agent: add Prompt method and Activity tracking

[Copilot]

Agents needed a way to receive on-demand prompts at runtime and a general mechanism to observe what they are doing.

Changes

Agent interface additions

• Prompt(text string) <-chan *model.Response — non-blocking; spawns a goroutine that runs the prompt through the model (including tool execution), then delivers the response on a buffered channel and closes it. Channel is closed with no value if no model is configured.
• Activity() []Activity — returns a safe snapshot copy of the agent's activity log, capped at 256 entries.

Activity tracking

• ActivityType string constants: evaluate, prompt, tool, response, error
• Activity struct: Time, Type, Prompt, Tool, Result, Err
• Internal record() helper — thread-safe, drops oldest entries beyond cap
• evaluate() now records each cycle's evaluate/tool/response/error events
• Tool call results are classified as ActivityError vs ActivityTool via JSON inspection of the content string

Usage

a := micro.NewAgent(
    agent.WithName("greeter-manager"),
    agent.WithModel(m),
    agent.WithDirective("Ensure greeter is healthy"),
)
go a.Run()

// Queue a prompt from any goroutine; non-blocking
ch := a.Prompt("how many nodes does greeter have?")
resp := <-ch // blocks only the caller, not the agent loop
fmt.Println(resp.Reply)

// Inspect recent activity
for _, act := range a.Activity() {
    fmt.Printf("[%s] %s %s\n", act.Time.Format(time.RFC3339), act.Type, act.Result)
}

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.