commit
Pass
Audited by Gen Agent Trust Hub on Oct 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local git commands to manage the repository workflow.
- The skill runs
git diff --stagedandgit diff --staged --statto inspect changes. - It performs destructive or state-changing operations including
git add -A(when the staging area is empty),git commit -m, andgit push origin HEAD. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection by processing external data as instructions.
- Ingestion points: The skill reads the output of
git diff --staged, which contains arbitrary text from the repository's files (SKILL.md, Step 1). - Boundary markers: There are no explicit boundary markers or instructions to the model to ignore potential directives embedded within the source code diffs.
- Capability inventory: The skill has the capability to write to the repository history via
git commitand upload changes viagit push(SKILL.md, Step 6). - Sanitization: There is no evidence of sanitization or filtering of the content retrieved from the git diff before it is passed to the LLM for summarization.
Audit Metadata