intelligems-profit-impact
Fail
Audited by Gen Agent Trust Hub on Feb 17, 2026
Risk Level: HIGHCOMMAND_EXECUTIONCREDENTIALS_UNSAFEDATA_EXFILTRATION
Full Analysis
- Persistence Mechanisms (HIGH): The skill attempts to create a macOS LaunchAgent via the
setup_automation.shscript to run reports daily. This provides a mechanism for persistent code execution on the user's host system. - Command Execution (HIGH): The instructions execute multiple bash scripts (
setup_workspace.sh,setup_automation.sh) using relative paths (../intelligems-core/) that point outside the skill's own directory. This behavior relies on external, unverifiable files being present on the host. - Credentials Handling (MEDIUM): The skill manages sensitive
INTELLIGEMS_API_KEYdata by reading from and writing to a.envfile. It usesgrepto check for existing keys, which may lead to the accidental display of credentials in the agent's output logs. - Indirect Prompt Injection (LOW): The skill ingests external data from A/B tests through the
impact.pyscript. Ingestion points: output ofimpact.pyand test IDs. Boundary markers: absent. Capability inventory: file writing, shell execution, and Slack webhook network requests. Sanitization: none. This surface allows untrusted experiment data to potentially influence the financial report presentation logic.
Recommendations
- AI detected serious security threats
Audit Metadata