The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from Instagram, including user bios, post captions, and comments, which could contain malicious instructions.
Ingestion points: Untrusted data enters the agent context through tools such as instagram_v3_user_profile, instagram_v3_post_comments, and instagram_v3_hashtag_posts (referenced in SKILL.md and references/sandbase-api-map.md).
Boundary markers: The skill does not provide specific instructions or delimiters to the agent to ignore or isolate potentially malicious instructions embedded within the fetched social media content.
Capability inventory: The skill facilitates data retrieval and analysis; however, the provided files do not show high-risk capabilities like local file writing, shell command execution, or arbitrary network requests outside the SandBase gateway.
Sanitization: There are no instructions for sanitizing or escaping the content retrieved from Instagram before it is processed or output by the agent.

instagram-research