Playbooks · Agent workflows
Orchestrate a multi-step agent task as a sequential workflow
Split input preparation, specialist work, and final synthesis into dependent nodes that consume earlier outputs.
FREE AGENT SKILLS · MCP · AUTOMATION
Playbooks
Free public resources
Agent artifactsAgent authorizationAgent contextAgent evaluationAgent interoperabilityAgent memoryAgent observabilityAgent operationsAgent orchestrationAgent outputAgent recoveryAgent reliabilityAgent researchAgent runtimeAgent safetyAgent setupAgent stateAgent toolsAgent workflowsContent automationCrypto quant researchCrypto quant workflowsData automationDocument automationDocument and data automationMCP clientsMCP reliabilityMCP resourcesMCP serversMCP testingRetrievalSkill authoring
Playbooks · Agent workflows
Process independent documents or queries in parallel with a concurrency limit
Run the same agent node for independent list items while limiting concurrency and preserving result order.
Playbooks · Agent workflows
Join parallel specialist branches before producing a final answer
Wait for all parallel branches, collect outputs by branch name, and pass them to a single synthesis node.
Playbooks · Agent workflows
Fan out agent subtasks from items discovered at runtime
Dynamically schedule child nodes when task count depends on data discovered during an agent run, then collect results.
Playbooks · Agent reliability
Retry workflow nodes only for selected transient exceptions
Set attempts, backoff, and an exception allowlist for one unreliable node; do not retry business denials.
Playbooks · Agent evaluation
Build agent regression evaluation from recorded conversations and tool traces
Use representative, redacted conversations to define expected answers and tool trajectories and detect regressions in pytest.
Playbooks · Agent evaluation
Measure tokens, tool calls, and latency during agent evaluation
Record agent resource use alongside quality evaluation to detect excessive cost for marginal quality gains.
Playbooks · Agent reliability
Stop an agent run when the user cancels or the client disconnects
Pass a client cancellation signal to the Runner so long tools or workflow nodes stop and the event stream closes cleanly.
Playbooks · Agent safety
Verify stored tool calls were not tampered with before resuming an agent run
When a paused tool call will execute on resume, validate its stored arguments and block tampered calls.
Playbooks · Agent authorization
Connect an agent to a third-party data tool with user authorization
Pause a run when a tool needs the user’s credential, complete the external consent flow, and resume the waiting call.
Playbooks · Agent interoperability
Expose an ADK agent as one MCP tool for other clients
Wrap an agent and its tools behind an MCP server so compatible clients can delegate through one bounded entry point.
Playbooks · Agent interoperability
Configure interceptors and response conversion for a remote A2A agent
Control outbound requests, required auth headers, and conversion of remote replies to local events when calling a remote agent.
Playbooks · Agent observability
Use event history to trace when agent state changed
Use ordered session events and state deltas to explain agent state transitions instead of guessing from prompts.
Playbooks · Agent safety
Intercept or redact content in a live bidirectional agent session
Inspect content before it reaches a live model stream and block or replace the turn when needed.
Playbooks · Agent workflows
Pause an agent workflow to ask for missing user information
Emit a structured input request when a workflow lacks a decision or field, then continue after the user responds.
Playbooks · Agent interoperability
Limit context and tool scope visible to a remote agent
Define fields and delegation scope before calling a remote agent and validate its response locally.
Playbooks · Agent workflows
Route an agent workflow through a safe branch based on validation
Encode explicit conditional routing in workflow edges so invalid inputs take a clarification or rejection path.
Playbooks · Agent runtime
Run a one-shot agent task with query
Run one task through the async message stream and distinguish text blocks from its final result.
Playbooks · Agent runtime
Build an interactive session with ClaudeSDKClient
Keep conversational context within one client, sending turns and consuming each full response.
Playbooks · Agent orchestration
Configure a least-privilege code-review subagent
Declare a focused reviewer with AgentDefinition, grant read tools only, and invoke it from the main task.
Playbooks · Agent orchestration
Verify that project-local agent definitions load
Read loaded agent names from the initialization message instead of assuming that a file on disk is active.
Playbooks · Agent safety
Distinguish tool auto-approval from tool availability
Verify that allowed_tools pre-approves listed tools and use disallowed_tools to block explicit exclusions.
Playbooks · Agent safety
Approve or rewrite tool inputs with a callback
Use can_use_tool to decide separately for reads, writes, and commands, and redirect writes to a safe directory.
Playbooks · Agent safety
Block risky commands with a PreToolUse hook
Match command input before tool execution and return deny through hookSpecificOutput.
Playbooks · Agent recovery
Feed tool errors back to the agent for recovery
Inspect tool responses in a PostToolUse hook and add actionable error context for the next model turn.
No matching entries.