test-coverage-improver_skill

Overview

This skill improves test coverage in the OpenAI Agents JS monorepo by running the coverage suite, inspecting artifacts, and proposing targeted tests. It identifies low-coverage files and branches, prioritizes high-impact areas, and asks for explicit user approval before writing tests or changing code.

How this skill works

Run CI=1 pnpm test:coverage to regenerate coverage artifacts, then parse coverage/coverage-summary.json (or coverage-final.json) and lcov data to spot gaps. The skill lists lowest files, branches under threshold, and uncovered lines, drafts test scenarios to exercise missing paths, and presents them to the user for confirmation. After approval it implements tests in the appropriate packages, reruns coverage, and verifies improvements.

When to use it

• Coverage thresholds are failing or regressions are suspected
• A release needs stronger quality gates for core packages
• New or changed code shows 0% or low coverage
• You want a prioritized plan for where tests will have highest impact
• Before merging risky bug fixes involving error handling, retries, or concurrency

Best practices

• Prefer coverage/coverage-summary.json for quick file-level totals; use lcov only for branch and line drilling
• Prioritize public APIs and shared utilities in packages/*/src before examples or docs
• Target uncovered branches: error paths, boundary values, optional flags, and cancellations
• Write unit tests next to the package (packages/<pkg>/test/*.test.ts) and avoid flaky async timing
• Present a numbered, concise list of proposed tests and expected coverage gains; require explicit user approval

Example use cases

• Detecting which packages dropped below 80% branch coverage after a refactor and proposing specific tests to restore it
• Adding tests to exercise error-handling and retry logic in a shared network utility identified as low-coverage
• Creating boundary-case inputs to cover combinational logic in an agent orchestration module
• Rerunning coverage after implementing tests to verify the reported improvement and remaining hotspots