altinity-profiler-clickhouse

Pass

Audited by Gen Agent Trust Hub on Jun 18, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits a potential surface for indirect prompt injection. It ingests data from the target ClickHouse cluster's system.query_log and system.tables (referenced in pipeline.md and SKILL.md). This data is then used to synthesize persistent markdown artifacts. Although Hard Rule 2 in SKILL.md requires query normalization and the stripping of comments, an adversary who can execute queries on the target cluster might attempt to influence the generated 'analyst' Skill by injecting specific patterns into the logs.
  • [COMMAND_EXECUTION]: The profiling pipeline relies on the execution of local Python helper scripts: tools/pareto_cut.py and tools/synthesize_conventions.py. These scripts are used for data processing and thresholding during the Pareto demotion and convention synthesis phases. Analysis of the script source code shows they use standard Python libraries for data manipulation and do not perform network requests, file system modifications outside the output folder, or unauthorized subprocess execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 18, 2026, 05:44 AM
Security Audit — agent-trust-hub — altinity-profiler-clickhouse