The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains a self-improvement loop that enables indirect prompt injection by persisting data from interactions into its own reference files.
Ingestion points: The agent is instructed in SKILL.md to read references/learnings.md at the start of every session to incorporate accumulated knowledge.
Boundary markers: There are no delimiters or specific instructions to isolate the data in references/learnings.md from the system instructions, increasing the risk of the agent interpreting stored data as commands.
Capability inventory: The skill utilizes file-writing capabilities to append user-influenced insights to the filesystem and references various external platforms for conversion tracking and marketing.
Sanitization: The skill does not implement any validation, filtering, or escaping for the content saved to or loaded from the persistence file.

sales-funnel