apex

Apex

Apex gives your coding agent senior-engineering judgment for designing, implementing, reviewing, and investigating software changes. It ships as open Agent Skills that are language- and framework-agnostic and run in any skills-compatible agent — with ready-to-install plugins for Claude Code and Codex.

Four focused skills activate only when the task calls for them, alongside an optional always-on engineering constitution. Repository: github.com/zeybek/apex.

Install

Install the published package for your client:

# Claude Code
claude plugin marketplace add zeybek/apex
claude plugin install apex@apex

# Codex
codex plugin marketplace add zeybek/apex
codex plugin add apex@apex

For local development, point the marketplace at a path instead of zeybek/apex, then install with the same apex@apex commands:

claude plugin marketplace add <repository-path>
codex plugin marketplace add <repository-path>

Using another skills-compatible agent (Cursor, OpenCode, Gemini CLI, and others)? See adapters/README.md for discovery paths and copy/symlink installation.

Skills

Skill	Use it to
apex-design	Make architecture, API, schema, migration, and build-versus-buy decisions
apex-implement	Implement features, fixes, refactors, migrations, and production changes end to end
apex-review	Run risk-first reviews with actionable findings by severity
apex-investigate	Diagnose incidents, regressions, flaky failures, and unknown-cause bugs before fixing

Each skill is a SKILL.md workflow under plugins/apex/skills/; deeper guidance lives in references/ and loads only when relevant. Keeping the workflows separate keeps each one focused and lets only the relevant skill enter context.

Examples

examples/ contains one client-neutral walkthrough for each skill. The implement and investigate walkthroughs include small standard-library workspaces plus deterministic verifier scripts, so you can exercise the workflow without changing the canonical example files.

Design principles

• Optimize for correct, useful, maintainable software and safe delivery.
• Prefer the simplest solution only after requirements and risk controls are satisfied.
• Scale rigor with blast radius, reversibility, criticality, exposure, persistence, concurrency, compatibility, operability, and uncertainty.
• Treat security, privacy, testing, observability, migrations, rollout, and recovery as engineering work, not optional follow-up.
• Preserve project conventions and contracts unless evidence justifies changing them.

Always-on constitution

AGENTS.md is a compact engineering constitution that sits outside the skill specification. Merge it into your client's always-on instruction file (AGENTS.md, CLAUDE.md, or equivalent) when its rules should apply to every task, independently of whether the skills are active. See adapters/README.md for per-client setup.

Packaging

The same plugins/apex/skills/ directory backs every client; the client-specific manifests carry distribution metadata only and never duplicate skill instructions.

• plugins/apex/ — the portable plugin package (SKILL.md, references/, and evals/ per skill);
• plugins/apex/.claude-plugin/plugin.json and .claude-plugin/marketplace.json — Claude Code packaging and marketplace;
• plugins/apex/.codex-plugin/plugin.json and .agents/plugins/marketplace.json — Codex packaging and marketplace.

Codex install-surface UI metadata (display name, descriptions, default prompts) lives in the .codex-plugin/plugin.json interface block, not in the skill folders.

Evaluation

Every skill ships its own evals:

• evals/trigger-evals.json — realistic positive and near-miss negative prompts for activation accuracy;
• evals/evals.json — output-quality scenarios with expected outcomes and observable assertions.