em-grid-scorer
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from user calendars and Slack message streams, making it susceptible to indirect prompt injection if an event title or message contains adversarial instructions.
- Ingestion points: Calendar event data loaded from the Calendar MCP (
list_events) and raw message content retrieved through the Slack MCP (conversations.listand text history sampling) as described inSKILL.md. - Boundary markers: Absent. The instructions do not define clear delimiters or add explicit instructions to ignore prompt injection strings within the ingested data during processing.
- Capability inventory: The skill uses file write/read tools to manage data within
/tmp/and relies on shell execution capabilities to invoke the bundled calendar processing script (scripts/process_calendar.py). - Sanitization: Absent. The local script
scripts/process_calendar.pynormalizes titles using standard regular expressions for grouping purposes, but it does not validate or sanitize text against malicious or deceptive language constructs.
Audit Metadata