How to use the Andru.ia Project Consultant agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent loads when a task matches the skill's description. This skill is meant for the very beginning of an AI project, or for a project you are taking over and want to improve.

Its behaviour depends on what it finds in the project folder. If the folder is empty, it runs an interview about what you are building, who it is for, and the result you expect. If code already exists, it scans the stack and possible technical debt, asks what to improve, and gives a short diagnosis before changing anything.

• Empty folder: asks what you will build, who it is for, and the desired result.
• Existing code: analyses the stack, architecture and possible technical debt, asks about pain points and quality goals, then gives a brief diagnosis.
• Proposes 3-5 existing skills by ID, declaring a gap instead of inventing an ID when none fits.
• Writes tareas.md (task backlog) and plan_implementacion.md (phased implementation plan), in Spanish by default.

Install the skill from the ZIP download

Download the ZIP for this skill from the page where you found this guide. Then choose one of the two installation locations below. The skill folder must contain SKILL.md.

For a personal install that applies to all your projects, unzip the download into ~/.claude/skills/. To limit the skill to a single project, unzip it into that project's .claude/skills/ folder instead.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal: ~/.claude/skills/
• Project only: .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills (upload)

How to trigger the skill with a natural request

You do not need a special command. Describe what you want at the start of a project, in plain language. The skill's description covers starting or improving an AI project, diagnosing code and constraints, proposing existing skills, and writing tasks and an implementation plan.

Say whether you want the diagnosis in the chat or saved as files. The skill only writes .md files when the task authorises saving them; otherwise it delivers the diagnosis in the conversation.

• "I'm starting a new AI app for nurses. Interview me before we write any code."
• "Diagnose this existing AI project and tell me what to improve first. Give me the diagnosis in the chat before changing files."
• "Create tareas.md and plan_implementacion.md for this project, in Spanish."

What to expect during a session

On an empty folder, expect questions about the goal, the audience and the desired result. On an existing project, expect a technical scan followed by three questions: what to improve, the biggest current limitation, and the quality standard you are aiming for. Answer what you can; the skill is designed to avoid repeating questions you have already answered.

After you reply, the skill checks the existing skill registry, reads the relevant skills, and proposes 3-5 of them. Proposing a skill does not create agents or authorise delegation. Finally it re-reads its diagnosis and files to check they match your goals and constraints, and separates verified observations from hypotheses. A generated file does not prove the implementation works.

Example of an existing-project diagnosis

Suppose you ask for a diagnosis of an existing AI project and want the result in the conversation first. A typical answer, based on the skill's own example, classifies the work as evolution, notes that an interface and API exist but there are no tests for the login flow, and proposes reproducing the login failure, adding a test for the expected behaviour, and reviewing access limits.

The skill would read the registry entries for any skills it recommends before naming them, and it would leave the choice of stack open until it knows your main constraint. Existing plans are not overwritten.

Safety notes: scripts and network

This skill does not run scripts and does not use the network, according to its listing. It requires nothing beyond the skill itself. It works by reading your project files and writing planning documents when you ask it to.

It is not a security or performance audit. The skill states it does not certify security or performance from reading a project, and it does not promise a universally optimal architecture. Treat its findings as hypotheses to test.

• Runs scripts: no
• Uses network: no
• Reads the project folder it is pointed at; writes .md files only when the task authorises saving them.
• Does not certify security or performance.

Tips for getting useful output

Start the skill in a clean project folder or in the real project you want to assess, so it reads the right files. Give constraints early, such as budget, hosting or language requirements, because the skill asks about constraints that would change its decision.

Read the planning files critically. Each task should have an observable completion criterion, and each phase should have a concrete check. If something is missing, ask the agent to name the gap rather than fill it with guesses.

• Give constraints at the start, not after the plan is written.
• Ask for the diagnosis in chat first if you want to review before files are created.
• Check that each phase in plan_implementacion.md has a concrete check.
• Ask for Spanish or another language explicitly if you need it; Spanish is the default.