git-safety-net
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data such as Git commit history, pull request metadata, and repository status.
- Ingestion points:
scripts/git_classify_refs.shingest JSON-formatted PR maps, andscripts/git_loss_audit.shparses output from several Git subcommands. - Boundary markers: The skill uses structured data formats (JSON, TSV) and clear markdown sections to separate instructions from data. It includes an 'Outcome contract' to explicitly bound the agent's authorized actions.
- Capability inventory: The skill possesses capabilities to modify the working tree (
scripts/git_align_checkout.sh), update Git references (scripts/git_preserve_danglers.sh), and generate backup archives. - Sanitization: Scripts use shell escaping (e.g.,
printf %q) and Git-specific guards (e.g.,--to separate refs from pathspecs) to handle potentially malicious filenames or ref names safely. - [DYNAMIC_EXECUTION]: The skill uses local Python execution for internal logic.
- Evidence:
scripts/git_classify_refs.shuses an inline Python script for JSON parsing, and the test suite utilizes the Pythonunittestframework to verify script behavior. These operations are restricted to processing local files or standard library functions and do not execute untrusted remote code.
Audit Metadata