The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input and external data which could be manipulated to inject malicious instructions into the agent's context, potentially influencing social media output.
Ingestion points: Raw reel ideas provided by the user, voice profiles read from ~/.uplika/<workspace-slug>/voice.md, and external "swipe files" referenced from the ig-viral skill (found in SKILL.md).
Boundary markers: The skill does not implement delimiters or explicit "ignore embedded instructions" warnings for the ingested content.
Capability inventory: The skill possesses the ability to publish content directly to Instagram using the publish tool, along with media management tools like media_presign, media_complete, and search_instagram_audio (found in SKILL.md).
Sanitization: There is no evidence of text sanitization, validation, or escaping for the external content before it is processed by the agent or published.

ig-reel