The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from user inputs and existing Harness pipeline configurations, which creates a surface for indirect prompt injection.
Ingestion points: User-provided values for chaos step names, durations, and runtime variables (Step 3 in references/components.md); existing pipeline YAML fetched via the harness_get tool (Step 4 in references/components.md and Step U3 in references/edit.md).
Boundary markers: Absent. The skill does not employ specific delimiters or instructions to prevent the agent from interpreting embedded content within the YAML or user inputs as new instructions.
Capability inventory: The skill has the ability to create and update pipelines via the harness_create and harness_update tools. These operations modify the configuration of chaos experiments executed against cloud infrastructure.
Sanitization: Partial. The skill includes a regex validation for auto-derived identifiers (^[a-zA-Z_][a-zA-Z0-9_$]{0,127}$), but does not provide sanitization or escaping for free-text fields or variable values that are interpolated into the final YAML manifest.

chaos-dr-test