Agentic Coding

A curated list of 130+ CLI coding agents and orchestration tools for AI-powered terminal workflows. The list covers terminal-native agents (open source like Hermes Agent, OpenCode, and Claw Code; closed source like Claude Code and Warp), orchestration platforms for running multiple agents in parallel (Orca, Multica, Paseo), and infrastructure tools for context management, sandboxing, and agent coordination. Key techniques highlighted include context compression via Headroom (reducing tokens 15–95%), persistent memory systems for maintaining project context across sessions, MCP (Model Context Protocol) integration for tool extensibility, skills and AGENTS.md for encoding best practices, and workflow frameworks for autonomous multi-agent loops with verification gates. For solo founders working with Claude and Cursor daily, relevant best practices include session managers that isolate work in git worktrees, durable memory layers to avoid context reset, skill optimization to reduce repeated prompting, and approval gates before risky operations—all implemented in projects like Omnigent, Concord MCP, and Harness Starter Kit.

Claude Opus 5.5 now features a 1M token context window, enabling developers to load entire codebases for comprehensive development sessions with minimal optimization overhead. Strategic model selection—reserving Opus for architecture planning and complex analysis while using Sonnet 5 for execution and routine edits—maximizes allocation efficiency. Key context management techniques include using the /compact command to reduce context size, batching related changes into focused sessions, and avoiding the final 20% of context windows for complex tasks due to performance degradation. Development workflows are optimized by timing intensive work around 5-hour reset cycles, breaking complex features into manageable phases, and leveraging built-in tools like /context (v1.0.86+) for debugging context issues. For agentic coding, Claude Code's max-turns parameter in non-interactive mode controls autonomous actions, with guidelines suggesting 40-200+ turns for sub-agent workflows. The Max 5x plan ($100/month) supports all-day development with generous Opus allocation, while the Max 20x plan ($200/month) provides unrestricted access suitable for parallel projects and team development scenarios.

When working with large codebases and AI coding assistants like Claude, the key challenge is determining how much context to provide. Rather than including all files, developers should focus on providing a dependency chain specific to the task: if modifying a feature, include the relevant Angular component, service, API layer, backend controller, business logic, and database interactions. Excessive context can reduce response quality, so a balanced approach using techniques like RAG, code indexing, or repository maps to retrieve only relevant dependencies is more effective than comprehensive context dumps. For enterprise applications, architectural understanding matters as much as code generation, making strategic context selection and dependency verification critical practices.