Running multi-agent and delegated work
Orchestrator and subagent instructions
Orchestrator and subagent instructions
- Multi-Agent Systems: When One Agent Is Not Enough
- Deciding When a Subagent Instruction Change Needs a Human Gate
- Diffing Two Subagent Instructions by Behavior
- Reviewing a Subagent Instruction Like a Code Diff
- Subagent Prompts: What the Orchestrator Sends and What Comes Back
- A Diff Format for Orchestrator-to-Subagent Instructions
- A Subagent Instruction Test Suite: Catching Regressions
- Canary Rules for Subagent Instruction Changes
- All of Orchestrator and subagent instructions
Delegation budgets and depth
Delegation budgets and depth limits
- Delegation Depth Limits: How Deep Should Subagents Go?
- Depth and Spend Limits for Delegating Agents
- Depth vs Breadth: Splitting a Delegation Budget
- Budgeting a Delegation: Tokens, Time and Depth per Subagent Call
- Concurrency Limits for Parallel Subagents
- Depth Limits vs Timeouts: Which Cap Fires First
- Splitting a Token Budget Across a Fan-Out: Weights and a Floor
- Three Budgets per Subagent Call: Tokens, Time and Depth
- All of Delegation budgets and depth
Subagent return schemas
Subagent return schemas and error passing
- When a Subagent Returns Partial Work: Accept or Fail Loudly
- A Subagent Return Schema Built to Survive a Model Swap
- A Subagent Return Schema: A Reason Code and a Retry Hint
- A Validator That Branches on Coverage, Not Just Validity
- Accept, Retry or Escalate: A Decision Table for Partial Returns
- Reason Codes vs Free Text: Which the Model Picks Right
- Reason Codes vs Free-Text Errors: A Test Suite for Return Schemas
- Versioning an Agent-to-Agent Return Contract
- All of Subagent return schemas
Tracing multi-agent runs
Tracing and logging multi-agent runs
- A Trace for a Multi-Agent Run: What to Record Across Hops
- A Partial Return That Reaches the Customer: Reproducing From Logs
- A Trace Schema for Tool Calls: Fields That Make a Run Replayable
- A Trace Schema That Proves Depth Was Obeyed
- A Trace UI for Multi-Agent Runs: What a Debugger Needs
- Audit Log Schema for Delegated Tasks: Proving What Happened
- Joining a Trace to the Audit Log: Linking Runs to Outcomes
- OpenTelemetry for LLM Agents: Mapping Spans to Conventions
- All of Tracing multi-agent runs
Concurency and deadlines