Finalize

Finalize

How it works

This skill runs a universal checklist, then checks the project's agent instructions file (CLAUDE.md, or AGENTS.md in harnesses that use it) for any project-specific finalization requirements (look for a "Finalize Checklist" section or similar). The project knows what it needs — this skill just makes sure nothing gets skipped.

Harness-neutral. The checklist is tool-agnostic: run the project's real lint/test/build commands with whatever shell and tooling the harness has. Steps that name a specific tool (e.g. the Impeccable UI pass, or the Memory step) apply only where that tool exists — skip them, with a note, where it doesn't. Nothing here requires subagents.

Universal Checklist

Work through each section. Report what passes and what needs fixing. Fix issues directly rather than just listing them — the user wants the work done, not a report.

1. Lint & Format

Run whatever lint/format tools the project uses. Check CLAUDE.md/AGENTS.md or package.json/Gemfile for the commands. Common ones:

• Ruby: standardrb, rubocop
• JS/TS: eslint, oxlint, prettier, oxfmt
• CSS: stylelint
• Python: ruff, black

2. Tests

Run the project's test suite. Check for:

• All tests pass
• No skipped tests that shouldn't be skipped
• New code has test coverage (or note if it doesn't)

3. Code Quality Sweep

Scan changed files for common issues:

• Raw/hardcoded values that should use project utilities or constants
• Inconsistent patterns with existing code (naming, structure, style)
• Dead code, redundant logic, TODO/FIXME comments left behind
• Debug statements (console.log, binding.pry, debugger, puts)
• Security: hardcoded secrets, unsafe patterns

4. Documentation

Check that new/changed code has corresponding documentation:

• README or docs pages updated if user-facing behavior changed
• Inline docs (JSDoc, YARD, docstrings) on public APIs
• Any project-specific docs (check CLAUDE.md/AGENTS.md for the list)

5. UI & Design Quality

Skip this section if the changeset has no view/template/component files.

Run the Impeccable skill on changed view files:

• $impeccable audit <target> — accessibility, performance, responsive behavior, theming
• $impeccable polish <target> — final UI quality pass: spacing, interaction states, typography, transitions, component-library fit, design-system compliance, tokens

Use the current Impeccable command routing rather than legacy standalone skills. normalize was folded into polish; do not try to invoke it separately.

Report which Impeccable commands were run and what they found. Fix issues directly.

Consistency with existing pages (not covered by Impeccable):

• New pages match the structure of similar existing pages
• Headers, content wrappers, and action placement follow established patterns
• If the page deviates from the dominant pattern, the deviation is intentional and justified

Visual verification:

• Changes render correctly in dev server, Storybook, or Lookbook
• Light mode and dark mode if applicable
• Mobile/responsive if applicable

6. Git & Issue Hygiene

• PR description reflects the actual scope of changes
• Closes #N or equivalent issue linking in PR body
• Parent issues/epics updated if this is part of a larger effort
• Follow-on issues created for deferred or discovered work
• Issues added to project board with correct status

7. Project-Specific Checks

Read the project's CLAUDE.md/AGENTS.md for a "Finalize Checklist" section (or similar). If it exists, work through those items too. These are project-specific requirements that the universal checklist can't know about.

8. Memory

If this is a Claude Code session with persistent memory:

• Update memory files with key learnings, decisions, or patterns discovered
• Update status tracking (component counts, phase progress, etc.)