The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest data from untrusted or external sources like Slack posts, Google Drive documents, and Emails. If these sources contain malicious instructions (e.g., hidden markdown or comments), the agent might follow them during the summarization or drafting process.
Ingestion points: Slack channels, Google Drive documents, and corporate Emails (specified in examples/3p-updates.md, examples/company-newsletter.md, and examples/faq-answers.md).
Boundary markers: None identified. The instructions do not tell the agent to ignore instructions embedded within the retrieved data.
Capability inventory: The agent has the capability to read extensive organizational data and re-share it in summarized formats (newsletters, FAQs).
Sanitization: None identified. The skill lacks filters or validation steps for the content fetched from external tools.
[DATA_EXFILTRATION]: While not directly exfiltrating data to an attacker's server, the skill encourages the agent to pull sensitive information from private company silos (Email, Drive, Slack) and redistribute it into broad formats like company-wide newsletters or FAQs. This could lead to accidental exposure of restricted information to the entire company if the agent summarizes sensitive executive communications improperly.
[SAFE]: The skill's primary logic is based on standard markdown templates and instructional guidance. It does not contain obfuscated code, remote script execution, or persistence mechanisms.

internal-comms