hooks-manager
Pass
Audited by Gen Agent Trust Hub on Aug 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent on how to configure and manage shell command hooks that execute in response to platform events (e.g., PreToolUse, Stop). It provides templates for automated tasks like linting with Prettier and logging command history to local files.
- [PROMPT_INJECTION]: The skill manages platform configuration files that define agent behavior, which constitutes an indirect prompt injection surface.
- Ingestion points: Platform settings files in project and user home directories (e.g., .claude/settings.json, .codex/hooks.json, .cursor/hooks.json) as described in references/router.md.
- Boundary markers: The instructions do not specify explicit delimiters for reading external configuration files.
- Capability inventory: Arbitrary shell command execution and local file writes (e.g., ~/.claude/bash-log.txt) via the configured hook system.
- Sanitization: Recommends using jq for JSON validation and implementing loop-prevention logic (stop_hook_active) to ensure stability.
- [SAFE]: All external URLs and documentation references point to official first-party developer resources for Claude Code, OpenAI, and Cursor. The skill emphasizes safety requirements like reasonable timeouts and the use of trusted absolute paths.
Audit Metadata