Navegacao Rapida
3.1
~50 min
Skills System
Workflows reutilizaveis como slash commands: discovery, parsing, substituicao de argumentos, execucao inline/fork e integracao MCP.
1
⚡ Skill Lifecycle: Six Stages
Discovery, Load, Parse, Substitute, Execute, Inject
2
📁 Four Sources & Priority Order
managed -> user -> project -> bundled
3
📝 SKILL.md Format & Frontmatter
15+ fields: tools, model, effort, paths, hooks
4
🔄 Argument Substitution Pipeline
Named args, indexed args, shell injection, special vars
5
🔌 MCP Skills & Bundled Skills
Remote skills sem shell, /simplify, /commit, /remember
6
🔒 Permissions, Auto-Allow & Live Reloading
Safe skills auto-approved, chokidar watch, 300ms debounce
3.2
~60 min
Agent System
Delegacao de trabalho entre instancias Claude: tres tipos de agente, sync vs async, fork path com cache sharing e worktree isolation.
1
🤖 Three Agent Types & Priority
BuiltIn, Custom, Plugin -- policySettings wins
2
🏃 Built-In Agents Deep Dive
General Purpose, Explore, Plan, Verification, Fork
3
🔄 Sync vs Async Lifecycle
shouldRunAsync boolean, 6 conditions, detached lifecycle
4
🔀 The Fork Path & Cache Sharing
buildForkedMessages(), byte-identical prefix, recursive guard
5
🌳 Worktree Isolation
Git worktree temporario, smart cleanup
6
📨 SendMessageTool & Agent Frontmatter
Inter-agent messaging, 15+ frontmatter fields
3.3
~45 min
Coordinator Mode
Modo de operacao onde Claude atua exclusivamente como dispatcher: 4 tools, workflow de 4 fases e a arte de synthesis.
1
📋 Core Concept & Activation
CLAUDE_CODE_COORDINATOR_MODE=1, matchSessionMode()
2
🎯 Four-Phase Task Workflow
Research -> Synthesis -> Implementation -> Verification
3
🔄 Continue vs Spawn Fresh Decision
6 situacoes, context overlap como criterio
4
💬 Worker Prompt Guidelines
Self-contained prompts com file paths e line numbers
5
🛠️ Tool Restrictions & Worker Context
4 coordinator tools, ASYNC_AGENT_ALLOWED_TOOLS
6
📝 The Scratchpad
Cross-worker knowledge, permission-free shared directory
3.4
~50 min
Teams & Swarm
Equipes nomeadas de Claude agents com config compartilhada, messaging file-based, tres spawn backends e lifecycle completo.
1
👥 TeamCreateTool & File Structure
5 sync steps, ~/.claude/teams/ layout, agentName@teamName
2
💻 Three Spawn Backends
tmux, iTerm2, in-process -- detection at module load
3
📬 Mailbox Messaging System
File-based JSON, 6 message types, inbox per agent
4
🔒 Permission Sync
Worker escalation, leader approval/denial
5
🔄 Full Lifecycle: 4 Phases
Setup -> Parallel Work -> Shutdown -> Cleanup Guard
6
🔑 Key Design Invariants
Team = TaskList, idle != dead, all messaging file-based
3.5
~50 min
Memory System
Tres camadas de memoria persistente: Auto Memory, Session Memory e Team Memory com sync server-side e secret scanning.
1
🧠 Three Memory Layers
Auto Memory, Session Memory, Team Memory
2
📂 MEMORY.md & Topic Files
200 lines/25KB index, YAML frontmatter, 4 types
3
🔍 Extraction Pipeline & Relevance Recall
Post-query extraction, Sonnet selector, staleness signals
4
📝 Session Memory Structure
10 fixed sections, extraction triggers, 12K token budget
5
🌐 Team Memory Sync
Delta push with SHA-256, secret scanning, scope routing
6
⚙️ Path Resolution & Disable Mechanisms
3-tier resolution, security boundary, 4 disable flags
3.6
~50 min
Auto-Memory & Dreams
Dois processos background: extraction per-turn e consolidation cross-session (Dream). Recall seletivo via Sonnet e o lock file elegante.
1
📁 Memory Directory & Four Types
Resolution order, MEMORY.md index, topic file YAML
2
⚡ Layer 1: Extract Memories
Per-turn extraction, forked agent, mutual exclusion
3
💭 Layer 2: Auto Dream (Consolidation)
3 gates, 4 phases, lock file mtime trick
4
🔍 Recall: findRelevantMemories
Sonnet side-query, max 5 files, tool suppression
5
🌐 Team Memory Scope Routing
Private vs team, prompt-guided enforcement
6
🌌 KAIROS Mode & Cache Efficiency
Append-only daily logs, /dream skill, prompt cache sharing