kelly-financial-services-intel
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes several utility scripts for lifecycle management, including
scripts/setup.mjsfor workspace initialization andscripts/execute_decisions.mjsfor processing human-approved actions. These scripts interact with the Busabase SDK and are intended for administrative use by the operator. - [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting untrusted external data from public sources like market news and regulatory updates, which presents a surface for indirect prompt injection attacks.
- Ingestion points: The agent is instructed to browse and summarize public financial news, central bank updates, and competitor announcements (SKILL.md).
- Boundary markers: The skill implements a strict review gate; signals and actions are held in a Busabase-backed "cockpit" (AirApp) for human approval before any external handoff is performed.
- Capability inventory: The skill has the ability to read and write to the Busabase data layer and provides local scripts to dry-run or finalize handoff statuses.
- Sanitization: The AirApp UI code includes standard HTML escaping for rendering data (
app.js) and utilizes a normalization layer infinancial-model.jsto enforce schema constraints on ingested data.
Audit Metadata