What this app covers

Coordinator and subagent patterns, multi-agent topology, task decomposition, and session-state management. Covers how agents delegate work, route to specialists, recover from failures, and maintain coherent state across a pipeline.

• ↳Coordinator / subagent orchestration and delegation boundaries
• ↳Hub-and-spoke vs. peer-to-peer multi-agent topologies
• ↳Fixed vs. dynamic task decomposition and workflow selection
• ↳Session state, resumption, and pipeline fork recovery
• ↳Escalation routing and authority-level handling
• ↳Agent loop lifecycle: plan, tool call, observe, decide

Designing effective tool interfaces, writing parameter descriptions that guide correct model behavior, MCP error protocol layering, and controlling tool availability via allowedTools and tool_choice.

• ↳Tool interface design: parameter descriptions and boundary definitions
• ↳MCP structured error responses and isError semantics
• ↳Tool distribution: allowedTools, tool_choice, and subagent scope
• ↳Companion tool patterns and error recovery flows
• ↳Tool schema token budget and selection reliability trade-offs
• ↳Protocol layer boundaries between MCP and application error handling

Claude Code CLI configuration, CI/CD integration, CLAUDE.md authoring, slash commands, MCP server setup, and production-grade agentic workflow patterns for developer productivity.

• ↳CLAUDE.md structure and project-level configuration
• ↳Claude Code in CI/CD pipelines (--output-format json, non-interactive modes)
• ↳MCP server configuration and tool registration
• ↳Slash commands and workflow automation
• ↳Permission scope and trust levels in Claude Code
• ↳Personal vs. project skill precedence and override behavior

JSON schema design for tool use, structured extraction patterns, validation-and-retry loops, semantic vs. syntax error handling, and prompt-level normalization for consistent structured output.

• ↳Structured output via tool_use and JSON schema definitions
• ↳Validation, retry, and feedback loops for extraction quality
• ↳Semantic vs. syntax error classification and handling
• ↳Prompt-level normalization for consistent field extraction
• ↳Schema envelope integrity and data completeness contracts
• ↳tool_choice: auto / any / specific — when to use each

Preserving facts across long interactions, multi-turn drift detection, ephemeral state injection, escalation for ambiguity resolution, and reinforcement injection patterns that degrade output quality over time.

• ↳Context preservation across long multi-turn interactions
• ↳Three-cause drift framework: compression, recency bias, reinforcement injection
• ↳Differential compression and "lost in the middle" mitigation
• ↳Ephemeral state injection patterns and session scope
• ↳Escalation vs. auto-resolution in ambiguous customer support flows
• ↳Within-conversation contradiction detection and consistency maintenance