Bug: claude-md-coverage-thresholds-and-toolchain-command-corrections #870

• Work Mode: full-bug

Summary

CLAUDE.md names a test command the repository deliberately never runs (#828), states coverage thresholds that the maintainer has now settled differently from two rules files (#563, decision recorded 2026-09-11), and cites .globalconfig twice as an analyzer-severity source when that file does not exist (#727 sub-finding 5). All three are documentation-only corrections to one file this repository owns.

Environment

• OS/version: not applicable (documentation)
• Python version: not applicable
• Command/flags used: CLAUDE.md lines 390 and 408 (vstest.console.exe <test-assembly-paths> /EnableCodeCoverage); UT2 coverage section; C#1 item 2 and C#7
• Data source or fixture: scripts/vscode/Invoke-MSTestWithCoverage.ps1, scripts/vscode/Invoke-MSTestWithCoverage.Threshold.ps1, .editorconfig

Steps to Reproduce

1. Read CUT3 step 4 and the "C# Toolchain" step 4: both name vstest.console.exe ... /EnableCodeCoverage. Read Invoke-MSTestWithCoverage.ps1 lines 19 to 26: the implemented route is an outer dotnet-coverage collect wrapping an inner vstest that is never given /EnableCodeCoverage, because the built-in collector conflicts with the outer instrumentation.
2. Read UT2: "Repository-wide line coverage must remain >= 80%" on a testable denominator, no branch floor stated. Compare .claude/rules/general-unit-test.md (85 line / 75 branch). The maintainer decision on Coverage threshold contradiction remains: CLAUDE.md/csharp.md say 80%, general-unit-test.md/quality-tiers.md say 85%/75%, and two live gates disagree #563 is 80 line, 75 branch, PowerShell 80 line, recorded in CLAUDE.md alone.
3. Search CLAUDE.md for .globalconfig: two references (C#1 item 2, C#7). Test-Path .globalconfig is false; .editorconfig is the actual severity source.

Expected Behavior

• CUT3 step 4 and the C# Toolchain step 4 name the route actually run: scripts/vscode/Invoke-MSTestWithCoverage.ps1 (or the VS Code task that wraps it), with a one-line note that /EnableCodeCoverage is deliberately omitted from the inner vstest call.
• UT2 states: C# line >= 80%, C# branch >= 75%, PowerShell line >= 80% (Pester measures no branch), new code >= 90%, the three exemption classes unchanged, and one sentence recording that these figures were settled on 2026-09-11 under Coverage threshold contradiction remains: CLAUDE.md/csharp.md say 80%, general-unit-test.md/quality-tiers.md say 85%/75%, and two live gates disagree #563 and that the .claude/rules/ figures are upstream-owned and not authoritative here.
• Both .globalconfig references read .editorconfig.

Actual Behavior

Reviewers citing CUT3 step 4 issue false PARTIAL verdicts (#809 AC6, and every later item citing it); the 80-versus-85 divergence forces a FAIL row on every review at the measured 84.56%; the .globalconfig citations point at a file that does not exist.

Logs / Screenshots

• Attached minimal logs or screenshot
• Snippet: none; documentation finding.

Impact / Severity

• Blocker
• High
• Medium
• Low

Source

From: docs/features/potential/2026-09-11-claude-md-coverage-thresholds-and-toolchain-command-corrections.md