kelly-retail-intel
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external public sources, including news, weather reports, and customer reviews. This ingestion of untrusted content represents a potential attack surface for indirect prompt injection.\n
- Ingestion points: Public web sources, competitor websites, and retail trend monitors as described in
SKILL.md.\n - Boundary markers: The workflow includes a mandatory human-in-the-loop review gate via the Busabase App UI; all actions must be approved before execution.\n
- Capability inventory: The skill utilizes file writing (Busabase records) and local shell execution for handoff operations.\n
- Sanitization: The agent is instructed to evaluate buyer intent and mark items with low confidence as 'blocked'.\n- [COMMAND_EXECUTION]: The skill involves the execution of local management scripts and development tools to operate the retail cockpit.\n
- Evidence:
SKILL.mdandREADME.mdprovide instructions for runningnode scripts/execute_decisions.mjsandpnpm devfor local previewing and decision application.\n - Risk Mitigation: The execution script defaults to a dry-run mode, requiring an explicit
--applyflag and human confirmation before marking actions as completed.
Audit Metadata