The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied justification text and media (images or videos) to perform its analysis. This creates a surface for indirect prompt injection where a user might attempt to embed instructions to redirect the agent's behavior. However, the rigid output format and specific analytical requirements significantly limit the potential impact of such attempts.
Ingestion points: User justification text and media uploads (SKILL.md).
Boundary markers: None explicitly defined to separate user content from instructions.
Capability inventory: Uses google_search to find educational resources.
Sanitization: No explicit sanitization of user-provided text or media metadata is performed.
[SAFE]: The skill uses the google_search tool for the benign purpose of finding educational links related to semiotics and visual communication, which is consistent with its stated persona.
[SAFE]: The skill-card references the author's personal GitHub repository (github.com/marioluciofjr/skills) for licensing and context, which is a standard practice and involves a well-known service.

analise-semiotica