The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes automation workflows that ingest untrusted data from external sources, which could contain malicious instructions designed to influence the agent's logic or downstream actions.
Ingestion points: Gmail triggers and HTTP module responses (SKILL.md).
Boundary markers: The provided examples lack explicit delimiters or warnings to the agent to disregard instructions embedded within the processed data.
Capability inventory: The skill enables capabilities such as uploading files to Google Drive, extracting data via OpenAI modules, and performing actions in third-party services like Slack and QuickBooks (SKILL.md).
Sanitization: While the skill mentions text manipulation functions like substring and replace, it does not include specific sanitization logic to mitigate instruction injection from untrusted inputs.

make-com