The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection as it processes untrusted data from various company sources without clear security boundaries or sanitization.
Ingestion points: The files examples/3p-updates.md, examples/company-newsletter.md, and examples/faq-answers.md instruct the agent to retrieve and summarize data from Slack, Google Drive, Email, and Calendar.
Boundary markers: The instructions do not define specific delimiters or directives to ensure the agent ignores potential commands or overrides embedded within the retrieved messages or documents.
Capability inventory: The agent is authorized to read sensitive communication history and file contents and then format them into reports or newsletters for wider distribution.
Sanitization: The skill lacks any requirement for content validation or filtering of the source data before it is incorporated into the generated communications.

internal-comms