blacksmith-testbox
Fail
Audited by Gen Agent Trust Hub on Apr 24, 2026
Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The installation instructions direct the agent to download a script from
https://get.blacksmith.shand pipe it directly to the shell (| sh). This executes remote code with the privileges of the current user without prior inspection or verification. - [COMMAND_EXECUTION]: The skill provides an interface to execute arbitrary shell commands on a remote VM via the
blacksmith testbox runcommand, effectively acting as a remote shell wrapper. - [COMMAND_EXECUTION]: The file synchronization mechanism uses
rsyncwith the--deleteflag. The instructions contain a critical warning that if the agent executes these commands from a subdirectory instead of the repository root, the sync process will delete local directories and files to match the remote state. - [DATA_EXFILTRATION]: The skill's primary function is to sync local repository data, including source code and untracked files, to a remote cloud environment managed by a third-party service (Blacksmith).
- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests data from local files and processes untrusted command strings for remote execution.
- Ingestion points: Local file system (via rsync sync), environment variables (
BLACKSMITH_CHANNEL,BLACKSMITH_ORG), and command-line arguments. - Boundary markers: The skill does not use delimiters or instructions to prevent the agent from obeying commands embedded within the files being synced or the commands being executed.
- Capability inventory: Remote command execution, local file synchronization (read/write), and network access to external cloud providers.
- Sanitization: There is no evidence of input validation or sanitization for the command strings passed to the
runordownloadfunctions.
Recommendations
- HIGH: Downloads and executes remote code from: https://get.blacksmith.sh - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata