The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest large volumes of untrusted data from Slack, Google Drive, Email, and Calendar to generate summaries (3P updates, newsletters, FAQs). This creates a surface for attackers to embed instructions in these sources that the agent might follow.
Ingestion points: examples/3p-updates.md, examples/company-newsletter.md, and examples/faq-answers.md instruct the agent to use tools to search and read posts, documents, emails, and calendar events.
Boundary markers: The instructions do not define strict boundary markers or include system-level warnings to ignore instructions found within the processed data.
Capability inventory: The skill itself does not specify dangerous tools like file-write or subprocess execution in its frontmatter, but it is designed for environments where Slack/Drive/Email tools are active.
Sanitization: There is no evidence of sanitization or filtering for embedded prompt injection triggers in the source data.

internal-comms