repos
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill possesses a standard attack surface for repository management tools as it must ingest and process untrusted configuration data from repository files.
- Ingestion points: Processes
.github/workflows/*and repository metadata during auditing and migration tasks (SKILL.md, references/ci.md). - Boundary markers: Relies on default agent parsing; no specific delimiters or ignore-instructions are defined for processing external file content.
- Capability inventory: Performs file writes to the local repository, executes GitHub API mutations (create, settings, archive), and manages Pulumi ESC environments.
- Sanitization: Explicitly instructs the agent to never print secret values in logs or chat and to avoid echoing sensitive data (references/secrets-esc.md).
- [EXTERNAL_DOWNLOADS]: The skill references official GitHub Actions from well-known and trusted providers (GitHub and Pulumi) for CI workflows and secret management. It also provides hardening guidance, such as pinning these actions to commit SHAs to mitigate supply-chain risks.
- [COMMAND_EXECUTION]: Utilizes administrative git commands, specifically
git filter-repo, to perform repository splitting and history management operations. These commands are within the scope of the skill's primary purpose.
Audit Metadata