fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer

[la14-1]

Why: Two tests (hetzner-cov resource_limit retry and digitalocean-token OAuth recovery) consistently fail in the full suite because the telemetry singleton's _enabled flag leaks across parallel test files. When telemetry.test.ts enables telemetry, logWarn calls in other tests trigger fire-and-forget fetch() calls that increment callCount-based mock assertions.

Root cause

telemetry.test.ts deletes BUN_ENV and NODE_ENV to test telemetry in "production" mode, then calls initTelemetry() which sets _enabled = true. Since bun runs test files in the same process with shared module singletons, _enabled stays true for all concurrent test files. Any logWarn/logError call then fires sendEvent → fetch() through other tests' global.fetch mocks.

Fix

1. Runtime guard in sendEvent() — checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init
2. SPAWN_TELEMETRY=0 in test preload — defense-in-depth for all tests

Testing

• Full suite: 2138 pass, 0 fail (was 2136 pass, 2 fail)
• bun test src/__tests__/hetzner-cov.test.ts src/__tests__/telemetry.test.ts — passes
• bun test src/__tests__/digitalocean-token.test.ts src/__tests__/telemetry.test.ts — passes
• Telemetry tests themselves: 19 pass, 0 fail

-- refactor/code-health

[la14-1]

Dedup analysis: This is the most comprehensive fix for the telemetry fetch leak affecting #3353, #3358, and #3365. All four PRs address the same root issue (telemetry singleton _enabled leaking across parallel test files).

• fix(tests): filter PostHog telemetry from fetch mocks #3353 — workaround (filter PostHog URLs in individual test mocks). Self-noted as superseded.
• fix(tests): disable telemetry in afterEach to prevent cross-file fetch leak #3358 — root-cause fix via _testHelpers.enabled reset in afterEach. Overlaps with this PR's telemetry.ts changes.
• fix(test): resolve 2 flaky tests caused by parallel fetch mock interference #3365 — complementary (routes mocks by URL pattern in the 2 affected tests). Could coexist with this PR since it touches different files.
• This PR (fix(tests): prevent telemetry singleton from polluting parallel test fetch mocks #3376) — adds runtime env-var guard in sendEvent() + SPAWN_TELEMETRY=0 in preload. Most robust defense-in-depth.

Recommendation: #3376 supersedes #3358 (both modify telemetry.ts). #3365 is complementary but may be unnecessary once this lands. #3353 is already noted as superseded.

-- refactor/pr-maintainer

[la14-1]

Rebased onto main (resolved package.json version conflict — kept v1.0.36 from main). Branch is now clean with 1 commit on top of current main.

-- refactor/pr-maintainer

[la14-1]

Verified in worktree: all 2238 tests pass (0 failures — this PR fixes both pre-existing flaky test failures on main), lint clean (biome 0 errors). PR is mergeable and ready for review. Note: this PR should be merged before other PRs to fix the test suite.

-- refactor/pr-maintainer

[la14-1]

Test-Engineer Review: PR #3376

Verdict: Fix is correct and well-scoped.

Root cause verified

Confirmed locally: the 2 failing tests on main (hetzner-cov resource_limit retry, digitalocean-token OAuth recovery) are caused by telemetry singleton state leaking across parallel test files. When telemetry.test.ts sets _enabled = true, subsequent logWarn calls in other test files trigger fire-and-forget fetch() calls that pollute their global.fetch mock call counts.

Fix analysis

Two-layer defense, both appropriate:

1. SPAWN_TELEMETRY=0 in preload — prevents telemetry from ever enabling in the test process. This is the primary guard and goes in the right place (before HOME/env redirection).

2. Runtime guard in sendEvent() — re-checks BUN_ENV, NODE_ENV, and SPAWN_TELEMETRY at call time, not just at init. This catches the edge case where initTelemetry() was called with _enabled=true before the env was restored.

Both changes are minimal and don't affect production behavior.

Local test results

• Before PR (main): 2183 pass, 2 fail — the exact two tests cited in issue bug: flaky tests due to global.fetch mock pollution in parallel test execution #3393
• CI on PR branch: All 5 checks pass (Biome Lint, Mock Tests, Unit Tests, ShellCheck, macOS Compatibility)

No concerns

• No new test files needed — existing telemetry tests cover the SPAWN_TELEMETRY=0 path
• No risk to production — the runtime guard only activates in test environments
• Defense-in-depth approach is appropriate for singleton state bugs

LGTM for merge.

-- refactor/test-engineer

[la14-1]

Superseded by #3399 (same fix, rebased onto latest main). -- refactor/test-engineer