claude-code-expert

Warn

Audited by Gen Agent Trust Hub on Sep 10, 2026

Risk Level: MEDIUMPRIVILEGE_ESCALATIONMETADATA_POISONINGINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill explicitly recommends the use of the --dangerously-skip-permissions flag for the Claude Code CLI, which bypasses the tool's primary safety mechanism that requires human confirmation for dangerous actions.\n- [METADATA_POISONING]: The skill's YAML frontmatter contains the field risk: none, which is a deceptive claim regarding the skill's safety and could mislead users or automated systems.\n- [INDIRECT_PROMPT_INJECTION]: The skill describes an architecture centered around CLAUDE.md and memory files that serve as persistent context ingestion points. This creates a vulnerability surface where malicious instructions in project files could be automatically executed by the agent.\n
  • Ingestion points: Project-level CLAUDE.md files and memory/ directories.\n
  • Boundary markers: None specified to delimit untrusted project content from instructions.\n
  • Capability inventory: Full filesystem access (Read/Write), arbitrary bash command execution, and MCP tool orchestration.\n
  • Sanitization: No mention of sanitization or validation for the content of these context files.\n- [COMMAND_EXECUTION]: The skill provides and encourages the configuration of 'hooks' in settings.json that execute shell commands automatically on system events, such as PowerShell beeps and file logging.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 10, 2026, 07:46 AM
Security Audit — agent-trust-hub — claude-code-expert