The Agent Skills Directory

[SAFE]: The skill is a legitimate tool for book authoring, using structured templates and a multi-agent workflow to ensure content quality and process transparency. It does not perform unauthorized network activity or access sensitive system credentials.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content (debates, notes) as research material, creating a surface for indirect prompt injection. Ingestion occurs through the research workflow where external files are read into the agent context from user-provided paths. Boundary markers include a strictly defined orientation brief and negative constraints like 'não invente'. The capability inventory includes workspace file operations and the use of the asciidoctor utility. Sanitization is managed through a sequence of independent review steps (editor, fact-checker) to verify output against source material.
[COMMAND_EXECUTION]: The skill instructs the agent to run the asciidoctor command for book compilation. This is a standard and expected utility for the skill's purpose and operates on project files created within the skill's directory structure.

teo-livro