The Agent Skills Directory

[DATA_EXFILTRATION]: The skill aggregates highly sensitive information from several sources including Gmail history, meeting transcripts from Granola, and CRM data from Attio.
It snapshots commercial data (such as pricing and proposals) and PII to local storage within the revops/customers/ directory.
Although no network exfiltration to unauthorized domains was detected, the persistence of aggregated sensitive data to the local file system increases the impact of a potential local security compromise.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted content from external communications without sufficient isolation.
Ingestion points: Raw email content retrieved via gmail_read_thread and meeting transcripts fetched via get_meeting_transcript (SKILL.md).
Boundary markers: The workflow lacks specific delimiters or instructions that would prevent the agent from interpreting instructions embedded within the emails or transcripts.
Capability inventory: The skill has powerful write capabilities, including local file system writes and the ability to create email drafts using gmail_create_draft.
Sanitization: There is no evidence of filtering or sanitization performed on external inputs before they are included in the prompt context or stored locally.

deal-reengagement