Step 7: Implement

Skill: fireskill-07-implement · Asks questions: no (goes to $fireskill-update-spec if scope changes) · Changes: production code, tests, the checklists and evidence of spec.md, .fireskills/*

When to use it

The spec is in specs/planned/ with Plan Gate: Passed and failing tests exist. Type the prompt again for each next task, or let the agent keep going.

What to type

Use $fireskill-07-implement para executar o próximo item do checklist em specs/<estado>/<NNNN>-<slug>/spec.md com BDD, TDD e evidência. Orquestre transições automáticas com anúncio, execução e retomada na mesma conversa.

What the agent does

Before the first change

1. Requires Status: Planned or Implementing, Definition Gate: Passed and Plan Gate: Passed. If a gate or a RED is missing, it goes back to $fireskill-05-tasks and does not touch production.
2. If there is a screen, it checks that the plan has tasks for screens, menus, forms, actions and @nuxt/ui components, and it loads $fireskill-specialist-nuxt-ui before writing any .vue with U* components.
3. Checks the base test command with check_database_safety.mjs; without a separate test database it runs no test that touches the database.
4. Sets Status: Implementing, Delivery Gate: In Progress and runs bun run fireskill transition <id> in-progress.
5. Picks the next ready task:
   node fireskills/skills/fireskill-07-implement/scripts/next_task.mjs specs/in-progress/<NNNN>-<slug>/spec.md
   

For each task (checked off one item at a time, never in a batch at the end)

1. PREP: confirms ID, type, references, dependencies, files.
2. Loads the specialist of the area before touching code: nuxt-ui for .vue, nuxt for pages and middleware, trpc for routers, drizzle for schema and migrations, nuxt-authorization for permissions, openauth for login, cloudflare-workers for Workers, sst for infra/. The table is in fireskills/SKILL.md.
3. EXECUTE: the smallest change that turns the TDD test green. A [CODE] task without its RED predecessor stops. For [MIGRATION], it creates the versioned file first, applies it to the test database and queries the state.
4. Runs the context monitor; PENDING blocks the task. Reviews .fireskills/PROJECT.md on any application change.
5. After each [CODE] task, hands off to $fireskill-documentator (see Step 8).
6. VERIFY: focal test, related suite and static checks, only after the command is SAFE.
7. VISUAL: for a screen, compares the rendered result with .fireskills/DESIGNSYSTEM.MD and .fireskills/INTERFACE.md: borders, spacing, margins, padding, typography, alignment, overflow, focus, short and long content, in the relevant viewports. Without a screen it writes Não aplicável and the concrete reason.
8. EVIDENCE: records command, result and IDs in sections 11–13 (and the specsfy:evidence JSON comment when Evidence Contract: 1, checked by verify_evidence.mjs).
9. IMPROVE: one safe improvement, or "nenhuma melhoria necessária" with the reason.

If scope changes (behavior not in the spec, contradiction of an AC, public interface, data or security): it stops, announces, and loads $fireskill-update-spec. See Late changes.

The completion gate

When every task in section 14 is checked, the agent runs, and you can run the same:

# tests of the whole repository (from the repository root)
bun run test

# types (from apps/app)
cd apps/app
bun run typecheck
bunx vue-tsc --noEmit -p server/tsconfig.json
cd ../..

# copy of the context and documentation checks
node fireskills/skills/fireskill-setup/scripts/monitor_context.mjs --project . --check
node fireskills/skills/fireskill-documentator/scripts/build_documentation.mjs --project . --check

bun run typecheck does not exist at the repository root: it is a script of apps/app.

Then it compares every AC, FR, NFR and Definition of Done item with current evidence, looks for open tasks, placeholders and skipped tests, and only then:

1. Delivery Gate: Passed, Status: Reviewing, and bun run fireskill transition <id> review.
2. $fireskill-04-validate does the final acceptance and moves the spec to specs/completed/.
3. $fireskill-progress shows the global picture.

To get a summary you can paste in a PR (it only prints, it writes no file):

node fireskills/skills/fireskill-07-implement/scripts/render_delivery.mjs specs/review/<NNNN>-<slug>/spec.md --format markdown

Things it will not do alone

• bun run deploy. For changes in apps/functions or layers, a real sst deploy is the check that matters, and it needs your explicit authorization.
• Run a test against the development database, or any destructive database command.

Next

Step 8: Documentator (runs automatically), then progress with bun run fireskill progress.