The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data (company blogs, social media posts) to extract rules and style guides (voice cards). These cards are saved as files and used to influence subsequent content generation tasks. If the ingested content contains hidden instructions, the agent might adopt them as stylistic or brand rules.\n
Ingestion points: references/blog-voice.md (Step 1) reads article bodies from discovered blog URLs; SKILL.md (Build Step 1) ingests social posts from the marketing-data-access skill or user-pasted text.\n
Boundary markers: The skill lacks explicit boundary markers or 'ignore' instructions for the data being analyzed, although it mentions screening for commercial claims.\n
Capability inventory: The skill uses gooseworks:file_write to save extracted rules to brand/voices/<handle>.md, which is then read by the agent for all future writing jobs associated with that handle.\n
Sanitization: The skill attempts to screen quotations for claims but does not explicitly sanitize against prompt injection payloads in the analyzed text.\n- [EXTERNAL_DOWNLOADS]: The skill fetches article content from external websites and social platforms to analyze writing styles.\n
Evidence: references/blog-voice.md instructs the agent to discover and read article bodies from company blogs using the host's public fetch or browser tools.

voice-card