project-health

Warn

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: MEDIUMDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill explicitly instructs its sub-agents (Permission Auditor and Context Auditor) to write custom Python scripts to .jez/scripts/ and execute them when multi-file scanning or aggregation is required. This dynamic generation and execution of code at runtime bypasses static prompt constraints.
  • [INDIRECT_PROMPT_INJECTION]: The skill defines a repository auditing mechanism that reads external, potentially untrusted files (such as CLAUDE.md, README.md, and other markdown/configuration documents).
  • Ingestion points: The Context Auditor sub-agent reads and scores all CLAUDE.md files, .claude/rules/ topic files, and public markdown documentation across the repository.
  • Boundary markers: There are no boundary markers or instructions telling the agent to ignore or isolate nested instructions found within the audited files.
  • Capability inventory: The skill possesses extensive system command access, file-writing capabilities, and explicit instructions to generate and execute local Python scripts.
  • Sanitization: Content from the repository files is parsed and processed without sanitization or escaping, creating a vulnerability surface where a malicious project file could hijack the sub-agent's execution flow.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 15, 2026, 07:26 AM
Security Audit — agent-trust-hub — project-health