194-202 SKILL.md:202-207 SKILL.md:207-211 SKILL.md:211-215 SKILL.md:215-220 SKILL.md:220-224 SKILL.md:224-226 SKILL.md:226-247 SKILL.md:247-252 SKILL.md:252-288 SKILL.md:288-292 SKILL.md:292-295 SKILL.md:295-299 SKILL.md:299-302 SKILL.md:302-307
📁 Acesso ao sistema de arquivos (29)
examples/CLAUDE_MD_TESTING.md:16 examples/CLAUDE_MD_TESTING.md:34 examples/CLAUDE_MD_TESTING.md:46 examples/CLAUDE_MD_TESTING.md:60 examples/CLAUDE_MD_TESTING.md:73 examples/CLAUDE_MD_TESTING.md:81 examples/CLAUDE_MD_TESTING.md:84 examples/CLAUDE_MD_TESTING.md:85 examples/CLAUDE_MD_TESTING.md:92 examples/CLAUDE_MD_TESTING.md:94 examples/CLAUDE_MD_TESTING.md:95 examples/CLAUDE_MD_TESTING.md:107 examples/CLAUDE_MD_TESTING.md:122 examples/CLAUDE_MD_TESTING.md:123 examples/CLAUDE_MD_TESTING.md:16 examples/CLAUDE_MD_TESTING.md:34 examples/CLAUDE_MD_TESTING.md:46 examples/CLAUDE_MD_TESTING.md:60 examples/CLAUDE_MD_TESTING.md:73 examples/CLAUDE_MD_TESTING.md:81 examples/CLAUDE_MD_TESTING.md:84 examples/CLAUDE_MD_TESTING.md:85 examples/CLAUDE_MD_TESTING.md:92 examples/CLAUDE_MD_TESTING.md:94 examples/CLAUDE_MD_TESTING.md:95 examples/CLAUDE_MD_TESTING.md:107 examples/CLAUDE_MD_TESTING.md:122 examples/CLAUDE_MD_TESTING.md:123 SKILL.md:151
Auditado por: codex Ver histórico de auditoria →
Compartilhar e citar este relatório

Compartilhe o relatório de avaliação versionado, o badge neutro, o cartão incorporável e as citações. A Skillstore apresenta evidências sem decidir se este Skill é seguro.

Abrir o relatório versionado
Avaliação de Segurança

Copiar link do relatório

https://skillstore.io/skills/dyai2025-testing-skills-with-subagents/audits/10?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_report

Badge Markdown

[![Skillstore security assessment](https://skillstore.io/badges/skills/dyai2025-testing-skills-with-subagents/security.svg)](https://skillstore.io/skills/dyai2025-testing-skills-with-subagents?utm_source=security_passport_badge)

Badge HTML

<a href="https://skillstore.io/skills/dyai2025-testing-skills-with-subagents?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/dyai2025-testing-skills-with-subagents/security.svg" alt="Skillstore security assessment" loading="lazy"></a>

Cartão incorporável

<iframe src="https://skillstore.io/embed/skills/dyai2025-testing-skills-with-subagents.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>
Citações acadêmicas (APA · BibTeX · CFF)

Citação APA

DYAI2025. (2026). testing-skills-with-subagents security audit report (audit version 10) [Author version unspecified]. Skillstore. https://skillstore.io/skills/dyai2025-testing-skills-with-subagents/audits/10

Citação BibTeX

@techreport{dyai2025-dyai2025-testing-skills-with-subagents-2026, author = {DYAI2025}, title = {testing-skills-with-subagents security audit report (audit version 10)}, institution = {Skillstore}, year = {2026}, number = {10}, url = {https://skillstore.io/skills/dyai2025-testing-skills-with-subagents/audits/10}, 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: "testing-skills-with-subagents security audit report (audit version 10)" version: "unspecified" type: report authors: - name: "DYAI2025" date-released: "2026-07-09" url: "https://skillstore.io/skills/dyai2025-testing-skills-with-subagents/audits/10" identifiers: - type: other value: "skillstore:dyai2025-testing-skills-with-subagents:audit:10" description: "Skillstore immutable audit report identifier"

Pontuação do Skillstore

Por que esta pontuação Confiança das evidências: Médio
55
Arquitetura
85
Manutenibilidade
85
Conteúdo
72
Comunidade
91
Conformidade com a especificação

O que você pode criar

Validar uma Nova Skill

Teste uma skill que impõe disciplina antes de publicá-la.

Comparar Variantes de Instruções

Meça qual variante de documentação melhora a conformidade do agente.

Revisar Skills da Comunidade

Avalie se uma skill lida com pressão sem comportamento inseguro.

Teste estes prompts

Executar um Teste de Linha de Base 
Crie um cenário de pressão para esta skill. Execute-o sem a skill e depois liste a escolha e a racionalização do agente.
Comparar Variantes de Documentação 
Teste duas variantes de instruções com o mesmo cenário. Informe qual variante melhorou a descoberta da skill e por quê.
Fechar um Ciclo de Racionalização 
Use a racionalização capturada para revisar a skill. Adicione um contraponto explícito e um sinal de alerta.
Projetar uma Campanha de Teste Completa 
Crie uma campanha RED, GREEN, REFACTOR para esta skill com três cenários de pressão, critérios de sucesso e etapas de reteste.

Boas práticas

  • Execute uma linha de base antes de editar a skill.
  • Use o mesmo cenário ao comparar variantes.
  • Registre as racionalizações exatas antes de escrever correções.

Evitar

  • Pular RED porque a regra pretendida parece óbvia.
  • Testar apenas cenários calmos sem pressão real.
  • Alterar muitas instruções antes de isolar a falha.

Perguntas frequentes

Que problema esta skill resolve?
Ela ajuda autores a provar que skills de processo funcionam quando agentes têm motivos para ignorá-las.
Ela executa testes automaticamente?
Não. Ela descreve um fluxo de trabalho manual de teste para subagentes e documentação.
Quais skills se beneficiam mais?
Skills que impõem disciplina, como TDD, revisão ou etapas obrigatórias de fluxo de trabalho.
Ela pode ser usada com Codex?
Sim. O fluxo de trabalho pode orientar sessões do Codex, Claude e Claude Code.
Por que usar cenários de pressão?
Eles revelam racionalizações que verificações simples de conhecimento não expõem.
Qual é a principal preocupação de segurança?
Os exemplos usam prompts persuasivos, portanto os testes devem ser executados apenas em contextos controlados de subagentes.

Detalhes do desenvolvedor

Autor

DYAI2025

Licença

MIT

Revisão da Skillstore

r1

Aviso de versão

O autor não declarou uma versão.

Repositório

https://github.com/DYAI2025/Stoppclock-page/tree/main/stoppclock_speckit/.claude/commands/skills/testing-skills-with-subagents

Referência

0519034dad657fb1f7706e0550e962beeda73fdf

Atualidade da manutenção

18/07/2026

Uso

6 downloads · 341 visualizações

Estrutura de arquivos

📁 examples/

📄 CLAUDE_MD_TESTING.md

📄 SKILL.md

📄 SKILL.zip