Skip to content

boonhow97/everything-codex

v1.0.1MIT

Codex-native skills and specialist agents for disciplined software delivery.

GitHubOpen Website

Everything Codex

Everything Codex is a production-oriented, Codex-native toolkit for planning, building, reviewing, testing, refactoring, and documenting software. It adapts the strongest ideas from Everything Claude Code to the surfaces Codex actually supports today.

This is an adaptation, not a search-and-replace port. Reusable workflows are skills, durable repository policy belongs in AGENTS.md, specialist roles are Codex custom-agent TOML files, hooks use the current Codex event schema, and MCP examples are opt-in and disabled by default.

Quick start

Requirements: Git and Node.js 20 or newer. The toolkit itself has no runtime npm dependencies.

git clone https://github.com/BoonHow97/everything-codex.git
cd everything-codex
npm test

Install the skills through Codex's supported marketplace flow:

codex plugin marketplace add BoonHow97/everything-codex
codex plugin add everything-codex@everything-codex
codex plugin list

Start a new Codex task after installation so the seven skills are loaded. The marketplace manifest points to this repository root.

The custom agents, AGENTS.md files, rules, hooks, and MCP files are templates, not automatically activated plugin components. Copy only the pieces you review and want in a project; see customization.

For development or vendoring, create a non-destructive standalone source copy:

node scripts/install.mjs --dry-run --target /absolute/path/to/everything-codex
node scripts/install.mjs --target /absolute/path/to/everything-codex

PowerShell:

.\scripts\install.ps1 -DryRun -Target 'C:\path\to\everything-codex'
.\scripts\install.ps1 -Target 'C:\path\to\everything-codex'

Bash:

./scripts/install.sh --dry-run --target "$HOME/.codex/plugins/everything-codex"
./scripts/install.sh --target "$HOME/.codex/plugins/everything-codex"

The installer refuses existing destinations. --force first renames an existing destination to a timestamped backup and restores it if copying fails. It never edits ~/.codex/config.toml, credentials, or marketplace configuration.

What is included

  • Seven focused skills under skills/ for delivery, review and security, E2E testing, safe refactoring, documentation and durable context, and frontend/backend patterns.
  • Nine project-scoped custom-agent templates under .codex/agents/ for planning, architecture, TDD, review, security, build failures, E2E testing, refactoring, and documentation.
  • A concise root AGENTS.md, modular policy templates under rules/, and scoped examples under examples/agents/.
  • Opt-in Codex hook and MCP configuration examples under examples/; none are silently activated.
  • Portable installation and validation scripts with automated tests.
  • A portable Agent Plugins 1.0 manifest, a supported .codex-plugin/plugin.json compatibility manifest, and a repository marketplace manifest for Codex installation.

See the component catalog, Claude-to-Codex compatibility map, and customization guide.

Recommended workflows

  • Ask Codex to use $delivery-loop for a feature, bug fix, or build recovery that should end with evidence.
  • Ask for $review-and-security when you want findings prioritized by real impact.
  • Ask for $e2e-testing for a browser journey or a flaky end-to-end test.
  • Ask for $refactor-safely before removing dead code or consolidating implementations.
  • Ask for $docs-and-memory to update durable documentation or capture a reusable project fact.
  • Ask for $frontend-patterns or $backend-patterns when the task needs domain-specific design guidance.
  • Explicitly request parallel subagents when independent exploration, review, or test work would help. Subagents cost additional tokens and are not useful for every task.

Validation

npm test
npm run validate

If the local npm launcher is unavailable, run the same zero-dependency checks directly:

node --test test/*.test.mjs
node scripts/validate.mjs

Validation checks manifests, skill frontmatter, custom-agent definitions, referenced paths, duplicate names, unsafe placeholders, accidental secret material, and installer behavior.

Safety and limitations

  • No MCP server is enabled automatically. Start with read-only tools and narrow allowlists.
  • Hook examples are inactive until deliberately copied into an active hook location and trusted in Codex.
  • The toolkit does not parse unstable transcript formats or claim a status-line API that Codex does not expose.
  • Available models, subagents, browsers, permissions, and MCP tools depend on the Codex surface and workspace policy. Agent files inherit the parent model instead of pinning a model name.
  • The instructions do not authorize commits, pushes, deployments, destructive changes, credential access, or external messages. Those actions still require the user's request and the active permission policy.

Read SECURITY.md before enabling hooks or MCP servers. For common setup issues, see troubleshooting.

Attribution and license

Everything Codex is independently written for Codex and inspired by the component taxonomy and workflow ideas in Everything Claude Code. See ATTRIBUTION.md. Released under the MIT License.