git-guardrails-claude-code
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFEPRIVILEGE_ESCALATIONPERSISTENCEINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to modify sensitive configuration files (
~/.claude/settings.jsonand.claude/settings.json) and executechmod +xto grant execution permissions to the guardrail script. - [PERSISTENCE]: The skill sets up a
PreToolUsehook, which is a mechanism that persists in the agent's environment to intercept and analyze every tool call before execution. - [INDIRECT_PROMPT_INJECTION]: The guardrail script ingests untrusted tool input data at runtime to perform pattern matching.
- Ingestion points: The
scripts/block-dangerous-git.shscript reads from standard input (stdin) to capture tool call data. - Boundary markers: None present; the script processes the raw tool input string.
- Capability inventory: The script itself is limited to pattern matching and exiting with status codes; it does not perform network operations or file writes.
- Sanitization: The script uses
jqto safely parse the JSON-formatted tool input before extracting the command string for inspection.
Audit Metadata