Capiva Blog
Articles and tutorials by Capiva on the work we do every day: building the harness around AI agents, configuring Claude Code, getting a company cited by ChatGPT and other AI assistants, and structuring AI adoption inside a company.
23 articles, newest first within each topic.
AI harness, Claude Code and AI agents
- Claude Code Hooks: How to Configure Them, With 3 Ready Examples (PreToolUse, PostToolUse and Stop)
How to configure Claude Code hooks, with three tested examples: a PreToolUse guard, a PostToolUse sensor and a Stop gate.
Published on
- CLAUDE.md: What It Is and How to Write One the Agent Follows
What CLAUDE.md is, how to write one the agent follows, its ideal size, and CLAUDE.md vs AGENTS.md.
Published on
- How to Build an MCP Server in Python and Configure It in Claude Code
Build an MCP server in Python, test it, and configure it in Claude Code with claude mcp add, .mcp.json and scopes.
Published on
- How to Create a Skill in Claude Code: SKILL.md Step by Step
How to create a Claude Code skill: SKILL.md step by step, with a script and arguments.
Published on
- How to Install Claude Code on Windows: PowerShell, CMD, WinGet or WSL
How to install Claude Code on Windows with PowerShell, CMD, WinGet or WSL, and how to fix the most common errors.
Published on
- Guardrails and Evals for AI Agents: What They Are and How to Know the Agent Got It Right
What guardrails and evals for AI agents are, and how to check an agent works before production.
Published on
- What Is Harness Engineering? The Discipline That Decides If an Agent Works in Production
Harness engineering is the discipline of designing everything around the model (rules, tools, memory, and verification) so an agent works in production. Named by Mitchell Hashimoto in 2026.
Published on
- Harness, Context Engineering, and RAG: What's the Difference and How They Fit
Context engineering is one layer of the harness, not a synonym, and RAG is one technique inside context engineering. Understand the containment: RAG inside context engineering, inside the harness.
Published on
- Examples of AI Harness: 5 Real Harnesses, From the Terminal to the Autonomous Engineer
Real examples of an AI harness: Claude Code, Cursor, Devin, LangChain's deepagents-cli, and capivaOS, a harness engineered on top of Claude Code.
Published on
- How to Build an AI Harness: The Practical Recipe, From Simple to Systematized
To build an AI harness, start simple and grow it from observed failures: each agent error becomes a permanent rule, then systematized into guides and sensors.
Published on
- Claude Code Is a Harness: CLAUDE.md, Hooks, MCP, and Subagents as Guides and Sensors
Claude Code is a harness: CLAUDE.md, hooks, MCP, subagents, and permissions are the guides and sensors around the model, and capivaOS is a harness engineered on top of it.
Published on
- AI Harness Benchmark: Why the Same Model Scores Differently
In agent benchmarks, the harness moves the score as much as the model. There are two senses: agent harness (the scaffold, which must be disclosed) and eval harness (the runner, which must be held constant).
Published on
- The Operational Harness: The AI Beyond Code Almost No One Has Built
The operational harness is harness engineering applied to running an entire business, not just code. Almost no one has built this layer yet.
Published on
- What is an AI Harness? Definition, components, and why it matters more than the model
An AI harness is the infrastructure around an LLM (tools, rules, memory and quality gates) that turns it into a reliable agent. Agent = Model + Harness.
Published on
- AI Harness in Practice: running a company on agents
Capiva built an operational AI Harness and open-sourced its development layer as capivaOS. What harness engineering is and how it works in practice.
Published on
GEO and AI citation
- llms.txt: What It Is and How to Create, Generate and Validate It (Next.js Example)
What llms.txt is and how to create one: the llmstxt.org v2 format, a Python generator and validator, and publishing in Next.js.
Published on
- How to Measure Whether ChatGPT Cites Your Company: A Python Script With the OpenAI Search API
A Python script that uses the OpenAI web search API to measure whether ChatGPT cites your company.
Published on
- What Is GEO / SEO for AI (and How It Works)
GEO (Generative Engine Optimization), AEO, and "SEO for AI" are the discipline of optimizing content to be cited by AI answers — distinct from classic SEO, which ranks links. The concept (Princeton, 2023), the difference between the terms, and what the AI rewards.
Published on
- How to Appear in ChatGPT: What Makes the AI Recommend Your Company
You don't register your company with ChatGPT — the AI recommends whoever has a citable, structured, authoritative presence. The three real levers (citability, authority, technical access) and how to test.
Published on
AI consulting and adoption
- AI Consulting for Businesses: What It Is, What It Covers, and How to Choose the Right One
AI consulting for businesses covers diagnosis, proof of concept, implementation, and ongoing governance. What each track includes, how to evaluate a vendor before hiring, and what separates a good engagement from a bad one.
Published on
- How Much Does AI Consulting Cost? What Determines the Price (and How to Ask for a Real Quote)
There is no fixed price table for AI consulting. The value depends on the billing model (hourly, project, retainer, scope), the seniority of the team, and the actual scope of the problem. The factors and the right questions to ask before accepting a quote.
Published on
- What Is an AI Center of Excellence? The Structure That Decides If AI Becomes Capability or Stays in Pilot
An AI Center of Excellence is the internal structure that drives AI adoption to valuable outcomes and prevents fragmented or ungoverned adoption (Microsoft CAF). Most fail by becoming bottlenecks.
Published on
- Claude Impact Lab Rio: Auditable AI for Public Health and Public Safety
The first Brazilian edition of the Claude Impact Lab brought teams together in Rio to build AI solutions for public health and public safety.
Published on
Guides, essays and tools live in the resource library. To apply this in your company, see how we work.