Removes kilo, kimi, kimi-code, copilot, windsurf, augment, trae, qwen, hermes, cline, codebuddy and pi end to end: capability descriptors, installer branches and converters (bin/install.js 14.9k -> 11.2k lines), TypeScript converters, hook surfaces and runtime homes, review lanes qwen/kimi-code, the two pi migrations, Kimi payload normalization in the hook guards, dead hostBehaviors vocabulary, launcher home probes, fixtures, runtime-specific tests and the prose that presented them as supported. Installer output for the six kept runtimes is byte-identical to before the prune. The Kimi tool-vocabulary tests in workflow-guard, read-guard and read-injection-scanner are left in place pending a decision.
5.5 KiB
How to resolve an ESLint glob-coverage finding
Goal: Turn a lint-eslint-glob-coverage failure into either real lint coverage or a recorded, reasoned exemption — so a source file can never again sit in the tree matching no rule at all, linted by nothing, while eslint . exits 0.
Prerequisites: A failing tests/eslint-glob-coverage.test.cjs, or a non-zero node scripts/lint-eslint-glob-coverage.cjs. You do not invoke the guard separately in CI — it runs as part of the test suite.
Why this guard exists
ESLint assigns rules by files: glob. A file matching no glob is not "linted and clean" — it is not linted at all, and ESLint reports nothing about it. There is no warning, no summary line, no exit code. The failure is perfectly silent.
That is not hypothetical. Before this guard, 62 tracked source files resolved to zero rules, including all 26 files under hooks/ — the enforcement machinery itself. One earlier escape hid a real Windows portability defect (a bare npm invocation without { shell: true }) in a shared test helper: a rule the repo bans as an error in sibling files, surviving purely because the helper was unreachable by the glob.
The guard checks rule reachability, not rule severity. A file covered entirely by warn rules passes here; that is a different gate (#1885 F17, --max-warnings 0).
Read a finding
Each violation names one path and one kind:
kind |
What it means |
|---|---|
uncovered |
The file matches no files: glob. Nothing lints it. |
allowlist_missing_reason |
An allowlist entry has no reason key. |
allowlist_empty_reason |
An allowlist entry's reason is empty or whitespace. |
allowlist_stale |
An allowlisted file now resolves to rules — the exemption is obsolete. |
allowlist_missing_path |
An allowlist entry names a path that is no longer tracked. |
allowlist_duplicate |
The same path is listed twice (usually a merge artifact). |
tracked_count_below_floor |
Fewer than 500 tracked source files were found — the guard refuses to report "clean" from an obviously broken file list. |
git_failed |
git ls-files failed or timed out. The guard degrades to a failure rather than a vacuous pass. |
An uncovered finding has exactly two legitimate resolutions. Pick the first one unless you can write down a permanent reason for the second.
Cover it — bring the file under a rule block
Choose this whenever the file is first-party source that should be linted. This is the correct resolution in almost every case, and it is what closed 56 of the original 62.
Open eslint.config.mjs and add the path to the files: array of the block whose ruleset fits:
- CommonJS Node code (
scripts/,eslint-rules/,bin/lib/,examples/,vscode/) → the CommonJS Node block. - Test code → the
tests/**/*.cjsblock. - TypeScript sources → the
src/**/*.ctsblock.
Then run npx eslint . and fix whatever it surfaces. Do not downgrade a rule's severity to make a new file pass — that converts a real finding into a silent one, which is the failure this guard exists to end.
If the file needs a genuinely different ruleset, add a new block with a comment saying why. hooks/** is the worked example: it is covered like other CommonJS Node code, but with n/no-process-exit deliberately off, because a hook's entire contract is its exit code and several of its exits are load-bearing (a setTimeout stdin guard where nothing else would terminate the process). That comment names the evidence, so the next reader does not have to re-derive it.
Allowlist it — record a permanent reason
Choose this only when the file must never be linted, and say why. Add an entry to scripts/lint-eslint-glob-coverage.allowlist.json:
{
"path": "tests/fixtures/brand-typing/bad-calibrated-as-sample-basis.cts",
"reason": "Deliberate MUST-NOT-COMPILE type-error fixture (#3059): type-aware linting would fail by design; it exists to prove the compiler rejects it."
}
The reason is structurally required — an entry with a missing, empty, or whitespace-only reason fails the guard. That is deliberate. The easy way to make a coverage metric look good is to add an exemption, so an exemption has to cost you a sentence you are willing to sign.
⚠️ "Not linted yet" is not a reason. An allowlist entry meaning we will get to it is a TODO wearing an exemption's badge, and it will outlive everyone who remembers it. If the file should eventually be linted, lint it now.
A file that is explicitly ignored — the msd-core/bin/lib/** tsc artifacts under ADR-457 — needs no allowlist entry. Ignored is a recorded decision; unmatched is an accident. The guard already tells them apart.
Note that bin/install.js also needs no entry: it resolves to two rules under ADR-1703's deliberately minimal coverage, which is non-empty and therefore passes.
Delete a stale entry
allowlist_stale means an allowlisted file now resolves to rules — someone widened a glob and the exemption became dead weight. Delete the entry. The allowlist only ratchets down; it can never quietly accumulate.
Verify
node scripts/lint-eslint-glob-coverage.cjs
A pass prints the count it actually checked:
ok lint-eslint-glob-coverage: 1175 tracked source file(s), 0 escapes
Read that number. If it is implausibly small, the guard's floor should have caught it — but a count that merely shrank is worth a second look before you trust the "0 escapes".