parallel-branches
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides bash snippets utilizing Git plumbing commands such as
git read-tree,git write-tree,git commit-tree, andgit update-ref. These commands are used to create commits on non-active branches without requiring a checkout, which is a legitimate use of Git's internal architecture. The skill also incorporates thegh stackextension for the GitHub CLI to handle stacked pull requests.\n- [INDIRECT_PROMPT_INJECTION]: The skill identifies a workflow where multiple subagents interact with and edit shared source code files, representing a potential surface for indirect instructions to enter the agent context.\n - Ingestion points: Source code files and disjoint file lists specified in
SKILL.md.\n - Boundary markers: Absent; there are no instructions to use delimiters or ignore embedded instructions within the files being edited.\n
- Capability inventory: The process involves executing shell commands, performing file system operations via Git, and interacting with remote repositories (documented in
SKILL.md).\n - Sanitization: No sanitization or validation mechanisms are mentioned for the content of the files processed by the agents.
Audit Metadata