The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of AI agents by generating task instructions from user-provided goals and workflows. This presents an indirect prompt injection surface if the requirements provided are influenced by untrusted external data.
Ingestion points: Agent goals, inputs, outputs, and workflows gathered during the interactive requirements phase in SKILL.md.
Boundary markers: The skill explicitly instructs the inclusion of a ## RULES section at the end of the PLUGIN_TASK to enforce constraints on the created agent's behavior.
Capability inventory: Generated agents have access to specific network domains via allowedDomains and external services through mcpConnectors and llmConnector.
Sanitization: The skill enforces a mandatory interactive review step (Phase 4) where the user must review the full YAML specification and provide explicit confirmation before the agent is created or updated.
[EXTERNAL_DOWNLOADS]: The skill references a Harness-owned container image to serve as the runtime environment for the created agents.
Evidence: The agent specification uses the image pkg.harness.io/vrvdt5ius7uwygso8s0bia/harness-agents/harness-ai-agent:latest.

create-agent