The Agent Skills Directory

[Indirect Prompt Injection Surface]: The skill processes potentially untrusted content from emails, internal chats, and meeting transcripts. To mitigate risks, the skill includes rigorous safeguards: it explicitly instructs the agent to treat this data as information rather than instructions, forbids the rendering of links found within untrusted text, and mandates user review for any content-originated actions. Evidence: Untrusted content handling instructions in General Rules and Steps 3, 4, and 5.
[Controlled Capability Access]: While the skill can perform actions like sending emails or DMs, it adheres to the underlying connector permissions and requires explicit user approval before executing actions suggested by or based on external content. Evidence: General Rules and the Owner check in Step 6.
[Safe Link Handling]: The skill uses a defensive approach to URL handling, using only verified web links from the tool connectors and specifically appending parameters like ?web=1 to ensure files open safely in a browser rather than being downloaded directly. Evidence: Link formatting rules in General Rules.

account-context