# Test Coverage Improvement > Autonomous Vitest coverage workflow for a chosen package under web/packages: baseline report, agent-curated shortlist of ~10 files, then per-file context from reading source, inferred behaviors, implemented tests, and verification (vitest + typecheck) before the next file; after the loop, **eslint only the files you created or edited** and fix issues. Human review happens outside this skill (e.g. MR, local diff). Use when improving unit coverage in the web monorepo. Trigger keywords — package coverage, vitest coverage, pnpm workspace tests, coverage baseline, raise coverage. - Skill: `nvidia-nemo/test-coverage-improvement` (Agent Skill) - Install (CLI): `npx skillmds@latest add nvidia-nemo/test-coverage-improvement` - Raw SKILL.md: https://api.skillmd.com/api/skills/nvidia-nemo/test-coverage-improvement/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: NVIDIA NeMo (https://skillmd.com/u/nvidia-nemo) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/nvidia-nemo/test-coverage-improvement --- # Package test coverage improvement (autonomous) **Scope:** A **single** package under **`web/packages/*`** that uses **Vitest** (TypeScript). Run commands from **`web/`** (pnpm workspace) or **`cd web/packages/`**. **Conventions:** Obey **`web/AGENTS.md`**. For **React** UI in `studio` / `common`, follow **`web/.agents/skills/unit-test/SKILL.md`**. For **plain TS**, use Vitest only; no watch until runs are green. **Human role:** The skill does **not** block on questions per file. The **user reviews** changes separately (PR, `git diff`, local run). If the user stated constraints up front (areas to focus, avoid), honor those; otherwise proceed using judgment below. ## 0. Choose the target package 1. If the user named a package, use it (e.g. `studio`, `common`, `sdk`). 2. Otherwise infer from open files or cwd under `web/packages/`. 3. Read **`package.json`**: **`name`**, **`test`**, **`test:ci`**, **`typecheck`**, **`lint`** / **`lint:fix`** (if present). 4. Read Vitest config (**`vite.config.ts`** / **`vitest.config.ts`**) → **`test.coverage`**: **`reportsDirectory`**, **`include` / `exclude`**. If there is **no** coverage tooling, add **`--coverage`** + reporters on the CLI for baseline, or build the shortlist without JSON (search + file importance only). ## 1. Baseline — current coverage 1. `cd web/packages/`. 2. Run (**no watch**): ```bash pnpm test:coverage ``` 3. Summarize **briefly** for the transcript: overall line (or statement) % from text or **`total`** in **`coverage-final.json`** (under **`reportsDirectory`**). ## 2. Shortlist ~10 files Combine **`coverage-final.json`** with judgment—**do not** use only the bottom ten by %: - Sort by low **`lines.pct`** / **`statements.pct`** among included source files. - **Prefer:** logic-heavy modules, hooks, API glue, non-trivial components, error paths. - **Skip or defer:** thin re-exports, empty barrels, generated-only files, trivial constants (unless user asked otherwise). - **Exclude:** `*.test.*`, `e2e-tests/**`, `node_modules/**`, outside **`coverage.include`**. Output a **numbered list** (path, %, one-line rationale). **Do not wait for approval** unless the user explicitly asked to confirm the list; proceed to the loop. ## 3. Per-file loop For **each** shortlist file, in order: ### 3a. Contextualize (agent) Read the implementation and, when useful, **callers or types** (imports, sibling modules). Note **exports**, **side effects**, **async** boundaries, **feature flags**, and **error handling**. ### 3b. Decide what to test (agent) Infer **high-value behaviors** without asking the user—for example: - Public functions / component outcomes users depend on. - Branches: success vs error, empty data, validation failures. - Integration with mocked **fetch** / **MSW** / **TanStack Query** where the module touches the network. - Stable contracts: inputs → outputs, not implementation trivia. If the file is **untestable without refactor**, add a **short skip note** in the summary and continue. ### 3c. Implement - **React:** **`web/.agents/skills/unit-test/SKILL.md`** (RTL, MSW, `findBy*`, `vi.mock`). - **Non-React:** Vitest + **`vi.mock`** as needed. - Colocate **`*.test.ts` / `*.test.tsx`** unless the package uses another established pattern. ### 3d. Verify before next file 1. `pnpm vitest --run path/to/File.test.ts` (and any related specs) until **exit 0**. 2. **`pnpm typecheck`**. Fix TypeScript issues (TanStack context, SDK types, etc.). 3. Optionally re-run coverage for the package and note improvement for that file. Keep a **running list** of every file path you **create or edit** during §3 (new **`*.test.*`**, and any production files you touch). You will pass that list to eslint in §4. You may fix obvious **ESLint** issues (e.g. **`import/order`**) during §3d when **`typecheck`** is already clean. ## 4. Lint after the loop (changed files only) When **all** shortlist iterations in §3 are finished (not after each file): 1. `cd web/packages/`. 2. From **`package.json`**, read the **`lint`** script and reuse the **same eslint flags** as the project (everything after the `eslint` command—e.g. **`--report-unused-disable-directives --max-warnings 0`** for Studio), but **replace the path glob** (e.g. `.`) with **only your tracked file paths**: ```bash pnpm exec eslint --report-unused-disable-directives --max-warnings 0 path/to/A.test.ts path/to/B.test.ts ``` 3. If the package uses **`lint:fix`** for local workflow, you may run **`pnpm exec eslint --fix ...`** with the **same file list** first, then re-run without **`--fix`** if needed to confirm **exit 0**. 4. Fix any reported issues until eslint exits **0** on that list. **Do not** run **`pnpm lint`** / **`eslint .`** on the whole package unless the user asked for a full-package lint or a changed file list is empty (then note “no files to lint”). If eslint is not available in the package, note that lint was skipped. ## 5. Stop - After **~10** files (or fewer if the shortlist was shorter), **§4 lint**, and a final sanity check. - If the user capped scope in the initial message, respect that. - End with a **short recap**: files touched, specs added/updated, any skips and why, and confirm **lint** passed. ## 6. Quick reference | Item | Notes | | ----------------- | ------------------------------------------------------------------------------------ | | Workspace root | `web/` — optional **`pnpm --filter