Vibe Test

Test analyzer and generator for vibe-coded apps. Classification-driven. Tiered by deployment context.

Catches the broken harnesses every other test tool assumes away.

Vibe Test reads a vibe-coded application, classifies it by maturity tier and deployment risk, and generates the tests that actually matter — proportional to what the app is, not what a spec would have asked for. It's the diagnostic-and-retrofit layer the ecosystem was missing: superpowers:test-driven-development owns discipline for new features, Tessl/Codecov/Coveralls measure what already exists — nobody owned the audit for apps that already shipped without tests.

The headline capability is F2 harness-break detection: vibe-coded apps routinely have package.json test scripts that reference uninstalled binaries, vitest configs with broken pool settings that time out and falsely report 0%, and coverage tools that cherry-pick the denominator (reporting 88% over 3 of 43 files). Vibe Test flags each as a distinct finding class with remediation guidance, separated from test-logic breaks.

Read the full thesis in framework.md — the 421-line argument for why this layer needs to exist and how the classification matrix and 5-level testing pyramid are structured.

Install

Claude Code marketplace (plugin — /vibe-test:* slash commands in-session):

/plugin marketplace add estevanhernandez-stack-ed/vibe-plugins
/plugin install vibe-test@vibe-plugins

npm (CLI — CI-safe deterministic commands, no LLM):

npm install -g @esthernandez/vibe-test-cli

The CLI ships audit / coverage / gate / posture — the deterministic operations that don't require Claude reasoning. generate and fix live in the plugin SKILL layer and require a Claude Code session (CLI exits 2 with a clear message pointing you at /vibe-test:generate).

Quick start

In a Claude Code session, inside your project:

/vibe-test:audit

That's the front door. The audit scans your app (routes, components, models, integrations, test infra), classifies it (app type × maturity tier × context modifiers), computes an honest-denominator coverage score using the locked weighted formula, and writes three artifacts:

• docs/vibe-test/audit-<timestamp>.md — the full human-readable report
• .vibe-test/state/audit.json — JSON sidecar for CI/tooling consumers
• Terminal banner — in-chat summary + next-step prompt

From there:

• /vibe-test:generate — retrofits tests for the gaps audit found (confidence-routed: auto ≥0.90 / stage 0.70–0.89 / inline <0.70)
• /vibe-test:coverage — honest-denominator coverage with adaptation prompt for cherry-picked setups
• /vibe-test:gate — CI-safe pass/fail against tier threshold (exit 0/1/2 + GitHub Actions annotations)
• /vibe-test:fix — diagnose + repair broken tests; catches harness-level breaks via F2
• /vibe-test:posture — read-only ambient summary, ≤40 lines, under 3s
• /vibe-test:evolve — agent-authored proposals based on accumulated friction / wins logs (Pattern #10)

After the first audit, Vibe Test writes (and later updates) docs/TESTING.md — a framework-agnostic runbook that survives uninstalling the plugin. See that file for per-project guidance on how your team actually runs + adds tests.

Command reference

| Command | Purpose | |---------|---------| | /vibe-test | Router — identity, state-awareness, next-step prompt | | /vibe-test:audit | Inventory + classify + score — diagnoses whole testing posture | | /vibe-test:generate | Confidence-tiered test generation (auto / stage / inline) | | /vibe-test:fix | Diagnose + repair broken tests or broken harnesses | | /vibe-test:coverage | Honest-denominator coverage measurement with tier interpretation | | /vibe-test:gate | CI-safe pass/fail against app tier threshold | | /vibe-test:posture | Read-only ambient summary of current testing posture |

Works better with

Vibe Test is built around Pattern #13 (Ecosystem-Aware Composition) — it announces deferrals to complementary skills/plugins instead of duplicating them. If any of these are installed, /vibe-test:* detects them and cedes the right slice of work:

| Complement | Install | Why Vibe Test defers | |------------|---------|----------------------| | superpowers:test-driven-development | /plugin install superpowers-plugin@superpowers | Owns discipline for new features; Vibe Test stays in the retrofit/audit lane | | superpowers:systematic-debugging | /plugin install superpowers-plugin@superpowers | /vibe-test:fix delegates complex failures instead of guessing | | superpowers:verification-before-completion | /plugin install superpowers-plugin@superpowers | /vibe-test:gate co-invokes for per-task "is this complete" decisions | | Playwright plugin + Playwright MCP | see Anthropic docs | E2E scaffolding; Vibe Test writes deferral stubs rather than native Playwright code | | Tessl analyzing-test-coverage | Tessl marketplace | Raw coverage parsing; Vibe Test overlays tier interpretation + honest denominator | | @esthernandez/vibe-doc | /plugin install vibe-doc@vibe-plugins | Co-authors docs/TESTING.md when both are present | | @esthernandez/vibe-sec (coming in v0.2) | /plugin install vibe-sec@vibe-plugins | Exchanges .vibe-test/state/covered-surfaces.json ↔ .vibe-sec/state/findings.jsonl for security-aware generation |

None of these are required — Vibe Test works fine standalone. When present, the announcement fires at command start so you know which skill is driving which decision.

Anchored registry lives at skills/guide/references/plays-well-with.md. Dynamic discovery is capped at one suggestion per invocation to prevent suggestion fatigue.

Environment

Vibe Test itself reads no environment variables. There is no .env.example by design — env-var detection runs against your app during /vibe-test:audit and /vibe-test:generate so we can warn when a test will depend on FIREBASE_API_KEY / STRIPE_PK / similar, and scaffold the CI stub's env: block with placeholders. The plugin itself has no secrets, no external services, no telemetry, no network calls beyond what Claude Code's agent runtime already does.

What's inside

• Router + 7 subcommand SKILLs — persona-aware, tier-adaptive language (experienced builders see dense technical detail; first-timers see plain-English framing — per story SE7)
• Deterministic primitives — scanner (typescript-estree AST walk), framework detector, route/component/model/integration inventory, weighted-score formula, reporter with three renderers (markdown + banner + JSON)
• Generate flow — confidence-tiered routing, --dry-run with 24h --apply-last-dry-run cache, env-var detection, HEAD-hash branch-switch check, rejection-pattern probe at ≥3 consecutive rejects, framework-idiom matching (vitest / jest detected)
• Builder-sustainable handoff — docs/TESTING.md (6 sections, framework-agnostic), docs/test-plan.md (chronological decision log), opt-in .github/workflows/vibe-test-gate.yml stub with env placeholders, graduating-to-next-tier guide
• Self-Evolving Plugin Framework — L1 builder profile, L2 session memory, Pattern #14 wins.jsonl, Pattern #15 active-path.json + RESOLVE.md, Pattern #16 blocking vs shaping prereqs declared per-SKILL
• CLI wrapper — @esthernandez/vibe-test-cli for CI (audit/coverage/gate/posture), emits ::error:: / ::warning:: annotations under GITHUB_ACTIONS=true

Non-goals (principled cuts)

Vibe Test is not:

1. A test runner (your framework — vitest / jest / playwright — runs tests)
2. A mutation testing tool
3. A visual regression platform
4. A load testing platform at scale
5. A 100%-coverage machine — we target meaningful, tier-appropriate coverage
6. A replacement for understanding your own tests (handoff artifacts survive plugin uninstall by design)
7. A feature-code modifier — /vibe-test:fix only edits tests

Full thesis in framework.md. Product spec in the monorepo docs/spec.md.

Links

• Thesis: framework.md — the 421-line argument
• Monorepo: vibe-plugins — sibling plugins (vibe-sec, future @626labs/plugin-core)
• Dogfood report: docs/dogfood-wseyatm-v0.2.md — first-patient validation against WeSeeYouAtTheMovies
• Changelog: CHANGELOG.md

License

MIT — © 2026 626Labs LLC, Fort Worth, TX.