We help your agents test edge cases in your software where rare production bugs hide.
Most AI-written tests optimize for coverage. They assert implementation details,
mock away the real risk, and pass even when the product breaks.
@workersio/skills gives coding agents a testing workflow that asks a stricter
question: will this test catch a real regression that users, operators,
reviewers, or maintainers care about?
@workersio/skills is one skill with five command modes:
$wio scan # Find high-value tests to add next.
$wio test # Run candidate discovery, strategy, implementation, and review.
$wio workload # Generate realistic, replayable workloads with new bug-finding value.
$wio review # Judge whether a test should be kept, redone, or removed.
$wio doctor # Audit suite health, weak assertions, flakes, mocks, and CI blind spots.
@workersio/skills is packaged for Codex and Claude Code today. The core skill
content is plain Markdown, so the testing workflow can also be adapted by other
coding agents that support project skills, instructions, or reusable prompts.
Direct skill install:
npx skills add workersio/skillsCodex plugin marketplace:
codex plugin marketplace add workersio/skillsClaude Code plugin marketplace:
claude plugin marketplace add workersio/skills
claude plugin install wio@workersio-skillsFor local Codex plugin testing from this checkout:
codex plugin marketplace add .| Area | What the agent gets |
|---|---|
| Test discovery | Maps product behavior, recent changes, existing tests, CI, and risk areas before choosing what to test. |
| Strategy selection | Chooses the right level: unit, component, integration, contract, E2E, workload, fuzz, property, mutation, resilience, or monitoring. |
| Test writing | Adds focused tests using the repository's own framework, naming, fixtures, helpers, and runner conventions. |
| Workload generation | Creates realistic, adversarial, seeded, replayable sessions or traffic that add coverage beyond wrappers and parameter sweeps. |
| Test review | Applies a strict value gate: KEEP, REDO, or REMOVE. Coverage alone is not enough. |
| Suite health | Finds weak assertions, over-mocking, broad snapshots, flakes, skipped tests, slow feedback loops, and CI blind spots. |
| Command | Use it when | Example |
|---|---|---|
$wio scan [target] |
You do not yet know what to test. | $wio scan checkout |
$wio test [target] |
You want the full write-and-review loop. | $wio test billing eligibility regression |
$wio workload [target] |
The risk lives in realistic user, API, CLI, job, or load behavior. | $wio workload onboarding session |
$wio review [target] |
A test exists and you need to decide whether it has real value. | $wio review tests/billing_eligibility_test.py |
$wio doctor [target] |
The suite is hard to trust or maintain. | $wio doctor API test suite |
An approved test should answer:
- What user, operator, customer, or API consumer failure does this prevent?
- What production, release, support, debugging, or review risk does it reduce?
- Would it fail for the regression that matters?
- Which assertion or invariant catches the plausible bug?
- Does the setup preserve the important dependency, state, permission, timing, or data risk?
- Does this belong in local development, PR CI, nightly, release, or production monitoring?
- If this is a workload, what existing workload gap does it fill?
If those answers are weak, the test should be redesigned or removed.
@workersio/skills includes three optional 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 and owns the final decision. Subagents inspect, challenge, and review; they do not duplicate the reference library or replace the main workflow.
| Path | Purpose |
|---|---|
plugins/wio/skills/wio/SKILL.md |
Source of truth for the skill workflow. |
plugins/wio/skills/wio/references/ |
Detailed testing guidance loaded only when relevant. |
plugins/wio/agents/ |
Claude Code plugin subagents. |
.codex/agents/ |
Codex custom-agent TOML files for project or user installs. |
plugins/wio/hooks/hooks.json |
Shared plugin hook config. |
.agents/plugins/marketplace.json |
Codex marketplace entry. |
.claude-plugin/marketplace.json |
Claude Code marketplace entry. |
public/ |
README assets, demo GIF, and agent icons. |
Codex plugin installs include the skill and plugin hook config. Codex custom
agents are a separate native surface loaded from .codex/agents/ in a project
or ~/.codex/agents/ for the user.
Enable the Codex agents globally:
mkdir -p ~/.codex/agents
cp .codex/agents/wio-*.toml ~/.codex/agents/Enable the Codex agents for the current project:
mkdir -p .codex/agents
cp /path/to/wio-skills/.codex/agents/wio-*.toml .codex/agents/Enable the Codex hook config for the current project:
mkdir -p .codex
cp /path/to/wio-skills/.codex/hooks.json .codex/hooks.jsonClaude plugin installs include the skill, plugin hooks, and Markdown
subagents from plugins/wio/agents/. Project-local Claude Code agents can also
be copied into .claude/agents/ when a repository is not using the plugin.
Detailed testing guidance lives only in
plugins/wio/skills/wio/references/. Reference topics cover behavior mapping,
risk-based testing, test levels, workload modeling, oracles, fixtures, mocks,
suite health, static analysis, security testing, fuzzing, property-based
testing, mutation testing, performance testing, resilience testing, and
regression selection.
Keep the public surface area small: one skill, wio, with command modes
scan, test, workload, review, and doctor.
Detailed testing guidance belongs in
plugins/wio/skills/wio/references/, not duplicated inside cloud folders,
subagents, hooks, or extra skill trees. When adding a reference topic, add both
overview.md and tools.md, then link it from
plugins/wio/skills/wio/references/index.md.
See CONTRIBUTING.md for contribution workflow and release expectations.
MIT. See LICENSE.
