test-sweep

name	test-sweep
description	Run all test suites (unit, integration, API, E2E) and aggregate results into a summary report. Use after completing execution slices or before the Review Gate.
argument-hint	[optional: specific test category to run]
tools	["run_in_terminal","read_file","forge_sweep"]

Test Sweep Skill

Trigger

"Run all tests" / "Full test sweep" / "Check test health"

Steps

1. Unit Tests

go test ./internal/... -v -count=1 -tags=unit 2>&1 | tee TestResults/unit.txt

Conditional: Unit Test Failure

If unit tests fail → skip integration/E2E tests, go directly to Report.

2. Integration Tests

# Requires database running (Docker or local)
go test ./tests/integration/... -v -count=1 -tags=integration 2>&1 | tee TestResults/integration.txt

3. API Tests

go test ./tests/api/... -v -count=1 -tags=api 2>&1 | tee TestResults/api.txt

4. E2E Tests (if available)

go test ./tests/e2e/... -v -count=1 -tags=e2e 2>&1 | tee TestResults/e2e.txt

5. Race Detection

go test ./... -race -count=1 2>&1 | tee TestResults/race.txt

6. Completeness Scan

Use the forge_sweep MCP tool to scan for TODO/FIXME/stub markers in the codebase.

7. Report

Aggregate results:

✅ Unit:        X passed, Y failed
✅ Integration: X passed, Y failed
✅ API:         X passed, Y failed
✅ E2E:         X passed, Y failed
✅ Race:        No data races detected
✅ Sweep:       N markers (TODO/FIXME/stub)
──────────────────────────────────────
Total:          X passed, Y failed

On Failure

• Show failed test names and error messages
• Read the failing test source to diagnose
• Check for race conditions with -race flag
• Suggest fixes (ask before applying)

Temper Guards

Shortcut	Why It Breaks
"Skipped tests are probably flaky"	Skipped tests hide real regressions. Each skip needs a documented reason and a linked issue.
"80% coverage is good enough"	Coverage thresholds prevent ratcheting down. If baseline is 85%, dropping to 80% means new code is untested.
"Integration tests cover the unit tests"	Integration tests are slow and brittle. Unit tests catch logic errors in milliseconds, not minutes.
"I'll fix the failing test later"	Broken tests normalize failure. The suite must be green before any code ships.

Warning Signs

• Skipped tests without documented reason — skip annotations present without explanation
• Coverage decreased from baseline — new code merged without maintaining coverage threshold
• No test output included in report — tests "passed" but no actual results pasted
• Test suite not run before PR — commit pushed without running the full sweep first
• Flaky test dismissed — intermittent failure ignored instead of investigated

Exit Proof

After completing this skill, confirm:

• All suites executed — go test ./... completes
• Zero unexplained failures (every failure has a documented reason)
• Coverage report generated — go test -coverprofile=coverage.out ./...
• Coverage not decreased from baseline
• forge_sweep found zero production code markers (TODO/FIXME/stub)

Persistent Memory (if OpenBrain is configured)

• Before running tests: search_thoughts("test failures", project: "<YOUR PROJECT NAME>", created_by: "copilot-vscode", type: "bug") — load known flaky tests, recurring failures, and environment-specific issues
• After test sweep: capture_thought("Test sweep: <N passed, N failed — key failure patterns>", project: "<YOUR PROJECT NAME>", created_by: "copilot-vscode", source: "skill-test-sweep") — persist failure patterns and flaky test discoveries