Common tasks

Use this page when you know what you want to build and need the shortest path through the docs. Each task points to the first page to read, then the pages that usually matter once the basic shape is working.

Agent and automation goals

GoalStart hereThen read
Build a Slack code-review botTutorial: code review agentSlack Events connector, Connector catalog, Trigger manifests
Run a release-audit cronTrigger manifestsCron connector, Flow predicate language, Maintainer release workflow
Create an evaluated agent pipelineTutorial: eval pipelineDebugging agent runs, Harn portal, Testing
Add human approval before risky actionsHuman in the loopTrust graph, Host boundary, Workflow runtime
Route work across multiple agentsWorkflow runtimeWorker dispatch, Agent state, Sessions
Reuse a proven prompt or personaPrompt library stdlibPersonas, Skills, Skill provenance

Integration goals

GoalStart hereThen read
Self-host an MCP server with custom toolsTutorial: MCP serverMCP, ACP, and A2A integration, Outbound workflow server, Protocol support matrix
Connect a Cursor automationOrchestrator MCP ServerProtocol support matrix, MCP, ACP, and A2A integration, Host boundary
Connect Harn to an external MCP tool serverRemote MCP quick startMCP, ACP, and A2A integration, Builtin functions, Configuring LLM providers
Handle a GitHub webhook workflowConnector catalogGitHub App connector, Trigger manifests, Orchestrator
Author a reusable connector packageConnector authoringConnector testkit, Package authoring, Rust connectors -> Harn packages
Expose a workflow over HTTP or stdioOutbound workflow serverMCP, ACP, and A2A integration, Bridge protocol, Agents Protocol v1

Operations goals

GoalStart hereThen read
Set up an agent operations consoleHarn portalOrchestrator observability, Trigger observability in the action graph, Debugging agent runs
Deploy a long-running orchestratorOrchestratorDeploy to Fly.io, Deploy to Render, Deploy to Railway
Manage secrets and OAuth for productionOrchestrator secretsConnector OAuth, Configuring LLM providers, Host boundary
Recover or replay failed trigger workOrchestrator DLQ managementTrigger dispatcher, Trigger registry, Debugging agent runs
Control throughput and backpressureOrchestrator backpressureTrigger budgets, Worker dispatch, Concurrency
Audit runtime behavior after a changeFlow predicate languageTranscript architecture, Trust graph, CLI reference

If you are starting from scratch

  1. Install Harn and run your first program with Getting started.
  2. Learn the core syntax with Language basics.
  3. Use LLM calls and agent loops for model-backed work.
  4. Pick one task above and follow only that path until the first version works.
  5. Come back to Best practices before making the workflow write to external systems or run unattended.