analyzing-system-throughput

Pass

Audited by Gen Agent Trust Hub on Apr 4, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes external system metrics and directory contents, which presents a surface for indirect prompt injection.\n
  • Ingestion points: The skill is designed to read metrics from ${CLAUDE_SKILL_DIR}/metrics/throughput/ and iterate through user-provided directory paths.\n
  • Boundary markers: No delimiters or instructions to ignore instructions within ingested data are present in the skill instructions or scripts.\n
  • Capability inventory: The skill has access to Read, Write, Grep, and Bash tools.\n
  • Sanitization: There is no evidence of data sanitization or validation in the provided analysis scripts.\n- [SAFE]: Documentation and implementation discrepancies were identified but do not pose a security risk.\n
  • The file scripts/identify_bottlenecks.sh contains Python code despite its shell extension.\n
  • The script's actual behavior (file iteration and size checking) does not match its description of using tools like top or iostat, indicating it is likely a placeholder.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 4, 2026, 01:52 PM