git-guardrails-claude-code
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes command strings to detect and block destructive Git operations.
- Ingestion points: The
scripts/block-dangerous-git.shscript parses the command string from the agent's tool input. - Boundary markers: The script operates on the full command string provided by the agent without specific delimiters.
- Capability inventory: The script's only capability is to signal a block by exiting with a non-zero code.
- Sanitization: The input is processed as data for regex matching and is not executed as a command within the guardrail script itself.
- [SAFE]: The skill's actions, such as modifying the agent's
.claude/settings.jsonand setting execution bits on its own script, are legitimate configuration steps for implementing a local security hook.
Audit Metadata