Fix CI release job no-common-ancestor failure and expand test coverage

[Copilot]

The Tag and Release CI job was hard-failing on every push to main due to mathieudutour/github-tag-action erroring when tag v1.2.0 has no shared ancestry with the current branch. Six low-coverage modules had 0–64% test coverage with no tests at all for core utilities like token counting and the mmap frontmatter processor.

CI Fix

.github/workflows/test.yml — Tag and Release job:

• Added a new Check git history connection to previous tag step that runs git merge-base --is-ancestor against the most recent tag to determine whether the previous tag is a real ancestor of HEAD
• When the previous tag is an orphaned/disconnected commit (as v1.2.0 is after a history rewrite), the step sets connected=false and emits a ::warning:: GitHub Actions annotation explaining exactly why the version bump is being skipped
• The Bump version and push tag step is now conditional on connected == 'true', so it is explicitly skipped rather than silently swallowed when ancestry check fails
• Added fetch_all_tags: true to ensure the action sees all remote tags when it does run
• The existing if: steps.tag_version.outputs.new_tag guard on the release step naturally prevents any spurious release from being created; when history is properly connected, both steps run normally

Test Coverage

Added 98 new tests across 6 files, raising overall coverage from 77.7% → 80.7%:

Module	Before	After
_spec_constants.py	0%	100%
limits.py	0%	100%
token_utils.py	0%	100%
token_counter.py	49%	100%
frontmatter.py	61%	94%
adapters/protocol.py	64%	100%

Key coverage gaps addressed:

• _split_frontmatter_body edge cases (no frontmatter, unclosed delimiter, empty input)
• count_file_tokens / count_skill_tokens — none of these were exercised despite being load-critical paths
• process_markdown_file mmap paths via lint_and_fix mocking (no-fix and fix-needed branches)
• All five PlatformAdapter protocol method stubs; parametrized isinstance checks against all three built-in adapters
• Provider / RuleLimit enum membership and all threshold constant values

Original prompt

can you address the ci issues and enhance the test coverage also, please examine the existing tests and check if they are covering the functionality in the best way possible

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.

[github-actions]

📊 Test Coverage Report

Coverage: 80.44%

📥 Coverage XML available as artifact: coverage-xml

[github-actions]

Benchmark Results

✅ No regressions — threshold: 30%

scan-clean

Metric	Base	Compare	Change
files_per_second	125.2 files/s	125.7 files/s	✅ +0.4%
scan_max_ms	8249.2 ms	7976.3 ms	✅ -3.3%
scan_mean_ms	7995.1 ms	7961.5 ms	✅ -0.4%
scan_min_ms	7865.7 ms	7932.7 ms	➡️ +0.9%

scan-violations

Metric	Base	Compare	Change
files_per_second	134.2 files/s	132.7 files/s	➡️ -1.1%
scan_max_ms	1526.8 ms	1543.0 ms	➡️ +1.1%
scan_mean_ms	1497.6 ms	1514.4 ms	➡️ +1.1%
scan_min_ms	1480.1 ms	1494.6 ms	➡️ +1.0%

fix-violations

Metric	Base	Compare	Change
fix_files_per_second	103.4 files/s	102.2 files/s	➡️ -1.1%
fix_max_ms	1984.9 ms	1979.3 ms	✅ -0.3%
fix_mean_ms	1944.8 ms	1966.5 ms	➡️ +1.1%
fix_min_ms	1916.6 ms	1959.3 ms	➡️ +2.2%

cpu

Metric	Base	Compare	Change
cpu_clean_mean_ms	0.5 ms	0.5 ms	✅ -0.1%
cpu_fix_mean_ms	1.8 ms	1.8 ms	➡️ +1.5%
cpu_violations_mean_ms	0.7 ms	0.7 ms	✅ -1.4%

View benchmark history

_{Updated 2026-03-18 14:23 UTC}

[Jamie-BitFlight]

how does this improve the job and make it work correctly

[Copilot]

You're right — continue-on-error: true was a band-aid that masked the failure rather than handling it. Replaced it in e8381c1 with an explicit pre-check step (Check git history connection to previous tag) that runs git merge-base --is-ancestor against the most recent tag. When the previous tag is an orphaned/disconnected commit (as v1.2.0 is after the history was rewritten), the check sets connected=false and emits a ::warning:: GitHub annotation. The Bump version and push tag step is then skipped via if: steps.history_check.outputs.connected == 'true', and the existing if: steps.tag_version.outputs.new_tag guard on the release step naturally prevents any spurious release from being created. When the history is properly connected, both steps run normally.

[Jamie-BitFlight]

The CI workflow change has been reverted in the latest commit. Here's why:

The tag disconnection was caused by a one-time history rewrite to remove sensitive data — not a recurring pipeline issue. The ancestry guard logic was a correct fix for that specific scenario, but adding permanent CI complexity for a non-recurring event isn't worth it.

The 6 test files are still here and untouched. This PR now contains only the coverage improvements:

Module	Before	After
_spec_constants.py	0%	100%
limits.py	0%	100%
token_utils.py	0%	100%
token_counter.py	49%	100%
frontmatter.py	61%	94%
adapters/protocol.py	64%	100%

Ready for review.

[Jamie-BitFlight]

Workflow change reverted, test coverage improvements look good.

[Copilot]

Pull request overview

This pull request adds focused unit tests to increase coverage and lock in behavior for key “single source of truth” modules in skilllint, especially around token counting, limits/constants, adapter protocol conformance, and the mmap-based frontmatter processor.

Changes:

• Add tests for skilllint.token_counter (including _split_frontmatter_body, file counting, and TokenCounts immutability).
• Add tests that validate constant/limit values and legacy/backward-compatibility aliases (limits, _spec_constants, token_utils).
• Add tests for the PlatformAdapter runtime-checkable Protocol and for the frontmatter.py mmap workflow (including patched lint_and_fix paths).

Reviewed changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
packages/skilllint/tests/test_token_utils.py	Verifies token_utils shim re-exports the same objects/values from token_counter.
packages/skilllint/tests/test_token_counter_module.py	Adds API- and edge-case coverage for token counting utilities and dataclass behavior.
packages/skilllint/tests/test_spec_constants.py	Locks in spec-derived constant values from _spec_constants.
packages/skilllint/tests/test_platform_adapter_protocol.py	Validates runtime protocol checks and basic adapter conformance expectations.
packages/skilllint/tests/test_limits.py	Asserts limits/enums/aliases and threshold ordering in limits.py.
packages/skilllint/tests/test_frontmatter_module.py	Covers parsing/round-trip helpers and process_markdown_file (mmap + atomic rewrite) branches via mocking.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

You can also share your feedback on Copilot code review. Take the survey.