Parent PRD

#2

What to build

Add agentskit run command to @agentskit/cli. This command executes runtime agents directly from the terminal — no interactive chat, just task → result.

Accepts a task string plus flags for provider, model, skill, tools, memory, system prompt, max steps, etc. Uses AgentRunner from @agentskit/runtime under the hood. Outputs the agent's final result to stdout (or streams steps with --verbose).

Acceptance criteria

• agentskit run <task> command added to CLI
• Flags: --provider, --model, --key, --skill, --tools, --memory, --system-prompt, --max-steps, --verbose
• Resolves provider from flags or environment variables (same as agentskit chat)
• Executes task via AgentRunner, outputs final result to stdout
• --verbose flag streams agent steps (tool calls, reasoning) to stderr
• Non-zero exit code on agent failure
• Unit tests for flag parsing and provider resolution
• Integration test: agentskit run with demo provider completes a task
• Help text and usage examples

Blocked by

• Blocked by Minimal AgentRunner with ReAct loop #6 (runtime — AgentRunner)

User stories addressed

• User story 31 (agentskit run executes agents from terminal)