Codex Plugin Marketplace

A repository-backed marketplace for reusable Codex plugins.

Available plugins

• Agent Fix (agent-fix) — fixes both style quality (CSS ownership and unused CSS) and unit test quality (ownership, coverage, and assertions) within a user-specified scope.
• Agent Review (agent-review) — reviews explicitly scoped code for correctness, reliability, security, performance, maintainability, architecture, dependencies, setup, configuration, comments, naming, language quality, and test infrastructure, or user-selected Markdown content for correctness, clarity, structure, grammar, and consistency with additional checks for prompts and skills.

Add the marketplace to Codex

Install the Codex CLI and add this GitHub repository as a marketplace:

codex plugin marketplace add aforemendude/skills

Verify that Codex knows about it:

codex plugin marketplace list

The marketplace declares both plugins as installed by default. If either one is not available in your Codex setup, install it manually:

codex plugin add agent-fix@aforemendude-skills
codex plugin add agent-review@aforemendude-skills

Codex detects newly installed skills automatically. If they do not appear, restart Codex.

Use a local clone

For marketplace development, clone the repository and register its root directory:

git clone git@github.com:aforemendude/skills.git
cd skills
codex plugin marketplace add "$PWD"
codex plugin add agent-fix@aforemendude-skills
codex plugin add agent-review@aforemendude-skills

Restart the ChatGPT desktop app after changing or installing a local plugin.

Update or remove the marketplace

codex plugin marketplace upgrade aforemendude-skills
codex plugin marketplace remove aforemendude-skills

Skills and example prompts

Agent Fix

$fix-tests-styles

$fix-tests-styles requires an unambiguous scope such as files, directories, packages, components, a diff, or the whole repository. It fixes both CSS and unit tests in that scope.

The skill may create, modify, or remove test-support files and edit setup or configuration required by the modified tests. It does not add or update dependencies. Its concise final report omits change statistics and details readily available from the diff. It reports checks, unresolved source defects, likely expected work that was not performed (including test-support refactoring blocked by effects outside the confirmed scope), recommended setup, configuration, or dependency changes, and other context the user would likely need.

Use $fix-tests-styles only in trusted repositories. The skill treats repository content as trusted and runs the repository's existing build and test commands, which may execute repository-controlled code.

Fix a component's styles and tests:

Use $fix-tests-styles to fix styles and unit tests for the Card component under packages/dashboard/src.

Optional opt-ins:

• Class-name renames and source-ownership moves: these remain report-only unless the prompt explicitly authorizes their corresponding fixes.

  Use $fix-tests-styles to fix styles and unit tests under packages/dashboard/src. Also fix inconsistent class-names
  and source-ownership problems in that scope.
  

• Production implementation fixes: production defects exposed by test updates remain report-only unless separately authorized.

  Use $fix-tests-styles to fix styles and unit tests under packages/dashboard/src/cards. If the test updates expose a
  production implementation defect in that scope, fix it without changing intended behavior.
  

Agent Review

$review-code

$review-code requires an unambiguous code review scope, such as files, directories, packages, applications, features, components, a diff or change set, a commit or commit range, or the whole repository. It requires a clean worktree and writes report-only findings progressively to CODE_REVIEW_<SCOPE_DESCRIPTION>.md. Large scopes are split into logical segments with one report per segment. The reviewer may run focused relevant static checks or tests when they provide useful evidence, using existing dependencies only.

Test-related findings are limited to dependencies, infrastructure, setup, and configuration. The skill does not review individual test cases, their fixture data, test logic, assertions, coverage adequacy, or missing test scenarios.

The review also checks project-wide dependency manifests and lockfiles for missing, unused, duplicate, incorrectly scoped, incompatible, or conflicting dependencies. It reviews application, build, packaging, deployment, runtime, CI, and local-development configuration and setup for concrete correctness, security, compatibility, reliability, and maintenance problems.

The code review also checks comments for inaccurate, stale, misleading, or contradictory content and flags grammar, spelling, typographical, and misleading identifier issues when they could cause misunderstanding or incorrect use. Formatting is left to the repository formatter.

Use $review-code only in trusted repositories. The skill treats repository content as trusted and may run the repository's existing static checks or tests, which may execute repository-controlled code.

Review the whole repository:

Use $review-code to review the whole repository.

Review a package's current code:

Use $review-code to review the current code under packages/dashboard.

Review project dependencies, configuration, and setup:

Use $review-code to review the repository's dependency manifests, lockfiles, configuration, and setup.

$review-markdown

$review-markdown requires one or more exact Markdown file paths and a review scope, such as complete current contents, uncommitted changes, a commit or commit range, or named sections or lines. It reviews correctness, clarity, structure, grammar, and consistency, with additional instruction-design, safety, metadata, and operational checks for AI prompts and skills. It writes MARKDOWN_REVIEW.md by default.

Review one complete file:

Use $review-markdown to review the complete contents of README.md.

Review a skill's uncommitted changes:

Use $review-markdown to review the uncommitted changes in plugins/agent-review/skills/review-markdown/SKILL.md.

Optional opt-ins:

• Apply fixes: reviewed Markdown files are not edited unless the prompt explicitly requests fixes.

  Use $review-markdown to review the complete contents of docs/setup.md and fix every finding you can resolve safely.
  

• Choose the output: provide a report path or request another output mode instead of the default generated Markdown report.

  Use $review-markdown to review the complete contents of README.md and return the complete review in chat instead of
  writing a report file.
  

• Run additional Markdown checks: $review-markdown runs link verification, web browsing, builds, tests, and live model inference only when explicitly requested.

  Use $review-markdown to review the complete contents of docs/prompt.md. Also verify its links, browse authoritative
  sources to check its factual claims, run the relevant build and tests, and use live model inference to evaluate the
  prompt's behavior.
  

• Validate external metadata schemas: metadata review otherwise covers only consistency and staleness within the repository.

  Use $review-markdown to review the complete contents of plugins/agent-review/skills/review-markdown/SKILL.md and
  perform full validation against the latest authoritative metadata schemas.