The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted evidence from external sources, creating a potential surface for indirect prompt injection attacks.
Ingestion points: Workflow Step 2 in SKILL.md identifies metrics, incidents, and customer signals as inputs.
Boundary markers: There are no specific delimiters defined to isolate external evidence from core instructions.
Capability inventory: The skill contains no executable scripts, tool permissions, or network access, preventing the exploitation of any injected instructions.
Sanitization: No validation or filtering of evidence content is specified.
[NO_CODE]: The skill consists strictly of markdown instructions and YAML metadata with no executable scripts or binary assets, minimizing the technical attack surface.

risk-register