amazon-working-backwards
Pass
Audited by Gen Agent Trust Hub on Apr 10, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill incorporates untrusted user data into its prompt instructions, creating a surface for indirect prompt injection.
- Ingestion points: User input is passed via the
$ARGUMENTSvariable into the<context>block inSKILL.md. - Boundary markers: The template uses
<context>tags to delimit user data from instructions. - Capability inventory: The skill does not explicitly invoke external tools or scripts, though the environment does not restrict tool access via the YAML frontmatter.
- Sanitization: No sanitization is applied to the
$ARGUMENTScontent before it is interpolated into the model prompt.
Audit Metadata