T03 · Remote Payload Retrieval and Execution
- Location
scripts/cluster_diagnostics.py:50- Finding
Runtime Retrieval and Execution of an Unpinned Remote MCP Server
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill has a coherent ByteHouse diagnostics purpose, but it runs mutable remote code and passes it the full local environment, which could expose more secrets than intended.
Review carefully before installing. Use only in an isolated environment with a dedicated least-privilege ByteHouse account, remove unrelated secrets from the environment before running it, and prefer a version pinned to an audited MCP server commit and exact dependencies. Treat generated reports as sensitive operational data.
scripts/cluster_diagnostics.py:50Runtime Retrieval and Execution of an Unpinned Remote MCP Server
scripts/cluster_diagnostics.py:46Complete Parent Environment Exposed to the MCP Child Process
scripts/cluster_diagnostics.py:20Unbounded MCP Dependency Version Allows Unreviewed Future Releases
The script copies the entire parent process environment and passes it wholesale into a child process that is launched from a remote GitHub-sourced MCP server package. This can expose unrelated secrets present in the environment, not just the intended ByteHouse credentials, to code outside the script's trust boundary; because the child code is fetched from @main, the risk is amplified by supply-chain drift and potential secret exfiltration.
print()
# 从环境变量获取配置
env = os.environ.copy()
# MCP Server参数
server_params = StdioServerParameters(
The README explicitly advertises query-history analysis and writing multiple JSON reports to a local output directory, but it does not warn that those reports may contain operational metadata such as query text, error details, host or cluster topology, and timing information. In a diagnostics skill, this omission can lead users to collect and persist sensitive telemetry unintentionally, increasing the risk of local disclosure, log leakage, or accidental sharing of infrastructure details.
The skill instructs users to export sensitive ByteHouse credentials, write diagnostic output files, and rely on an MCP-backed service, but it does not declare any explicit tool scope such as permissions or allowed-tools. This creates a trust and containment gap: an agent may access environment secrets, perform file writes, or invoke MCP capabilities without the skill clearly constraining or documenting those actions.
The skill’s natural-language interface and instructions are entirely presented in Chinese, including the manifest description and invocation guidance, with no indication that users may interact in other languages or opt into Chinese. Under the policy rule for language or locale constraints, this is a natural-language policy concern unless the locale restriction is explicitly justified as region-specific.
The natural-language instructions in this file are fully Chinese, which effectively forces a specific language for users without opt-in or an alternative. Under the stated policy, language constraints should either offer user choice or be explicitly justified as region-specific.