chore: import upstream snapshot with attribution

cli-tool/components/loops/engineering/test-coverage-loop.md

@@ -0,0 +1,42 @@
+---
+name: test-coverage-loop
+description: Adds meaningful tests pass after pass until the suite reaches a target coverage threshold and stays green.
+category: engineering
+interval: 20m
+stop-condition: Coverage meets or exceeds the target (e.g. 90%) with all tests passing and no flaky additions.
+components: [agent:performance-testing/test-automator, command:testing/test-coverage, command:testing/generate-tests]
+tags: [testing, coverage, quality, loop]
+---
+
+# Test Coverage Loop
+
+> **Loop Engineering** — a recurring goal that drives the agent to grow real test coverage incrementally instead of in one risky batch.
+
+## 🎯 Goal
+Raise test coverage toward a defined target by adding **meaningful** tests for the least-covered, highest-risk code first — never gaming the metric with empty assertions.
+
+## ⏱️ Schedule
+Suggested interval: `20m`.
+
+## ▶️ Run it
+```
+/loop 20m "Measure current test coverage. Pick the lowest-covered, highest-risk module and add real, behavior-focused tests for it. Re-run the suite. Continue until coverage is at least 90% with everything passing."
+```
+
+## 🔁 Iteration steps
+1. **Perceive** — run the coverage report and rank uncovered code by risk.
+2. **Reason** — identify branches, error paths and edge cases that lack tests.
+3. **Plan** — choose one module/area to cover this pass.
+4. **Act** — generate tests with `test-automator` + `/generate-tests`.
+5. **Observe** — re-run `/test-coverage`; if the suite is green and coverage rose, continue; if flaky, fix before moving on.
+
+## 🛑 Stopping condition
+Coverage ≥ target, full suite passes, and no newly-added test is flaky.
+
+## 🧩 Referenced components
+- `agent:performance-testing/test-automator` — authors the tests.
+- `command:testing/test-coverage` — measures progress each pass.
+- `command:testing/generate-tests` — scaffolds new test cases.
+
+## 💡 Example
+Starting at 61%, the loop targets the untested error branches of the payment module first, climbing a few points per pass until it crosses 90% with a green suite.