The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill relies on ingesting and summarizing data from multi-user environments, which introduces a potential vector for malicious actors to influence the agent's output through specially crafted content.
Ingestion points: The agent is instructed to gather information from Slack channels, Google Drive documents, Email communications, and Calendar events to draft newsletters, 3P updates, and FAQ responses (referenced in examples/3p-updates.md, examples/company-newsletter.md, and examples/faq-answers.md).
Boundary markers: There are no specific instructions or delimiters provided to the agent to treat external content as untrusted or to ignore instructions that might be embedded within those sources.
Capability inventory: The skill directs the agent to search for and read potentially sensitive communications (e.g., leadership updates, critical project docs) and output summaries or newsletters directly to the user.
Sanitization: The skill lacks any guidelines for validating or sanitizing the data retrieved from external tools before it is processed and presented.

internal-comms