The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements an interactive workflow that ingests untrusted data, representing a potential indirect prompt injection surface.
Ingestion points: Free-text inputs for pipeline identifiers (Phase 1), image references, and repository URLs (Phase 8) are collected from the user. Additionally, the skill reads existing pipeline YAML using the harness_get tool (Phase 1).
Boundary markers: The generated YAML for the SscaOrchestration step does not appear to use specific boundary markers or 'ignore' instructions when interpolating these user-provided strings.
Capability inventory: The skill possesses the capability to modify remote configurations via the harness_update MCP tool. Across the provided scripts, capabilities include file system access and network operations (implicit in the Harness MCP tools).
Sanitization: No explicit validation or escaping logic for the user-supplied strings is defined in the instruction set before they are included in the harness_update payload.

create-sbom