学習パス
s01からs20へ：段階的エージェント Harness 設計
レイヤー凡例
The Agent LoopOne Loop Is All You Need
“The smallest useful agent is a loop that calls the model, runs tools, and feeds results back.”
詳細を見るTool UseAdd a Tool, Add Just One Line
“The loop stays stable while capabilities register into a dispatch table.”
詳細を見るPermissionCheck Permissions Before Execution
“Dangerous actions need a harness decision point before the shell runs.”
詳細を見るHooksHang on the Loop, Don't Write into It
“Cross-cutting behavior belongs around the loop, not tangled inside it.”
詳細を見るTodoWriteAn Agent Without a Plan Drifts Off Course
“Explicit plans keep long-running work visible and correctable.”
詳細を見るSubagentBreak Large Tasks into Small Ones with Clean Context
“Subagents give each subtask a clean message history while preserving the main thread.”
詳細を見るSkill LoadingLoad Only When Needed
“Inject specialized knowledge only when the task actually needs it.”
詳細を見るContext CompactContext Will Fill Up
“Compression keeps the conversation usable when the context window gets crowded.”
詳細を見るMemoryKeep a Layer That Doesn't Lose Details
“Some facts should survive summarization and future sessions.”
詳細を見るSystem PromptAssembled at Runtime, Never Hardcoded
“The system prompt is a generated product of policy, tools, skills, and context.”
詳細を見るError RecoveryErrors Are the Start of a Retry
“A robust harness classifies failures and decides what kind of retry is worthwhile.”
詳細を見るTask SystemBreak Big Goals into Small Tasks
“A task graph turns vague goals into ordered, observable work.”
詳細を見るBackground TasksSlow Operations Go to the Background
“The agent can keep reasoning while slow work completes elsewhere.”
詳細を見るCron SchedulerProducing Work on a Schedule
“Recurring work should be created by the harness, not remembered by the model.”
詳細を見るAgent TeamsOne Agent Isn't Enough, Form a Team
“Persistent teammates let work continue in parallel without stuffing every thought into one context.”
詳細を見るTeam ProtocolsTeammates Need Agreements
“Multi-agent systems need explicit message contracts, not vibes.”
詳細を見るAutonomous AgentsCheck the Board, Claim the Task
“Teammates become useful when they can discover and claim work themselves.”
詳細を見るWorktree IsolationSeparate Directories, No Conflicts
“Parallel agents need isolated filesystems as much as isolated conversations.”
詳細を見るMCP ToolsExternal Tools, Standard Protocol
“External services can become agent tools through a standard discovery and call protocol.”
詳細を見るComprehensive AgentAll Mechanisms, One Loop
“The final harness is still one loop, now surrounded by the systems that make it production-shaped.”
詳細を見る