The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill is designed to ingest vast amounts of untrusted data from Slack, Email, and Google Drive and use it to generate company-wide communications.
Ingestion points: Identified in examples/3p-updates.md, examples/company-newsletter.md, and examples/faq-answers.md (Slack messages, Email threads, Google Drive docs, Calendar events).
Boundary markers: Entirely absent. There are no instructions to the agent to distinguish between its operational instructions and the content within the documents/messages it reads.
Capability inventory: The skill generates outputs for high-stakes audiences (executives, company-wide newsletters, leadership updates). A malicious actor could embed instructions in a Slack post or email to influence the content of the official newsletter or FAQ.
Sanitization: No validation or sanitization of ingested content is mentioned.
Data Exposure (HIGH): The skill explicitly instructs the agent to seek out sensitive internal data points.
Evidence: Instructions in examples/company-newsletter.md and examples/3p-updates.md specifically direct the agent to find "docs written from critical team members," "emails from executives," "company-wide vision docs," and "product reviews."
Risk: While the intended use is summarization, this creates a pattern where the agent is trained to aggregate high-value internal secrets and private communications, which could be exploited or accidentally leaked if the agent is also tasked with external-facing roles.

internal-comms