The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that involves processing external data sources, specifically YouTube links and .mp4 video files, creating a potential surface for indirect prompt injection attacks.
Ingestion points: External media content provided by users for phonetic dissection as specified in SKILL.md.
Boundary markers: Partially present; the instructions in SKILL.md direct the agent to restrict its focus to phonetic analysis and ignore other metadata (lighting, posture), but there is no explicit instruction to disregard or identify adversarial commands spoken within the audio.
Capability inventory: The skill frontmatter does not request access to high-risk tools (e.g., shell execution, file system writes), limiting the potential impact of an injection.
Sanitization: Absent; no methods are defined to sanitize or validate the content extracted from the media before processing.
[EXTERNAL_DOWNLOADS]: The file references/ref.md contains several references to external educational and reference websites (e.g., Wikipedia, Toda Matéria, Fonologia.org, Português.com.br). These links are used for academic reference purposes related to phonetics and linguistics and do not involve the download or execution of scripts or binaries.

contexto-sotaque