forecast
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection Surface]: The skill is designed to process content from potentially untrusted external sources, such as email threads and chat transcripts, which could contain malicious instructions.
- Ingestion points: Untrusted data enters the agent's context via the CRM and external documents as described in the "Rules" and "Step 6" sections of SKILL.md.
- Boundary markers: The skill includes explicit instructions to treat these sources as "untrusted content" and to interpret them as data rather than instructions.
- Capability inventory: The skill can perform writes to the CRM using the
update-opportunitytool and can generate links to external records. - Sanitization: The instructions require that any "content-originated action" be presented to the user for review before execution and prohibit automated writes from untrusted sources during scheduled runs.
- [Instructional Directives]: The skill contains a directive advising the agent to "never refuse an action the user asked for on the plugin's own authority," deferring instead to the connector's settings. While this is intended to prevent the agent from over-restricting legitimate tasks, such language can sometimes be used to encourage the agent to bypass internal behavioral constraints if a user makes a request that is permitted by a connector but violates other safety principles.
Audit Metadata