Learning Path
s01 to s12: Progressive Agent Design
Layer Legend
The Agent LoopBash is All You Need
“The minimal agent kernel is a while loop + one tool”
Learn MoreToolsOne Handler Per Tool
“The loop stays the same; new tools register into the dispatch map”
Learn MoreTodoWritePlan Before You Act
“An agent without a plan drifts; list the steps first, then execute”
Learn MoreSubagentsClean Context Per Subtask
“Subagents use independent messages[], keeping the main conversation clean”
Learn MoreSkillsLoad on Demand
“Inject knowledge via tool_result when needed, not upfront in the system prompt”
Learn MoreCompactThree-Layer Compression
“Context will fill up; three-layer compression strategy enables infinite sessions”
Learn MoreTasksTask Graph + Dependencies
“A file-based task graph with ordering, parallelism, and dependencies -- the coordination backbone for multi-agent work”
Learn MoreBackground TasksBackground Threads + Notifications
“Run slow operations in the background; the agent keeps thinking ahead”
Learn MoreAgent TeamsTeammates + Mailboxes
“When one agent can't finish, delegate to persistent teammates via async mailboxes”
Learn MoreTeam ProtocolsShared Communication Rules
“One request-response pattern drives all team negotiation”
Learn MoreAutonomous AgentsScan Board, Claim Tasks
“Teammates scan the board and claim tasks themselves; no need for the lead to assign each one”
Learn MoreWorktree + Task IsolationIsolate by Directory
“Each works in its own directory; tasks manage goals, worktrees manage directories, bound by ID”
Learn MoreAgent EvalsTDAD + LLM-as-Judge
“An agent that can't verify its own work is just guessing; verifiable output is the key to success”
Learn MoreWorkflow PatternsFive Orchestration Strategies
“Don't build agents — build workflow patterns; start simple, add complexity only when needed”
Learn MoreContext EngineeringBudget, Not Compress
“Context Engineering > Prompt Engineering; proactive budget allocation, not reactive compression”
Learn MoreLong-Running HarnessGAN-Style Three Agents
“Separate the generator from the evaluator; fresh context per iteration solves context anxiety”
Learn MoreMCPStandard Tool Protocol
“MCP is USB for AI tools — standardized discovery and invocation across any service”
Learn MoreAuto ModePermission Tiers + Sandbox
“Safety should not be the enemy of efficiency; 95% auto-approve, 5% human oversight”
Learn MoreThink ToolA Tool That Does Nothing
“The simplest tool can be the most useful; giving the model permission to pause improves quality”
Learn MoreParallel TeamsFile-Lock Task Board
“The key to scaling is decoupling; task board + file locks = linear scalability”
Learn MoreTool DesignACI Principles
“Good tools are hard to misuse; tool design impacts agent quality more than prompt design”
Learn More