skill-developer
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFENO_CODEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes a hook-based architecture (
UserPromptSubmit) where user prompts are processed and results are injected directly into the agent's context. While this represents an injection surface, it is a documented system feature described for developer use.\n - Ingestion points: User prompts are ingested via the
UserPromptSubmithook as defined in.claude/hooks/skill-activation-prompt.ts.\n - Boundary markers: Present; the guide includes templates that format injected content with decorative visual delimiters.\n
- Capability inventory: The architecture described includes hooks capable of blocking file edits (
PreToolUsein.claude/hooks/skill-verification-guard.ts) and modifying session state.\n - Sanitization: Absent; the skill documentation does not explicitly detail input sanitization or escaping mechanisms for prompt data.\n- [COMMAND_EXECUTION]: The skill provides standard bash commands for testing and validating the hook scripts, utilizing tools like
npx tsxandjq.\n- [NO_CODE]: The skill consists entirely of markdown documentation and reference files; it does not include any executable scripts, binaries, or active hooks in the distributed package.
Audit Metadata