b2-cloud-storage
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill interacts with the Backblaze B2 service using the official
b2CLI and a Python audit script. All subprocess calls inscripts/storage_audit.pyuse list-based arguments (e.g.,subprocess.run(["b2", *args])), which effectively prevents shell injection even when processing user-supplied bucket names. - [DATA_EXPOSURE]: The instructions establish strong security boundaries by explicitly forbidding the agent from reading the B2 account database file (
~/.b2_account_info) or Mutation of credentials viab2 key *commands. It includes instructions to warn the user if secrets are inadvertently shared in the chat transcript. - [INDIRECT_PROMPT_INJECTION]: The storage audit functionality ingests file metadata (names, sizes, timestamps) from external B2 buckets. The skill handles this data surface safely by parsing the information as structured JSON and treating all bucket-sourced strings as data rather than instructions. Mandatory safety rules (Rule 3 and 5) ensure that the agent does not perform destructive actions without explicit, verified human consent.
- [SAFE]: The skill does not contain any obfuscated code, persistence mechanisms, or unauthorized network exfiltration. External dependencies are limited to the official B2 CLI and the Python standard library, with instructions for the user to handle installation themselves to maintain system integrity.
Audit Metadata