SG-005: Indirect Prompt Injection
Severity: HIGH
Instruction sets embedded in metadata fields (description, name)
Fires when
- The skill's metadata 'description' or 'name' field contains an instruction-override pattern (e.g. ignore previous instructions, persona replacement, safety override)
- Metadata fields are processed by the AI, so an override planted there is an indirect injection even though it never appears in the visible skill body
Related CVEs
Remediation
No safe remediation: a file that triggers this rule should be rejected, not patched. Read the flagged lines in the source before deciding.