# Test Specialist > Test orchestrator for PwnDoc. Analyzes a feature or the full project to determine test gaps across backend (Jest), frontend unit (Vitest), and frontend E2E (Playwright), then delegates to /test-backend, /test-frontend-unit, and /test-frontend-e2e skills. - Skill: `pwndoc/test-specialist` (Agent Skill) - Install (CLI): `npx skillmds@latest add pwndoc/test-specialist` - Raw SKILL.md: https://api.skillmd.com/api/skills/pwndoc/test-specialist/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Web & Frontend - Author: pwndoc (https://skillmd.com/u/pwndoc) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/pwndoc/test-specialist --- # Test Specialist - Orchestrator You are the PwnDoc test orchestrator. Your job is to analyze what tests are needed and coordinate the three test sub-skills. ## Arguments `$ARGUMENTS` - Either: - A feature name (e.g., "audit", "client", "vulnerability") to test that feature across all layers - `gaps` or `all` to analyze and fill all test coverage gaps ## How to Operate ### Step 1: Analyze Scope **If feature name provided:** Identify all files related to that feature across all layers: - Backend: `backend/src/routes/{feature}.js`, `backend/src/models/{feature}.js` - Frontend service: `frontend/src/services/{feature}.js` - Frontend pages: `frontend/src/pages/{feature}*/` - Frontend components: related components - Existing tests in all three layers **If "gaps" or "all":** Scan all three test directories and compare against source files to build a complete gap list, then proceed to Step 1.5. ### Step 1.5: Prioritize Scope (gaps/all mode only) **This step only applies when the argument is "gaps" or "all".** Skip this step for feature-specific arguments. After building the gap list, you MUST ask the user before proceeding: 1. **Summarize gaps** — Print a concise summary showing the number of missing tests per category: ``` Test gaps found: - Backend: X missing test files - Frontend Unit — Components: X missing (list names) - Frontend Unit — Pages: X missing (list names) - Frontend E2E: X missing specs (list names) ``` (Omit categories with no gaps, e.g. skip "Services" if all are covered.) 2. **Ask the user to pick focus areas** — Use `AskUserQuestion` with `multiSelect: true` to let the user choose which categories to work on. Options should be the categories that have gaps (e.g., "Frontend Unit — Components", "Frontend Unit — Pages", "Frontend E2E"). 3. **Batch limit** — Cap the session to a maximum of **5 test files** total. If the user's selected categories contain more than 5 missing items, either: - Ask the user which specific items to prioritize, OR - Default to the top 5 items from the Priority Order section below Tell the user how many items you'll work on (e.g., "I'll create 5 test files this session based on priority order."). 4. **Proceed** — Only generate tests for the selected/prioritized items. Skip everything else. ### Step 2: Determine What Needs Tests For each layer, check what exists: **Backend** (tests in `backend/tests/`): - Already ~90% covered. Only delegate to `/test-backend` if the feature has NO existing test file or if you're adding tests for a new feature. - Check: Does `backend/tests/{feature}.test.js` exist? **Frontend Unit** (tests in `frontend/tests/unit/`): - Check for missing service tests in `frontend/tests/unit/services/` - Check for missing component tests in `frontend/tests/unit/components/` - Check for missing page tests in `frontend/tests/unit/pages/` - Check for missing store tests in `frontend/tests/unit/stores/` **Frontend E2E** (tests in `frontend/tests/e2e/`): - Check for missing `*.spec.js` files for major user flows - E2E tests cover critical user journeys, not every single page ### Step 3: Delegate to Sub-Skills **In gaps/all mode:** Only delegate for the items selected/prioritized in Step 1.5. Do NOT generate tests for unselected categories or items beyond the batch limit. For each layer that needs tests, invoke the appropriate sub-skill with `--no-run` so they generate tests without running them (you will run tests in Step 4): - `/test-backend {feature} --no-run` — for backend API tests - `/test-frontend-unit {type}:{name} --no-run` — for frontend unit tests - `service:audit --no-run`, `component:breadcrumb --no-run`, `page:profile --no-run`, `store:index --no-run` - `/test-frontend-e2e {feature} --no-run` — for E2E specs ### Step 4: Run and Verify (Fix Loop) After ALL sub-skills have finished writing code, run the test cycle below. **CRITICAL RULES:** - **ONE test run per cycle.** Run `./pwndoc-cli test` ONCE, capture the full output, then analyze the stored results. Do NOT run the command multiple times to "investigate" — all information you need is in the single run's output. - **NEVER run `pwndoc-cli` in parallel.** Only one instance of `./pwndoc-cli` may run at a time across all agents and subagents. No two test commands may overlap. - **Fix agents must NEVER run tests.** When you delegate fixes to subagents/parallel agents, they must ONLY edit code. They must NOT run `./pwndoc-cli test` or any test command. Only YOU (the main test-specialist orchestrator) run tests, and only AFTER all fix agents have completed. - Always use `./pwndoc-cli test` commands. Never use `npm run test` or `npx` directly. - Use `--coverage` only for backend and frontend unit suites (not E2E). **Fix loop procedure:** 1. **Run tests once** — Execute the relevant test suites and store the full output: - `./pwndoc-cli test --backend` if backend tests were created/modified - `./pwndoc-cli test --frontend-unit` if frontend unit tests were created/modified - `./pwndoc-cli test --frontend-e2e` if E2E tests were created/modified 2. **Analyze stored results** — Read through the captured output to identify all failures. Do NOT re-run tests to get more information. 3. **Fix all failures** — Delegate fixes to the relevant sub-skills using `--fix` mode with `--no-run`. Group failures by layer, then invoke in parallel: - `/test-frontend-unit --fix {test-file} "{error}" --no-run` for frontend unit test failures - `/test-frontend-e2e --fix {spec-file} "{error}" --no-run` for E2E spec failures - `/test-backend --fix {test-file} "{error}" --no-run` for backend test failures Sub-skills must NOT run tests (ensured by `--no-run`). Wait for ALL to complete before proceeding. 4. **Re-run tests once** — After ALL fixes are written, run the test command again (once) to verify. Repeat from step 2 if there are still failures. 5. **Maximum 2 iterations.** If tests still fail after 2 fix cycles, report the remaining failures to the user rather than continuing to loop. ## Coverage Thresholds When running coverage tests or filling gaps, aim to reach these thresholds: **Backend** (`./pwndoc-cli test --backend --coverage`): - Statements: `75%` - Lines: `75%` - Functions: `70%` - Branches: `60%` (advisory) **Frontend Unit** (`./pwndoc-cli test --frontend-unit --coverage`): - Statements: `65%` - Lines: `65%` - Functions: `60%` - Branches: `50%` (advisory) **E2E** (`./pwndoc-cli test --frontend-e2e`): - No coverage percentages - Focus on critical user flows (auth, audit CRUD, report generation) - Validate stable pass/fail behavior ## Automatic Threshold Enforcement When adding features or filling gaps: 1. **Run coverage**: Execute `./pwndoc-cli test --backend --coverage` or `./pwndoc-cli test --frontend-unit --coverage`. 2. **Parse results**: Extract statements, lines, functions, and branches percentages from the coverage output. 3. **Compare to thresholds**: Check each metric against the targets above. 4. **If below threshold**: - Identify uncovered paths from the coverage report. - Prioritize: critical business logic (auth, permissions, CRUD), shared services/stores, error handling and validation, high-reuse components. - Add targeted tests to close gaps. - Re-run coverage once. 5. **Completion rule**: - Task is complete when thresholds are met, or measurably improved (`>= 5%` gain in any below-threshold metric). - If thresholds are still not met after one gap-filling round, report current coverage and remaining gaps to the user instead of looping indefinitely.