rules-distill

Warn

Audited by Gen Agent Trust Hub on Apr 14, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses bash to execute local scripts (scan-skills.sh and scan-rules.sh) located in the skill's own directory (~/.claude/skills/rules-distill/scripts/). These scripts utilize find, grep, awk, and jq to traverse the user's home directory to locate .claude/skills and .claude/rules directories.
  • [EXTERNAL_DOWNLOADS]: While no hardcoded external URLs are present, the skill's architecture involves feeding the full text of local skill and rule files into a sub-agent (LLM). This pattern creates a surface for Indirect Prompt Injection if the skills being scanned contain malicious instructions designed to influence the distillation process.
  • [COMMAND_EXECUTION]: The shell scripts use environment variables (RULES_DISTILL_DIR, RULES_DISTILL_GLOBAL_DIR, RULES_DISTILL_PROJECT_DIR) to determine scan paths. While intended for testing, this pattern allows for dynamic path configuration that could be misused to scan unintended directories if the environment is compromised.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 14, 2026, 12:19 AM
Security Audit — agent-trust-hub — rules-distill