gh-repo-mirror
Pass
Audited by Gen Agent Trust Hub on Oct 31, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPRIVILEGE_ESCALATION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/mirror-repo.zshscript uses thesedcommand with a pipe|delimiter to perform template substitution (e.g.,sed -e "s|{{DESCRIPTION}}|$description|g"). If user-provided inputs like--description,--hero-eyebrow, or--custom-domaincontain the|character or othersed-special characters (like&), it can corrupt the command or lead to unintended content injection in the generated repository files. - [PRIVILEGE_ESCALATION]: The skill's default behavior is to mirror all team access and direct collaborators from the reference repository to the newly created repository. This presents a risk where an agent, if directed to mirror an untrusted repository, might inadvertently grant administrative or write access to the malicious repository's owners on the user's new repository.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes metadata (descriptions, settings, etc.) from external GitHub repositories. A malicious repository could be crafted with instructions in its metadata designed to influence the agent's behavior when it reviews the mirror's plan or verification diffs.
- Ingestion points: Repository metadata and settings fetched via
gh apiinSKILL.mdandscripts/mirror-repo.zsh. - Boundary markers: The script uses distinct section headers (e.g., "Plan", "Done", "Settings drift") in its output, but lacks explicit instructions to the agent to disregard embedded commands in the mirrored data.
- Capability inventory: The skill has extensive capabilities including repository creation, setting modification, GitHub Pages configuration, Cloudflare DNS record management, and collaborator invitation.
- Sanitization: There is no evidence of sanitization or escaping of the metadata retrieved from the reference repository before it is used in templates or displayed to the agent.
Audit Metadata