The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest "five recent posts" provided by the user in Step 1. This external content is then processed to create a writing sketch and final draft. The instructions lack boundary markers (e.g., XML tags or delimiters) or specific directives for the agent to ignore potentially malicious instructions embedded within the user-provided examples, creating a surface for indirect prompt injection.
Ingestion points: User-provided social media posts in Step 1.
Boundary markers: None mentioned.
Capability inventory: publish_post, schedule_post, get_voice_profile, score_voice_match.
Sanitization: None mentioned.
[DATA_EXFILTRATION]: The skill describes an optional VoiceMoat MCP connector that, when active, reads from or sends "post text and account data" to voicemoat.com. This network communication is a declared behavior for the skill's core functionality and is managed via OAuth/user sign-in rather than hardcoded credentials.
[COMMAND_EXECUTION]: The skill utilizes high-privilege tools such as publish_post and schedule_post to interact with the user's real Twitter/X or LinkedIn accounts. To mitigate the risk of unauthorized actions, the skill enforces a safety protocol requiring the agent to display a preview and obtain a second confirmation call containing a one-time code before the action is finalized.

write-in-my-voice