Tooling Landscape
This page is a mid-2026 snapshot, not a buyer's guide. The roster of agents, spec tools, and MCP servers changes faster than any book tracks well. Use the categories here to orient yourself, then verify current support against each tool's documentation.
Capability-class agents
This book targets coding agents that combine a thinking-capable model, real tool use, and a plan or architect mode. The vendor roster matters less than those capabilities, but examples make the boundary concrete. As of mid-2026, examples in scope include:
| Agent | Type | AGENTS.md support |
|---|---|---|
| Claude Code (Anthropic) | CLI | Native |
| Codex CLI (OpenAI) | CLI | Via AGENTS.md standard |
| OpenCode | CLI, open-source | Via AGENTS.md standard |
| Junie (JetBrains) | IDE-embedded | Via AGENTS.md standard |
| GitHub Copilot coding agent | IDE / PR | As of Aug 2025 |
Concrete model examples from this snapshot include Claude Sonnet 4.5 and GPT-5.2-class setups. Treat those as dated examples, not permanent cut lines. A newer or stronger model in the same capability class is still in scope.
IDE-only completion tools and chat-only assistants are out of scope. The practices in this book target the capability class, not a frozen vendor list. When a new tool combines planning, tool use, file edits, and reviewable output, evaluate it against the same patterns.
Sources: Anthropic Docs, "Claude Code overview" (ongoing), Claude Code capability class. Anthropic, "Introducing Claude Sonnet 4.5" (Sep 29, 2025), Claude Sonnet 4.5 as a dated capability example. OpenAI Docs, "Codex CLI" (ongoing), Codex CLI capability class. OpenAI, "Introducing GPT-5.2" (Dec 11, 2025), GPT-5.2 as a dated capability example. GitHub Changelog, "Copilot coding agent now supports AGENTS.md custom instructions" (Aug 28, 2025), Copilot AGENTS.md support. OpenCode Docs (ongoing), OpenCode capability class. Junie documentation (ongoing), Junie capability class. AGENTS.md support statuses are a mid-2026 snapshot.
Spec-driven tools
| Tool | Scope | Why it appears here |
|---|---|---|
| OpenSpec (openspec.dev) | Change-folder lifecycle (proposal, design, specs, tasks) | Used end-to-end in this book |
| LeanSpec (lean-spec.dev) | Lightweight spec | Small-spec discipline |
| GitHub Spec-Kit (github.com/github/spec-kit) | Enterprise toolchain | Full lifecycle, GitHub-integrated |
| GSD | Minimal ceremony | See Hightower's comparison for trade-offs |
No dominant framework has emerged as of mid-2026. ThoughtWorks Radar Vol 34 describes the SDD tool field as fragmented. Treat the table as a vocabulary map, not a recommendation list. Pick the lifecycle discipline your team will keep current under deadline pressure.
Sources: Hightower, "Agentic Coding: GSD vs Spec Kit vs OpenSpec vs Taskmaster AI" (Medium, Feb 2026), the tool-tradeoff map. ThoughtWorks, Technology Radar Vol 34 (April 2026), the fragmented SDD tool assessment.
MCP servers
The Model Context Protocol (modelcontextprotocol.io) lets agents reach external tools during a session. The durable pattern is external context fetched on demand. The server names below are perishable examples from mid-2026:
| Server | Purpose | Notes |
|---|---|---|
Atlassian Rovo MCP / mcp-atlassian | Jira and Confluence access | Fetches story context during spec drafting |
| Framelink MCP (GLips/Figma-Context-MCP) | Figma design data | For front-end specs, 14.9k stars as of mid-2026 |
MCP server availability, permission scope, and reliability vary by environment. Verify connectivity before relying on any external MCP server in a CI or automated context. What an agent reaches through MCP is determined by configuration and permissions, not by the tool existing.
Sources: Model Context Protocol (modelcontextprotocol.io, ongoing), MCP as the agent-tool bridge. Atlassian Rovo MCP Server and sooperset mcp-atlassian documentation (ongoing), Jira and Confluence connector examples. Framelink Figma-Context-MCP (github.com/GLips/Figma-Context-MCP, ongoing), Figma connector example and star count as of mid-2026.
Companion tool
iec is the companion CLI for this book. Its history is the evidence trail for the practices the book claims are demonstrated. See Companion Repo for phase tags and how to browse it.
Model selection and cost
This book targets capability class, not a specific model. Any agent combining a thinking-capable model, real tool use, and plan or architect mode falls in scope. These models are well suited to spec-driven development and architecture-heavy work because they hold longer contexts, plan across several steps, and handle codebase-spanning tasks better than earlier coding assistants. None of that means the agent follows ADRs by magic. ADR compliance still depends on whether the repo surfaces the decision, whether the agent loads it, and whether review and verification catch drift. For direct-API users, per-token cost matters: longer instruction files and specs have a real per-call cost, and the practices in this book add tokens deliberately to improve output quality. Some commercial tools hide direct per-run accounting behind seat licenses, but the engineering tradeoff stays the same: context has cost, and stale context has a different cost. DevOps, SRE, and cloud infrastructure costs are out of scope for this book.
Sources: OpenAI, "Introducing GPT-5.2" (Dec 11, 2025), stronger long-context and coding performance as a reason these models fit spec-driven and architecture-heavy work. Anthropic, "Introducing Claude Sonnet 4.5" (Sep 29, 2025), longer-horizon coding and agent work as a reason these models fit spec-driven and architecture-heavy work. Paula Hingel, "AI Agent Loop Token Costs: How to Constrain Context" (Augment Code, Apr 6, 2026), input-token cost growth in long agent loops. The model-selection boundary and ADR-compliance caveat are this book's capability-class synthesis.