skill-developer

Pass

Audited by Gen Agent Trust Hub on Sep 18, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill documents an auto-activation system that ingests untrusted data, such as user prompts and file contents, to trigger context injection or tool blocking.\n
  • Ingestion points: The system ingests user prompts via the UserPromptSubmit hook and file metadata/content via the PreToolUse hook, as detailed in HOOK_MECHANISMS.md.\n
  • Boundary markers: The guide illustrates the use of visual delimiters for injected suggestions but does not mandate or describe sanitization of the untrusted input before regex processing.\n
  • Capability inventory: The described hooks possess the capability to inject arbitrary text into the agent's context and prevent the execution of file-editing tools.\n
  • Sanitization: The documentation lacks instructions for escaping or validating content extracted from user prompts or project files.\n- [COMMAND_EXECUTION]: The skill includes shell and npx command examples for verifying the functionality of local hook scripts within the .claude/hooks directory.\n
  • Evidence: The guide provides commands such as 'npx tsx .claude/hooks/skill-activation-prompt.ts' and shell heredocs to test hook logic against simulated inputs.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 18, 2026, 03:47 PM
Security Audit — agent-trust-hub — skill-developer