crabbox
Pass
Audited by Gen Agent Trust Hub on Oct 12, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill makes extensive use of the
crabboxandblacksmithCLI tools to configure remote backends, perform environment preflight checks, and execute verification commands. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to handle external data from forks and pull requests, representing a potential attack surface. It mitigates this risk by providing strict boundary instructions.
- Ingestion points: Untrusted data enters the agent context through the
<check-command>and code referenced by--fresh-pr. - Boundary markers: The instructions explicitly direct the agent to unset sensitive AWS and Tailscale environment variables and use the
--network publicflag to ensure isolation. - Capability inventory: The skill utilizes shell execution via the
crabboxbinary for all remote operations. - Sanitization: It mandates the use of a
<trusted-bootstrap-script>to verify the integrity of the remote environment and pin toolchains before running any user-provided code. - [DATA_EXPOSURE]: The skill provides specific warnings regarding the collection of artifacts and logs, advising the agent to treat them as potentially containing sensitive information until manually reviewed.
Audit Metadata