checkpoint-commit

Pass

Audited by Gen Agent Trust Hub on Mar 3, 2026

Risk Level: SAFE
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes standard Git commands (git diff, git add, git commit) to manage repository state. These operations are performed locally and align with the skill's primary function of maintaining work-in-progress checkpoints.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface as it analyzes untrusted repository content (via git diff) to generate commit summaries. This is an inherent part of the skill's workflow and is mitigated by the agent's internal safety filters.
  • Ingestion points: File content and diffs are read in Step 2 of SKILL.md using git diff and git ls-files.
  • Boundary markers: No explicit delimiters are used to wrap the diff content before summary generation.
  • Capability inventory: The skill can stage files and create commits locally. It lacks network access or arbitrary system command execution.
  • Sanitization: No sanitization is performed on the diff output before it is processed by the language model.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 3, 2026, 12:33 PM