The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and transforms it into persistent operational rules, which is a common attack surface for indirect prompt injection.
Ingestion points: The skill explicitly ingests user messages and orientations as defined in SKILL.md ('Receba a mensagem do usuário').
Boundary markers: The skill instructs the agent to 'Separate the literal orientation from the interpreted operational rule,' which provides a structural boundary between raw input and executed logic, though it remains a conceptual guardrail rather than a technical sandbox.
Capability inventory: The skill allows the agent to write and update project configuration files including .inboundfy/context/aprendizado.md, dicionario.md, and proibicoes.md. It also invokes local processing tools such as inboundfy-anti-slop and inboundfy-voz.
Sanitization: The workflow includes classification (sugestão, correção, alinhamento, dica) and an interpretation step intended to limit the scope of the rule to what the message explicitly permits, reducing the risk of over-privileged rule creation.

inboundfy-aprendizado