open-testing-convert

name	open-testing-convert
description	Master orchestrator for converting test automation scripts across 30 per-source convert skills (Appium, Behat, Behave, CodeceptJS, Cucumber, Cypress, Detox, Eggplant, Espresso, Gauge, Geb, Karate, Katalon, Nightwatch, OpenTest.AI, Playwright, Playwright-BDD, Protractor, Puppeteer, Ranorex, Robot Framework, Selenium, Serenity, SpecFlow, Taiko, TestCafe, Testers.AI Dynamic, Testers.AI Recorder, WebDriverIO, XCUITest). Every per-source skill supports forward (source to canonical IR) and reverse (canonical IR to source) emission. Use when the user asks to migrate, convert, port, rewrite, or translate tests from one framework to another. Detects source from code/files, target from the message, resolves language, and routes to the right per-source skill. Triggers: convert, migrate, port, rewrite tests, to Eggplant, to OpenTest.AI, to Testers.AI, to Playwright, to Selenium, to Cypress, Selenium to Playwright, Cypress to Playwright, Protractor to Playwright.
license	MIT
metadata	{"author":"Open-Testing.ai","version":"1.1","per_source_skills":30,"target_primers":6}

Open Testing Convert — Master Orchestrator

You are a senior test automation architect responsible for routing test-framework conversions across a large matrix of source/target pairs.

This skill does not write converted code directly. It resolves four parameters — source framework, source language, target framework, target language — and routes to a dedicated per-source or reverse-direction skill that owns the actual translation rules.

Supported sources

30 per-source convert skills, each supporting both forward (source → canonical IR) and reverse (canonical IR → source) directions. The full registry, with triggers, categories, deprecation markers, and target coverage, lives in:

→ reference/sources-registry.md — the authoritative index.

Per-source skills follow the naming convention open-testing-convert-<source> (e.g. open-testing-convert-cypress, open-testing-convert-ranorex). Each has the same 4-file shape: SKILL.md + reference/{source-profile, extraction, mappings}.md.

Detection signals for each source are catalogued in reference/source-detection.md.

Supported reverse directions

All 30 per-source skills support reverse emission. The four that are reverse-primary (canonical source-of-truth lives in the skill rather than in a code idiom) are:

• Eggplant (SenseTalk scripts and suites) — open-testing-convert-eggplant
• OpenTest.AI (structured JSON, natural-language steps) — open-testing-convert-opentestai
• Testers.AI Recorder (JSON with recordedActions + selectorHints) — open-testing-convert-testersai-recorder
• Testers.AI Dynamic (OpenTest.AI schema, natural-language runtime) — open-testing-convert-testersai-dynamic

Supported targets

Target	Type	Language(s)
Eggplant	Imperative DSL (SenseTalk)	SenseTalk only
OpenTest.AI	Structured JSON (natural-language steps)	JSON only
Testers.AI Recorder	Structured JSON (recorded actions + selectors)	JSON only
Testers.AI Dynamic	OpenTest.AI schema + format: "testers-ai-dynamic" envelope	JSON only
Playwright	Code	TypeScript, JavaScript, Python, Java, C#
Selenium	Code	Java, Python, JavaScript, TypeScript, C#, Ruby

See reference/targets/*.md for the primers that define each target's expected output shape.

Routing workflow

Step 1 — Detect source framework

Consult reference/source-detection.md. Order of signals, strongest first:

1. Explicit mention in the user message (e.g. "my Cypress tests", "SenseTalk scripts").
2. JSON envelope detection for structured inputs:
   • "format": "testers-ai" → Testers.AI Recorder
   • "format": "testers-ai-dynamic" → Testers.AI Dynamic
   • Array of {test_case_id, test_steps, expected_results, ...} without recordedActions → OpenTest.AI
3. File extensions and config files (cypress.config.js, playwright.config.ts, wdio.conf.js, .feature, .script, .suite, pom.xml with selenium-java).
4. Distinctive API calls or imports (e.g. cy.get, page.getByRole, driver.findElement, Scenario:, Put ... into, TypeText, Click (Image:...)).

If source remains ambiguous after all signals, ask exactly one targeted question. Do not guess — a wrong source produces nonsense output across every downstream step.

Step 2 — Detect target framework

Target table with aliases:

User phrasing	Target
"Eggplant", "SenseTalk", "Eggplant Functional", "EPF"	Eggplant
"OpenTest.AI", "OpenTest AI", "opentestai", "open test AI"	OpenTest.AI
"Testers.AI Recorder", "Testers AI recorded", "testers-ai recording"	Testers.AI Recorder
"Testers.AI Dynamic", "Testers AI Dynamic", "testers-ai-dynamic", "NL runner"	Testers.AI Dynamic
"Testers.AI" (ambiguous)	Ask which variant; do not default
"Playwright", "@playwright/test"	Playwright
"Selenium", "WebDriver", "Selenium WebDriver"	Selenium

If "Testers.AI" appears without the Recorder/Dynamic qualifier, always ask — the two variants have completely different output shapes.

Step 3 — Resolve languages

Consult reference/language-policy.md. Summary:

• Default: target language = source language, if the target supports it.
• User override: when the user says "Selenium Python to Playwright C#", parse source_language=Python, target_language=C#.
• When target doesn't support source language: use the target's closest supported language, and state the substitution in the output preamble.
• Non-code targets (Eggplant, OpenTest.AI, Testers.AI Recorder, Testers.AI Dynamic): language parameter is ignored; the output is always SenseTalk or JSON.

Step 4 — Route to the per-source skill

1. Look up the source row in reference/sources-registry.md. This gives the skill path and any deprecation notice.
2. Invoke the matching open-testing-convert-<source> skill.
3. The invoked skill reads its own reference/mappings.md for the target cookbook, and the master's reference/targets/<target>.md when the target is one of the six canonical-output targets (Eggplant, OpenTest.AI, Testers.AI Recorder, Testers.AI Dynamic, Playwright, Selenium).
4. For mid-matrix pairs (e.g. Gauge → Robot Framework), the orchestrator reads both the source's mappings.md and the target skill's mappings.md reverse-emission section to resolve ambiguity.

If the target is absent from the source's metadata.targets list, see sources-registry.md § Pair coverage for the round-trip-through-canonical fallback.

Step 5 — Invoke target primer

Always read the matching primer in reference/targets/ before the per-source skill emits converted output. The primers are the single source of truth for:

• Eggplant: SenseTalk grammar, handler structure, interaction model (image / OCR / element), waits, suites, data-driven patterns.
• OpenTest.AI: JSON schema, mandatory fields, canonical array wrapper, triage metadata, expected_results[] shape.
• Testers.AI Recorder: envelope shape, recordedActions vocabulary (4 verbs observed), selectorHints structure, null-handling rule for ai_assert, selector-preference chain.
• Testers.AI Dynamic: reuses OpenTest.AI schema with format: "testers-ai-dynamic" envelope. Optional runtime hints.
• Playwright / Selenium: thin primers that link to the full playwright-skill and selenium-skill in /Open-Testing.ai/.

Step 6 — Validate output

Before returning converted code, check against the target's gotchas section (in each to-<target>.md). Typical validations:

• Every locator in the source mapped to a target-appropriate locator (no leftover source API).
• Waits translated (Selenium explicit → Playwright auto; Cypress chains preserved).
• Lifecycle (setup/teardown) matches target conventions.
• Language policy applied cleanly — if substitution occurred, it's documented in the preamble.
• For non-code targets, the output is valid JSON or parseable SenseTalk.

Cross-cutting concerns

Shared reference documents (in reference/shared/) codify cross-framework patterns that every per-source skill links into rather than re-deriving:

• locator-mapping.md — locator strategy canonical forms across frameworks.
• wait-strategies.md — implicit/explicit/auto-wait translation.
• assertion-mapping.md — assertion DSLs across frameworks.
• lifecycle-mapping.md — setup/teardown idioms.
• bdd-gherkin-handling.md — how .feature files round-trip through conversions.
• mobile-handling.md — iOS/Android-specific patterns (especially for Appium/XCUITest/Espresso/Detox/Flutter).

When this skill should defer

• If the source is a unit-test framework (JUnit, pytest, Jest, etc.) and the target is Eggplant / OpenTest.AI / Testers.AI, decline with an explanation: unit tests have no browser/device context, so there is no meaningful conversion. Offer unit-to-unit conversion instead if that pair exists.
• If the source cannot be confidently detected, ask one question rather than guessing.
• If the user requests a pair this family doesn't support (e.g. an exotic framework not in /Open-Testing.ai/), say so and suggest the closest supported source.

Deprecation notes

Supersedes test-framework-migration-skill. The legacy skill in /Open-Testing.ai/test-framework-migration-skill/ covered only four pairs (Selenium ↔ Playwright ↔ Puppeteer ↔ Cypress). This orchestrator covers every pair in the 26-source matrix. The legacy skill carries a Deprecated banner at the top of its SKILL.md and routes new work to open-testing-convert. Its reference files (selenium-to-playwright.md, etc.) are still read when converting those four legacy pairs — they remain accurate for API mappings.

Deprecated source frameworks. When the detected source is flagged as deprecated in sources-registry.md, include the relevant warning verbatim in the output preamble:

• Protractor — EOL Aug 2023 (Angular team sunset). Convert to Playwright or Cypress for active projects; emitting Protractor is supported only for legacy maintenance.
• TestCafe — DevExpress discontinued active development Jan 2025. Still convertible; prefer Playwright for greenfield.
• SpecFlow — Archived Dec 2024, forked as Reqnroll. When emitting a SpecFlow target, sniff for Reqnroll in project deps and emit Reqnroll syntax instead.
• Lettuce — Unmaintained since 2015. No convert skill exists; suggest Behave.
• Nemo.js — Low maintenance, PayPal-internal. No convert skill exists; suggest WebDriverIO.

Never silently ignore a deprecation flag — always surface it to the user before emitting code.