Activation recipes

Problem-led quickstarts that move from a first AgentsKit result to a second ecosystem component.

Activation recipes are short, executable paths for discovering AgentsKit through a result. Each path names at least two components, shows an observable output, and ends with one next action.

These recipes are local and advisory. They do not publish, merge, comment on pull requests, or change the Registry or distribution ledger.

#Start here: first result without a provider key

Install the runtime packages and run the existing first-agent fixture:

npm install @agentskit/core @agentskit/runtime tsx
cat > agent.ts <<'EOF'
import type { AdapterFactory } from '@agentskit/core'
import { createRuntime } from '@agentskit/runtime'

const localAdapter: AdapterFactory = {
  createSource(request) {
    const task = request.messages.at(-1)?.content ?? 'your task'
    return {
      async *stream() {
        yield { type: 'text' as const, content: `Agent ready. I received: ${task}` }
        yield { type: 'done' as const }
      },
      abort() {},
    }
  },
}

const runtime = createRuntime({ adapter: localAdapter })
const result = await runtime.run('Plan my first production agent')
console.log(result.content)
EOF
npx tsx agent.ts

Expected output includes:

Agent ready. I received: Plan my first production agent

Continue with Chat so the same task becomes interactive:

npx @agentskit/chat-cli init ./chat-app --renderer react --yes

The first success metric is time_to_first_result. The next-component metric is second_component_completion.

#Prioritized catalog

Priority	Recipe	Components	Observable result	Next action	Metric
R0	First result without a key	AgentsKit + Chat	runtime output and Chat task	open Chat starter	`time_to_first_result`
R0	Registry agent in Chat	Registry + AgentsKit + Chat	copied `research` agent runs in the same renderer	ask the docs-first question	`second_component_completion`
R0	Docs-first local Chat	Playbook + Chat + Doc Bridge	verified answer with a source path	resolve the handoff	`deterministic_answer_rate`
R0	Handoff before editing	Doc Bridge + Playbook	`startHere`, edit roots, and checks	run advisory review	`handoff_resolution_rate`
R0	Review before merge	Code Review + Playbook + Doc Bridge	advisory findings and a separate gate	add cited retrieval	`review_completion_rate`
R1	Cited document answer	AgentsKit + Chat + Doc Bridge	answer with a valid source ID	compare adapters	`citation_coverage`
R1	Provider swap evaluation	AgentsKit + Playbook + Code Review	fixture invariants preserved across adapters	run bounded agent	`provider_swap_pass_rate`
R1	Sandboxed code agent	AgentsKit + Registry + Code Review	bounded run, cleanup, and review result	return to zero-key start	`sandboxed_run_success`

#Component handoffs

#Registry to Chat

npx agentskit add research

Run the copied source-owned agent with a local adapter or cassette, then keep the Chat renderer unchanged. A provider key is optional for the fixture path.

#Docs-first handoff

npx ak-docs demo --text
npx agents-playbook list
npx agents-playbook run no-any named-exports --cwd "$PWD"

The result should expose a source path, edit root, checks, and an explicit unresolved path for questions outside the deterministic fixture.

#Review before merge

npx @agentskit/code-review --help

For a real local review, use an already configured provider, keep the review advisory, limit files, and preserve the report. Do not grant the recipe permission to comment, merge, release, or publish.

#Quality contract

Every recipe has two or more named components.
Every recipe has a copy/paste command, expected output, CTA, and metric.
The zero-key path remains available without Ollama or another provider.
Analytics may record only recipe identity, component, CTA, duration, and safe error codes.
Model suggestions remain advisory; deterministic checks and human review retain authority.

Explore nearby

✎ Edit this page on GitHub·Found a problem? Open an issue →·How to contribute →