Build with Claude
For engineers, integrators, and anyone shipping Claude into production.
Practitioner content for engineers shipping Claude into production. Claude Code workflows, Claude API patterns, MCP integrations, agentic-development verification loops, and the patterns that separate prototypes from systems you'd let run unsupervised. If you're integrating, this is the track that's writing for you.
Posts in the Build with Claude track
The Lethal Trifecta Has Three Legs. Least Privilege Covers One.
Least privilege shrinks your AI agent blast radius but leaves the egress and untrusted-input paths ungoverned. Here is how to close both gaps in production.
Deep diveMCP Authorization Scope: Least Privilege Before You Connect
MCP server permissions are an architecture decision, not connector plumbing. Scope your Jira, GitHub, and Confluence access before you connect, not after the incident.
Deep diveWhen to Trust an LLM Judge (and When to Pull It Off the Gate)
Uncalibrated LLM-as-a-judge verdicts share the generator's blind spots. A practitioner's guide to judge failure modes, calibration, and when a judge earns gate authority.
Deep diveThe Approval Budget: A Model for Spending Human Attention in Agentic Workflows
Approval fatigue in agentic AI turns every permission prompt into a rubber stamp. The fix is a budget: spend human attention where judgment is irreplaceable and push the rest into deterministic allow and deny layers.
Deep diveClaude Code Subagents vs Skills: One Teaches the Session, the Other Staffs It
Claude Code subagents vs skills is a context architecture decision, not a capability one. Learn when to define a .claude/agents file vs. a SKILL.md.
Deep diveThe Changelog Says 5 Levels. My Probe Went 9 Deep. Inside Claude Code's Nested Subagents.
Claude Code subagents can now spawn their own subagents. Here's what the depth limit actually does, what every delegation level costs, and the few trees worth growing.
Deep diveClaude Fable 5's Silent Degradation: The Safety Tier You Couldn't See, Log, or Turn Off
Claude Fable 5 shipped a silent-degradation tier with no stop_reason, no notice, and no opt-out, then reversed it in 48 hours. A field map of the three safety tiers and what to instrument now.
Deep diveClaude Fable 5 Is 'Mostly Drop-In.' The Word Doing the Work Is 'Mostly.'
Migrating from Opus 4.8 to Claude Fable 5? Adaptive thinking costs, a new refusal API, and ZDR incompatibility mean it's not just a model-name swap.
Visual guideClaude Code Hooks, Visualized: The Lifecycle Map, the Deny Gate, and the Enforcement Ladder
Claude Code hooks, visualized: the session lifecycle map, the PreToolUse allow/ask/deny gate, payload anatomy, and the enforcement ladder vs CLAUDE.md.
Deep diveI Ran Seven Claude Code Sessions on One Repo. Three Spiraled. Git Worktrees Were Only Half the Fix.
Running parallel Claude Code sessions safely requires git worktrees, not just extra tabs. Here's the structural isolation model every practitioner needs.
Opus 4.8 vs 4.7, One Week Later: The Upgrade Call I Couldn't Make on Day One
Opus 4.8 launched as the honesty release, then misreported its own tools in launch week. A week on: the changelog fix, whether the honesty held, and the 4.7 upgrade call.
Deep diveHow to Control Claude Code Token Spend Without Sacrificing Output
Claude Code token optimization is a quality problem, not just a cost one. The structural levers that govern both: model and effort, the shape of your context, and how wide you fan out.
Deep diveHow to Instrument Observability for Production Claude Agents
Agent View shipped this spring atop older CLI telemetry, and together they cover a lot. Here is the slice you still instrument yourself, and the principle that outlasts any dashboard.
Deep diveClaude Opus 4.8 in Claude Code: I Couldn't Trust What It Said About Its Own Tools.
Opus 4.8 is the honesty release, but in my Claude Code loops I could not trust what it reported about its own tool calls. The measurement, the mechanism, and how to catch it.
Deep diveI Read the Files Claude Code Wrote About Me: How Auto-Memory Actually Works
Claude Code auto-memory writes Markdown files about your project. Where they live, what triggers a save, and how to read, steer, and govern what Claude writes.
Deep diveThe Most Important Number in Claude Opus 4.8 Isn't a Benchmark
Claude Opus 4.8's headline improvement is honesty, not raw capability. Why a model that flags its own uncertainty changes the economics of autonomous agents.
Deep diveClaude Code Can Run a Thousand Subagents. The Verification Contract Is the Part You Design First.
Claude Code dynamic workflows fan out up to 1,000 subagents. At that scale no human reads every diff, so the verification contract is what you design first.
Deep diveClaude Code in GitHub Actions: A Containment Checklist for CI Agents
Running Claude Code in GitHub Actions is a containment problem first. A production checklist for secrets, isolation, permission scoping, and merge ownership.
Deep diveBuild Your First MCP Server: The Hard Part Isn't the Code
Build your first MCP server step by step. Purpose-built beats a stretched connector, and tool-description design is where it counts.
Visual guideWhen to Orchestrate Claude Code Subagents: A Four-Gate Decision Map
A visual decision guide to Claude Code subagent orchestration: the scope-of-work test, context blast radius, and write-safety gates that decide whether to split a task and which pattern to use.
Deep diveClaude Code /advisor: When to Call It and Where It's Blind
Claude Code's /advisor sends your full conversation to Opus mid-task. After 159 calls in 17 days, here's the rule that beats the official guidance.
Deep diveClaude Code Can Auto-Fix Your Pull Requests Unattended. Who Owns the Review Bar Now?
Claude Code auto-fix watches your PR, fixes CI failures, and responds to review comments overnight. Here's what that means for your merge gate.
Deep diveHow to Audit an MCP Server Before Its Upstream API Breaks It
MCP servers are a maintained production surface, not one-time plumbing. A guide to security audits, token governance, and surviving upstream API deprecation.
Deep diveWhat Does 'Critical' Actually Mean in an AI Code Review?
Review automation treats Claude's severity labels as absolutes. They're local ranks inside one context. A deep-dive on AI code review severity calibration and what to do about it.
The AI Diligence Operating System: Schema First, Subagents Second
Before you hire a Claude architect for diligence, settle the AI diligence operating system question: memo schema first, subagents second. Here's why.
Deep diveClaude Code iPhone: The Practitioner Playbook for Shipping Without a Laptop
Claude Code on iPhone runs production-grade personal projects, if you stop fighting the workflow. A practitioner playbook for shipping without a laptop.
Deep diveWhen to Reach for Claude Code's /batch (and When claude-squad, Agent Teams, or Superpowers Wins)
A reference for senior engineers deciding whether to adopt Claude Code's /batch command on real work. Mechanics, suitability criteria, anti-patterns, comparison against claude-squad, Agent Teams, the Anthropic Agent SDK, and Superpowers v5, plus the failure modes the launch announcement skipped.
Deep dive90 Minutes With 400 Engineers, PMs, and Ops on Claude Code: Here's What They Wanted to Know
Six questions kept recycling when I trained 400+ people on Claude Code. The taxonomy reveals more about rollout failure than the answers do.
Deep diveHow to Create a Claude or Claude Code Skill: A Hands-On Guide for First-Timers and Optimizers
Skills feel like a mini-program in a new tech stack. They aren't. Here is how to create your first Claude or Claude Code skill, three creation paths from easiest to most rigorous, and how to make the same SKILL.md run in Codex, Cursor, and beyond.
Deep diveThe free Claude Code skill that audits your CLAUDE.md, hooks, and subagents for Opus 4.7 breaking changes
Your 4.6 CLAUDE.md looks fine. Opus 4.7's literal reading means it isn't. Free compatibility scanner audits hooks, settings.json, and subagents in minutes.
Deep diveWhere Does That Rule Go? A Decision Tree for CLAUDE.md, Settings, Skills, and Hooks in Claude Code
Stop overloading CLAUDE.md. Learn when Claude Code directives belong in settings, rules, skills, or hooks. Includes a decision tree and worked examples.
Deep diveClaude vs Claude Code: How to Pick the Right One (and When to Use Both)
Claude vs Claude Code: understand the core differences, when each tool is the right choice, and how the most effective users run both together.
Deep diveThe Four Sub-Agent Orchestration Patterns That Cover 90% of Production Claude Workloads
Anthropic ships no canonical multi-agent reference architecture. Here are four Claude sub-agent orchestration patterns with production decision criteria, cost math, and failure modes.
Claude Opus 4.7 Is a Split, Not an Upgrade
Opus 4.7 vs Sonnet 4.6: Anthropic just split its lineup into agent operators and fast-turn assistants. Route them interchangeably and you'll pay for the mismatch.
Deep diveBeyond the Wrapper: Five Claude API Patterns That Separate Prototypes from Production
Most Claude API integrations are wrappers around a string function. Five architectural patterns separate prototypes from production: here's how to ship them.
Your Claude API Integration Is Probably a Wrapper Around a String Function
Most teams treat the Claude API as stateless text-in/text-out. The integration mistakes that follow cost money, add latency, and leave Claude's strongest features unused.
Deep diveThe Agentic Development Starter Guide: Plan, Audit, Implement, Verify
A practitioner's guide to agentic coding workflows. Learn the plan-audit-implement-verify cycle that separates disciplined AI development from vibe coding.
Connecting Claude to Jira, GitHub, and Confluence: An MCP Server Setup Guide
Connect Claude to Jira, GitHub, and Confluence with MCP servers. The full three-tool setup, a cross-tool workflow demo, and the credential security other guides skip.
Your PRs Are Missing Context. Claude Code Fixes That.
A third of pull requests ship with no description at all. Claude Code generates rich, structured PR descriptions from your codebase and commit history, giving reviewers everything they need without extra effort from you.
Claude Code Custom Skills: Encoding Your Team's Standards Into AI Workflows
Your team's coding standards live in a wiki nobody reads. Claude Code custom skills turn those standards into instructions Claude follows on every task.
Quick readWhy Claude Code Changes How SaaS Teams Ship
Most teams use Claude as a chatbot. The productivity lift comes from Claude Code workflows that plug into your CI/CD pipeline, Jira board, and code review process.