agent-hook-guard
Warn
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: MEDIUMDYNAMIC_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DYNAMIC_EXECUTION]: The skill generates custom scripts (e.g., shell or python) based on natural language problem descriptions found in reports and executes them locally to verify the hook logic.
- [PERSISTENCE]: The skill installs automated hooks into agent configuration files (such as Claude Code settings), ensuring the checks run automatically in all future sessions.
- [INDIRECT_PROMPT_INJECTION]: The skill reads from potentially untrusted external reports to determine the implementation of automated guards.
- Ingestion points: Reads data from
.scratchpad/<slug>/redteam-report.md,verification-report.md, andbrowser-test-report.md. - Boundary markers: The skill does not implement specific delimiters or safety instructions to prevent embedded malicious prompts in the reports from influencing the agent's script generation.
- Capability inventory: The skill can write files, execute arbitrary local scripts, and perform git commits.
- Sanitization: No explicit sanitization or validation of the input report data is performed before it is used to generate executable code.
- [COMMAND_EXECUTION]: The skill executes generated scripts and git commands to test and deploy the automated hooks.
Audit Metadata