internal-docs

Warn

Audited by Gen Agent Trust Hub on Apr 7, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill defines a 'Companion check' procedure that instructs the agent to run a shell command (ls) against hidden directories in the user's home directory (such as ~/.claude/skills/ and ~/.agent/skills/) to discover other installed software upon activation.
  • [REMOTE_CODE_EXECUTION]: The skill directs the agent to suggest and provide commands for installing additional third-party packages from an unverified repository using the npx skills add command, facilitating the execution of remote code.
  • [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes untrusted user documents while maintaining active command execution and package installation capabilities. Ingestion points: Processes user-provided engineering document drafts and incident logs. Boundary markers: Absent; there are no instructions to ignore or isolate embedded directives in the content being reviewed. Capability inventory: Shell command execution (ls) and remote software installation recommendations (npx) specified in SKILL.md. Sanitization: Absent; the skill does not specify any validation or filtering for document content.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Apr 7, 2026, 03:04 PM
Security Audit — agent-trust-hub — internal-docs