# Finalizer

> Use when the user asks to run the Finalizer workflow, execute the finalizer skill, or finalize a branch through the finalizer subagent. Gathers the exact checkout context and then invokes the `finalizer` subagent correctly.

- Skill: `bnema/finalizer` (Agent Skill)
- Install (CLI): `npx skillmds@latest add bnema/finalizer`
- Raw SKILL.md: https://api.skillmd.com/api/skills/bnema/finalizer/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: bnema (https://skillmd.com/u/bnema)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/bnema/finalizer

---


# Finalizer Skill

**Required baseline:** finalization must include the thermo-nuclear code-quality standard from `code-review-release/subskills/thermo-nuclear-code-quality-review/SKILL.md` as the first-priority maintainability bar. The finalizer's specialized subagents add lenses; they do not replace the requirement to catch structural regressions, missed code-judo simplifications, file-size blowups, spaghetti branching, weak boundaries/types, canonical-helper duplication, and unnecessary abstraction/cast churn.

This skill is a wrapper around the `finalizer` subagent.

Use it when the user asks for any of these:
- run finalizer
- execute the finalizer skill
- finalize this branch
- do the finalizer pass

## Goal

Avoid the two common failure modes:
1. treating Finalizer like a skillless concept search instead of a subagent workflow
2. invoking `finalizer` without the exact checkout context it requires

## Required procedure

### 1. Resolve the exact checkout context first

Before invoking the `finalizer` subagent, determine all four values it requires:

- exact working directory to use
- repo root for that working directory
- current branch for that working directory
- base branch to compare against

Use focused git commands in the intended repo, for example:

```bash
git branch --show-current
git rev-parse --show-toplevel
git merge-base --fork-point origin/main HEAD 2>/dev/null
git merge-base --fork-point main HEAD 2>/dev/null
```

Prefer the repo path explicitly given by the user. If the user did not provide one and the repo is ambiguous, ask.

### 2. Invoke the `finalizer` subagent explicitly

Do **not** output plain text like `subagent finalizer`.
Do **not** search the filesystem for a SKILL named Finalizer once this skill is loaded.

Call the `subagent` tool in SINGLE mode with:
- `agent: "finalizer"`
- `agentScope: "user"`
- `cwd`: the exact working directory
- `task`: a prompt that includes all four required values verbatim

### 3. Task template to send to the subagent

Use this structure in the `task` field:

```text
Run the full finalization pass for this exact checkout.

Required review baseline:
- apply code-review-release/subskills/thermo-nuclear-code-quality-review/SKILL.md first for all review judgments
- treat structural simplification/code-judo opportunities, file-size blowups, spaghetti branching, boundary/type cleanliness, canonical helper reuse, and unnecessary abstraction/cast churn as first-class finalization concerns

Required context:
- exact working directory to use: <WORKDIR>
- repo root for that working directory: <REPO_ROOT>
- current branch for that working directory: <CURRENT_BRANCH>
- base branch to compare against: <BASE_BRANCH>

Stay pinned to that exact checkout.
```

### 4. Let Finalizer orchestrate the rest

The `finalizer` subagent is responsible for launching the required child subagents.
Its configuration expects parallel `general` subagents with these explicit skills:
- `thermo-nuclear-code-quality-review`
- `hexarch`
- `simplify`
- `coderabbit`
- `vibesec`

Do not replace that with manual review unless the user asks.

## Success criteria

A correct run means:
- the `finalizer` subagent is invoked through the tool, not as free text
- all four checkout values are passed in the task
- the finalizer can then spawn the required subagents automatically