Fireskills Quick Start Guide

Fireskills ships inside the Nuxfire repository. There is nothing to install: the skills (fireskills/skills/), the technical specialists (fireskills/specialists/), the CLI (fireskills/cli/) and the project context (.fireskills/) are already in the code you downloaded.

This page tells you what to type in your IDE agent for each thing you may want to do. Every step of the flow has its own page in this section; the table below links to them.

How you talk to Fireskills

You do not run the flow from a terminal. You type a prompt in your AI agent's chat (the agent panel of your IDE, opened on the repository root). The prompt names a skill with a leading $:

Use $fireskill-01-inbox para registrar este texto imediatamente, sem perguntas, em specs/inbox/.

• $fireskill-… is a text convention, not an IDE command. This repository does not register skills as slash commands: they live as plain files in fireskills/skills/<name>/SKILL.md. The agent resolves the name to that file.
• The repository ships AGENTS.md and CLAUDE.md (which imports AGENTS.md). An agent that reads either file is sent to fireskills/SKILL.md, which maps every request to the right skill and specialist. If your agent does not read those files, point it at the skill directly: Leia fireskills/skills/fireskill-01-inbox/SKILL.md e siga.
• The prompts and the agent's answers are in Brazilian Portuguese. The shipped prompts (fireskills/skills/<skill>/agents/openai.yaml) are Portuguese, and the contract in fireskills/skills/Spec.md requires every question, option and artifact to be in Português do Brasil. The prompts on these pages are quoted verbatim.
• After your prompt, add your own content on the next lines (your idea, the spec path, the diff). The shipped prompt is the starting text.
• Skills hand off to each other on their own ("Transição automática: $fireskill-02-backlog → $fireskill-03-specify …"). You keep answering in the same conversation; you do not retype commands.

Before your first prompt: see the state of the app

Two commands you can run yourself, from the repository root:

bun run fireskill progress
node fireskills/skills/fireskill-setup/scripts/monitor_context.mjs --project . --check

• progress lists every spec with its state, task counts and gates. On a fresh copy with no specs it prints Summary 0 specs 0/0 tasks 0/0 items 0%.
• The monitor compares your uncommitted and untracked files with the context files and prints Context monitor: CURRENT, or lists which .fireskills/* document needs an update. After you commit, it reports CURRENT again.

To understand what the application is and how it is built, read (or ask the agent to read) these, in this order: AGENTS.md → fireskills/SKILL.md → .fireskills/PROJECT.md → .fireskills/STACK.md → .fireskills/RULES.md. That is the same path a new agent follows, so a new engineer and an agent start with the same picture. See Project Context.

The first prompt (once)

Type this the first time, or whenever context looks stale:

Use $fireskill-setup para detectar o stack, consultar respostas confirmadas, adaptar as perguntas ao nível de conhecimento e monitorar pendências documentais.

Every other flow skill runs the setup for you before it starts, so you do not have to repeat it. Details: Step 0: Setup.

What do you want to do?

Lines ending in … continue with the sentence Orquestre transições automáticas com anúncio, execução e retomada na mesma conversa. in the shipped prompt (the full text is on each step page).

The flow at a glance

idea → inbox → backlog → spec → validate → tasks → TDD/BDD → implement → documentator → progress
        (1)      (2)      (3)      (4)       (5)      (6)        (7)          (8)

Three acts, three gates:

The skills move the spec between folders for you by running bun run fireskill transition <NNNN>-<slug> <folder>. The identifier is the spec number (0001) or the full directory name (0001-convite-de-membros-por-e-mail).

Rules you should know before you start

• One question per round, always numbered, always ending with Escrever outra resposta, Gere outras opções and Avançar. See Workflow & Governance.
• A spec lives in one file: specs/<state>/<NNNN>-<slug>/spec.md. There is no plan.md or tasks.md.
• The agent loads the specialist for the area (UI, tRPC, Drizzle…) before touching code. See Technical Specialists.
• Deploy, deleting data and other irreversible actions always need your explicit approval. bun run deploy is never run by the flow on its own.
• When code and documentation disagree, the code wins.