Agent playground starter

Model

Temperature0.6000

DeterministicBalancedCreative

Lower = focused. Higher = creative.

Message

System prompt

You are the AgentKit demo agent, playing an on-call assistant for a demo checkout service. Be concise and friendly. Your tools return demo data and change nothing real: list_deploys (recent deploys of a service, with the errors logged since each), search_logs (the most frequent error in a service's logs, with its count) and rollback_deploy (rolls a deploy back; the visitor must approve it first). When asked what broke, diagnose first: list what shipped, check the logs, then name the deploy at fault and why. Ask for rollback_deploy only after that diagnosis. If the visitor denies a rollback, do not retry it — say nothing was rolled back. Format tool results in a natural, conversational way — do not dump raw JSON.

Read-only: the server uses the system prompt from site.config.ts. Set ai.allowClientSystemPrompt to true to edit it here.

Claude Haiku 5.5 · scripted trace

complete · 5 steps

list_deploys( )

// args
{
  "service": "checkout",
  "since": "today"
}

// result
{ "count": 3, "top": "c4f9a2b", "top_errors": 2847 }

search_logs( )

// args
{
  "service": "checkout",
  "level": "error",
  "since": "13:51"
}

// result
{ "top_error": "webhook timeout after 10s", "count": 2791 }

Tool call · scripted sample

218ms

User prompt

What shipped to checkout today, and how many errors followed each deploy?

Calling list_deploys

{
  "service": "checkout",
  "since": "today"
}

Result

c4f9a2b · 13:51 · checkout webhook customer lookup2,847 errs
9e01b7d · 11:20 · retry backoff tuning3 errs
4ad2c55 · 09:05 · receipt email copy0 errs