Agent Workflow Patterns
These pages cover workflow patterns for Claude Code and similar coding agents.
They are not a replacement for tool-specific documentation. Use them as operating patterns: how to give an agent enough context, a safe enough action surface, and a clear enough verification target.
In This Series
Section titled “In This Series”- Codebase Onboarding with Coding Agents
- Finding Unknowns Before Agentic Work
- Spec-Driven Development with Coding Agents
- Plan Before You Code
- Verification-Driven Agentic Coding
- Capability Overhang and Tools
- Teaching Agents Your Local Tools
- Prompting Long-Horizon Coding Agents
- Parallel Coding Agents for Large Refactors
- Claude Code as a Unix-Style Tool
- Understanding Coding-Agent Benchmarks
- Agent Identity and Delegated Access
- Durable AI Agents with Temporal
What Makes These Workflows Useful
Section titled “What Makes These Workflows Useful”Most coding-agent failures are not caused by missing syntax knowledge. They usually come from:
- weak task scope
- too little repository context
- no runnable verification signal
- vague permissions around what the agent may do
- storing repeatable knowledge in the wrong place
The patterns in this section are meant to reduce those failure modes.
What This Section Does Not Cover
Section titled “What This Section Does Not Cover”The basics are already covered elsewhere:
- Claude Code
- Gemini CLI
- Skills Overview
- Skills with MCP, Tools, and Subagents
- AI Agents: Concepts, Patterns, and Frameworks
For commands and flags, always verify against the current tool documentation or run /help in the installed CLI.