git-guardrails-claude-code
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFEPERSISTENCECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PERSISTENCE]: The skill establishes a persistent hook by instructing the agent to modify the global (~/.claude/settings.json) or project-specific (.claude/settings.json) configuration files, which ensures the verification script runs for all subsequent bash tool uses.
- [COMMAND_EXECUTION]: The setup process requires marking the provided blocklist script as executable and integrating it into the agent's command execution pipeline.
- [INDIRECT_PROMPT_INJECTION]: The safety logic processes untrusted tool inputs through jq and grep to identify dangerous patterns. This ingestion of external data for security enforcement constitutes an attack surface where specifically crafted inputs might attempt to bypass the regex blocklist.
- Ingestion points: The script scripts/block-dangerous-git.sh reads tool input from standard input.
- Boundary markers: None present.
- Capability inventory: The script utilizes jq and grep for parsing and matching.
- Sanitization: No sanitization or normalization is performed on the command string before matching against the blocklist.
Audit Metadata