high-concurrency
Fail
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: HIGHINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONSAFE
Full Analysis
- [REMOTE_CODE_EXECUTION]: Automated scanners flagged a pattern involving
curlpiped topython3. A technical review of the code inSKILL.mdconfirms this is a false positive. The commandcurl -s "http://$be:8040/api/query_cache/stat" | python3 -m json.tooluses the local Pythonjson.toolmodule to format JSON output from an internal StarRocks Backend (BE) metrics API. This is a standard administrative procedure and does not execute remote scripts. - [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze
fe/log/fe.audit.log. This file contains raw SQL queries submitted by users, which serves as a surface for indirect prompt injection. If an attacker submits a query containing malicious instructions, an AI agent reading the log might inadvertently follow them if proper isolation or sanitization is not applied. - Ingestion points:
fe/log/fe.audit.log - Boundary markers: Absent
- Capability inventory: File system read, local command execution (grep, awk, curl), database administration via SQL
- Sanitization: Absent
- [COMMAND_EXECUTION]: The skill contains several bash snippets using
grep,awk, andsortto extract performance metrics from log files and configuration files (e.g.,fe/conf/fe.conf). These are standard operational tasks for managing a StarRocks cluster.
Recommendations
- HIGH: Downloads and executes remote code from: http://$be:8040/api/query_cache/stat - DO NOT USE without thorough review
Audit Metadata