The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process and modify Nika workflow files (.nika.yaml) from the user's workspace, creating an attack surface for indirect prompt injection. \n
Ingestion points: The skill interacts with existing .nika.yaml files via commands like nika list, nika inspect, and nika check (SKILL.md). \n
Boundary markers: The skill documents the use of a permits: block to establish security boundaries and a default-deny posture (SKILL.md). \n
Capability inventory: The language supports subprocess execution (exec:), tool invocation (invoke:), and LLM interaction (infer:, agent:) as detailed in SKILL.md. \n
Sanitization: It instructs the agent to use nika check for auditing and describes the lift: mechanism for explicit handling of untrusted data (SKILL.md). \n- [DYNAMIC_EXECUTION]: The skill facilitates the creation and execution of dynamic workflows. \n
Evidence: It documents the invoke: { workflow: ... } syntax for dynamic composition and the lift: law: data-as-code parameter for handling executable payloads (SKILL.md). \n- [COMMAND_EXECUTION]: The skill provides documentation for executing system commands. \n
Evidence: Detailed instructions are provided for the exec: verb, including the use of command: arrays and shell: environments (SKILL.md). \n- [EXTERNAL_DOWNLOADS]: The skill references mechanisms for fetching external templates and components. \n
Evidence: Mentions the use of nika new to fetch example workflows and the registry:owner/name@version syntax for importing remote components (SKILL.md).

nika-authoring