The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages the instructions field for agents, which is defined as the durable behavior contract and system prompt for the agent's runtime. While this is the intended purpose of the management tool, it represents a surface where user-supplied content could influence agent behavior.
Ingestion points: multica agent create --instructions and multica agent update --instructions in SKILL.md.
Boundary markers: The documentation distinguishes between the description (metadata) and instructions (behavior contract), clarifying that only the latter is injected into the runtime prompt.
Capability inventory: The skill uses the Bash tool to execute multica CLI commands for agent management.
Sanitization: Standard CLI validation is performed on numeric fields like concurrency, but the instruction content is passed as the runtime behavior contract.
[COMMAND_EXECUTION]: The skill references administrative and testing commands used to verify the platform's agent creation contracts against its source code.
Evidence: go test ./internal/service -run TestCreatingAgentsSkillCoversAgentCreationContracts and other verification commands located in references/creating-agents-source-map.md.
Context: These commands are provided for conformance evaluation by developers working within the platform's ecosystem.

multica-creating-agents