What problem does it solve? Repositories with no test suite cannot be changed safely by an agent or a person, because nothing detects unintended behaviour change. This Skill fills that gap by ranking the riskiest units and writing real tests that pin current behaviour, each proved able to fail before it is kept. ## Core Features & Use Cases - Risk-ranked test authoring: A bundled offline ranker scores units by churn and blast radius across Python, node/TypeScript, Go, and Rust, then triages each into four testability tiers. - Red-to-green proof loop: Every test is written with a deliberately wrong assertion, required to fail, corrected, and required to pass, so the assertion provably binds to real output. - Runtime I/O guards: Per-stack guards (pytest plugin, node --require preload, go test -overlay wrapper, cargo preload hook) enforce that no generated test performs real I/O, with tier-aware allow lists for controlled boundaries. - Honest reporting: Suspected bugs are pinned and reported rather than blessed, and untestable code becomes a ranked seam list handed to clean-code. - Use Case: Before a migration on a legacy Go module with no _test.go files, run the ranker, confirm the top 10 units, and let the agent build a safety net of characterization tests one unit at a time. ## Quick Start Ask the agent to add a safety net to this repository before we refactor it, then confirm the ranked list of units it proposes to test.
