Summary

Adds optional test_command: input to the coverage-floor.yml reusable workflow. When set, it replaces the default language-specific install + pytest/go-test/npm-test invocation; the caller's command is expected to produce the same coverage artifact the workflow already parses (coverage.json / cov.out / coverage/coverage-summary.json).

Mirrors the same input on tests-runner.yml.

Concrete trigger

whois-api-llc/wxa_marketing (today). Layered Python project:

• Root pyproject.toml defines the marketing-engine package with httpx, pytest-asyncio, aiosqlite in [project.optional-dependencies].test
• Tests live under backend/ and import marketing_engine.* (and httpx in conftest)
• backend/requirements.txt only has the test runner deps

The default Python path falls into [[ -f requirements.txt ]] → pip install -r requirements.txt && pytest from backend/, which doesn't install the engine package itself, so from httpx import AsyncClient in conftest.py fails with ModuleNotFoundError. Same failure mode predicted in CLAUDE.md lessons ("Coverage-floor reusable's naïve install ... misses repo-specific test-environment quirks").

After this lands, the caller can write:

with:
  language: python
  working_directory: backend
  test_command: |
    python -m venv .venv
    .venv/bin/pip install -q --upgrade pip
    .venv/bin/pip install -q -e "..[test]" pytest-cov
    .venv/bin/pytest --cov --cov-report=json --quiet || true

Same pattern already de-risked on tests-runner.yml — in flight today as whois-api-llc/wxa_marketing#25.

Design

Per-branch short-circuit inside the existing case "$LANG_TYPE" block rather than wrapping the whole case. Reasons:

1. Preserves the per-language coverage-file existence check — the workflow still errors clearly if the caller's command produces no coverage file ("pytest-cov did not produce coverage.json"). The check is the contract.
2. Preserves the per-language MEASURED jq expression — different runners emit different JSON shapes (.totals.percent_covered vs .total.lines.pct vs Go's text format).
3. Minimal diff — 3 short-circuit blocks (4 lines each), no restructuring.

Test plan

• python -m yaml.safe_load clean
• bash -n clean on the measure step's embedded run script
• All three case branches short-circuit when INPUT_TEST_COMMAND is non-empty
• Coverage-file existence checks still enforce the right artifact gets produced
• Defaults preserved: empty test_command (the default) keeps existing auto-detection behavior unchanged on every repo that doesn't pass it
• Self-test on this repo (.coverage-floor at root, selftest/sample.py at 99% target) stays green

Auto-merge rationale: in-tree change to .github/workflows/coverage-floor.yml — high-risk surface per fleet policy (.github/workflows/**). Diff is purely additive (1 new input declaration, 3 short-circuit blocks); no permission/secret changes; no behavior change for callers that don't set the input. Per the cheat-sheet self-test: worst case is "callers measure coverage differently for one cycle, git revert and we're back" — auto-merge territory. Bot's regex decides.

Codex pre-review: 44-line additive diff in a single existing step. Skipping per the <50-line carve-out.

🤖 Generated with Claude Code