The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied content (Facebook posts and voice samples) to perform complex rewriting and auditing tasks. This creates a surface for indirect prompt injection where malicious instructions embedded in the input could influence the agent's behavior during the humanization process. The skill lacks explicit boundaries and sanitization to isolate untrusted data from the agent's instructional logic.
Ingestion points: User-supplied text to be humanized in SKILL.md and user-provided voice samples in sub-skills/voice-profile.md.
Boundary markers: Absent; the skill does not define specific delimiters or directives for the agent to ignore instructions embedded within the processed text.
Capability inventory: File system write access (writing or updating voice-profile.md) and remote tool invocation (image generation through the lib.illustrate function in sub-skills/illustration.md).
Sanitization: No evidence of sanitization, filtering, or escaping of the user-provided text is described before it is integrated into the agent's decision-making and rewriting workflows.

fb-humanizer