draft-email

Pass

Audited by Gen Agent Trust Hub on Sep 4, 2026

Risk Level: SAFEDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill reads sensitive local files, specifically the team directory located at .local/team/users.md and processed meeting notes at .local/meetings/. This information is then sent to an external email provider to create drafts as part of its primary functionality.
  • [COMMAND_EXECUTION]: The skill executes the local script scripts/save-draft.py using python3 to transmit the drafted email content to the email provider.
  • [INDIRECT_PROMPT_INJECTION]: The skill has a potential injection surface as it ingests untrusted data from external meeting notes, tickets, and wiki pages. 1. Ingestion points: Reads from .local/meetings/processed/, conversation context, and external documentation URLs. 2. Boundary markers: The skill uses strict HTML templates for drafting but lacks explicit delimiters to segregate untrusted data from instructions. 3. Capability inventory: Executes python3 scripts/save-draft.py and reads local system files. 4. Sanitization: No specific sanitization or filtering of the ingested content is defined before it is formatted into the final email draft.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 4, 2026, 06:29 AM
Security Audit — agent-trust-hub — draft-email