wio

Better tests, not more tests

Most AI-written tests optimize for coverage. They assert implementation details, mock away the real risk, and pass even when the product breaks.

WIO is one testing workflow skill with four commands: $wio scan, $wio test, $wio review, and $wio doctor.

Structure

WIO uses one skill, command routing inside SKILL.md, and one shared references tree. The canonical source lives inside the plugin so plugin installs and direct skill installs do not drift:

plugins/wio/
  .codex-plugin/plugin.json
  .claude-plugin/plugin.json
  skills/
    wio/
      SKILL.md
      scripts/
        test-review-reminder.py
      references/
        index.md
        <topic>/
          overview.md
          tools.md
  agents/
    wio-candidate-scout.md
    wio-strategy-critic.md
    wio-test-reviewer.md
  hooks/
    hooks.json

There are no separate scan, test, review, or doctor skills, no symlinked skill copies, and no copied reference trees.

Install Surfaces

WIO supports the official install surfaces for each host:

Host	Official shared install	Includes
Codex	Codex plugin via .agents/plugins/marketplace.json and plugins/wio/.codex-plugin/plugin.json	Skill and plugin hook config. Codex plugin hooks require plugin_hooks to be enabled in the current release.
Claude Code	Claude plugin via .claude-plugin/marketplace.json and plugins/wio/.claude-plugin/plugin.json	Skill, Claude subagents, and Claude plugin hook config.
Agent skills installer	npx skills add workersio/skills --skill wio	Skill only: SKILL.md, scripts/, and references/. The installer discovers plugins/wio/skills/wio directly.
Codex project config	.codex/agents/ and .codex/hooks.json	Codex custom subagents and project hooks. Codex custom agents are not installed by skills add.
Claude project config	.claude/agents/ and .claude/settings.json	Project-local Claude agents and hooks for development or non-plugin use.

The repo-level marketplaces are:

.agents/plugins/marketplace.json      # Codex marketplace
.claude-plugin/marketplace.json       # Claude Code marketplace

Install

Direct skill install:

npx skills add workersio/skills

Claude Code plugin development:

claude --plugin-dir ./plugins/wio

Claude Code marketplace install:

claude plugin marketplace add workersio/skills
claude plugin install wio@workersio-skills

Claude plugin installs include the WIO skill, plugin hooks, and Markdown subagents from plugins/wio/agents/.

Codex marketplace add:

codex plugin marketplace add workersio/skills

For local Codex plugin testing, add this checkout as the marketplace root:

codex plugin marketplace add .

Codex plugin installs include the WIO skill and plugin hook config. Codex custom agents are a separate native surface: they are loaded from .codex/agents/ in a project or ~/.codex/agents/ for the user.

To enable WIO Codex agents globally:

mkdir -p ~/.codex/agents
cp .codex/agents/wio-*.toml ~/.codex/agents/

To enable WIO Codex agents for the current project:

mkdir -p .codex/agents
cp /path/to/wio-skills/.codex/agents/wio-*.toml .codex/agents/

To enable the WIO Codex hook config for the current project:

cp /path/to/wio-skills/.codex/hooks.json .codex/hooks.json

Verify Codex agent files:

find .codex/agents ~/.codex/agents -name 'wio-*.toml' 2>/dev/null

Commands

Command	What it does
$wio scan [target]	Maps product behavior, existing tests, CI, and risk areas to find the highest-value tests to add next.
$wio test [target]	Runs the full loop: discover candidate, pick strategy, write test, validate, review, then keep only if valuable.
$wio review [target]	Reviews a test for customer value, developer-flow value, signal quality, maintainability, and false confidence.
$wio doctor [target]	Audits test-suite health: weak assertions, flakes, excessive mocks, broad snapshots, slow feedback, skipped tests, and missing critical behavior coverage.

Subagents

WIO includes three focused subagents:

Subagent	Role
wio-candidate-scout	Read-only discovery of high-value test candidates before implementation.
wio-strategy-critic	Read-only challenge of the selected strategy before editing tests.
wio-test-reviewer	Read-only post-write review that returns KEEP, REDO, or REMOVE.

The main agent still writes the test. Subagents gather evidence, challenge the strategy, and review value. They do not duplicate reference content and they do not own the workflow.