craft-experiment-readout
Pass
Audited by Gen Agent Trust Hub on Apr 15, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as an instructional template for the agent to draft experiment readouts and stakeholder communications based on provided metrics. It contains no executable scripts or remote dependencies.
- [PROMPT_INJECTION]: The template incorporates user-provided metrics via the
$ARGUMENTSvariable. While this is an ingestion point for external data, the template uses<context>boundary markers to separate data from instructions. The skill has no capabilities beyond text generation (no subprocess calls, file writes, or network operations), which significantly limits the risk of indirect prompt injection. - [DATA_EXFILTRATION]: There are no tools or commands that could be used to exfiltrate data to external servers. The skill operates purely within the text-based conversational context.
- [COMMAND_EXECUTION]: The skill does not include any shell scripts, binary files, or commands that would allow for arbitrary code execution.
Audit Metadata