fix: exclude sleep.ts from coverage and add pre-push hook

[adawalli]

Summary

• Fixes CI failure where bun test --coverage exits 1 despite all tests passing
• Adds a pre-push hook so coverage failures are caught locally before CI

Root cause

src/sleep.ts is always mocked via mock.module("../src/sleep", ...) in tests. On Linux (CI), bun reports 0% coverage for the real module, failing the 90% line coverage threshold. Locally on macOS the coverage tracking behaves differently, so it wasn't caught pre-push.

src/stable-wait.ts is already in the ignore list for the same reason - src/sleep.ts belongs there too.

Changes

• bunfig.toml - add src/sleep.ts to coveragePathIgnorePatterns
• .husky/pre-push - run bun test --coverage before every push

[claude]

Claude finished @adawalli's task in 1m 45s —— View job

---

PR Review

• Gather context
• Read changed files
• Review code quality and correctness
• Post review feedback

---

LGTM overall. This is a minimal, well-scoped fix with solid reasoning.

bunfig.toml

Adding src/sleep.ts to coveragePathIgnorePatterns is the right call. The module is a one-liner setTimeout wrapper that is always replaced by mock.module("../src/sleep", ...) in both test/stable-wait.test.ts and test/llm.test.ts — the real implementation never runs. This is exactly the same situation as src/stable-wait.ts, which was already excluded. The fix is consistent with the existing pattern.

.husky/pre-push

The hook correctly mirrors the CI command (bun test --coverage), which is the right approach to catch platform-specific coverage differences before they hit CI. One minor note: the hook is missing a shebang line (left an inline suggestion). The pre-commit hook also omits one, so this is consistent with existing project convention, but adding #!/bin/sh is safer across environments.

No functional issues. The tradeoff of running full coverage on every push is intentional and appropriate given this was a CI-only failure previously.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[claude]

Claude finished @adawalli's task in 57s —— View job

---

PR Review

• Gather context
• Read changed files
• Review code quality and correctness
• Post review feedback

---

LGTM. This is a clean, minimal fix with good reasoning.

bunfig.toml

Adding src/sleep.ts to coveragePathIgnorePatterns is correct. The module is always replaced by mock.module("../src/sleep", ...) in both test/stable-wait.test.ts and test/llm.test.ts — the real implementation never executes under test. This is identical to the existing src/stable-wait.ts exclusion, so the change is consistent with the established pattern.

The current ignore list:

coveragePathIgnorePatterns = ["src/index.ts", "src/llm.ts", "src/stable-wait.ts", "src/sleep.ts"]

All four entries have legitimate reasons to be excluded (entry point, always-mocked modules). No concerns here.

.husky/pre-push

The hook correctly includes a shebang (#!/bin/sh) and mirrors the CI coverage command exactly — the right approach to surface platform-specific coverage differences before they reach CI. The pre-commit hook omits a shebang (pre-existing inconsistency), so this hook is actually more correct than its sibling.

One tradeoff worth noting: running bun test --coverage on every push adds overhead to the push workflow. Given the PR's stated motivation (a CI-only failure that was invisible locally), this is an acceptable and intentional tradeoff.

No issues found.