The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8). It is designed to ingest and summarize large volumes of untrusted data from Slack channels, emails, and shared documents.
Ingestion points: The skill actively searches Slack messages (specifically looking for high engagement/reactions), Google Drive documents, and company-wide emails to generate 3P updates, newsletters, and FAQs (as defined in examples/3p-updates.md, examples/company-newsletter.md, and examples/faq-answers.md).
Boundary markers: The instructions do not define clear delimiters (e.g., XML tags) or specific 'ignore embedded instructions' warnings when processing this external context.
Capability inventory: The skill has read-access to sensitive communication platforms and produces structured output that could influence company-wide information flow. While it does not appear to execute code or perform network operations, the generated content is intended for wide internal distribution.
Sanitization: There is no explicit instruction to the agent to sanitize or validate the content found in these third-party sources for malicious instructions aimed at overriding the agent's behavior during the drafting process.

internal-comms