The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements a self-improvement feature that maintains state by reading from and writing to a local file (references/learnings.md). This creates a potential surface for indirect prompt injection.
Ingestion points: The agent is instructed to read references/learnings.md at the start of every session if it exists.
Boundary markers: Absent; there are no specific instructions or delimiters used to prevent the agent from treating the contents of the learnings file as authoritative instructions.
Capability inventory: The skill allows appending content to the references/learnings.md file based on discoveries during usage. It does not possess dangerous capabilities like arbitrary shell execution, network exfiltration, or access to sensitive system directories.
Sanitization: There is no evidence of validation or sanitization of the data before it is written to or read from the learnings file.

sales-webinar