exploratory-testinglisted

# exploratory-testing Learn how the feature actually behaves before choosing automation. This skill may edit test code. It never changes expected behaviour to match an implementation. The only permitted product-code edit is the temporary one-line P0 sensitivity injection below, on a local working copy, reverted before anything else. Read repository instructions and the current Test Contract at `.harness-ship/test-contract.md` first. Exploration may proceed against a draft, but do not author automated expectations until the relevant behaviour is user-approved. ## Candidate boundary Use a runnable **non-production** surface: local, preview, simulator, or QA. Record: - full source SHA, or the candidate identifier where the contract's **Release surface owner** is not this repository; - artifact/build/environment revision and how it maps to that SHA or identifier; - access path, fixtures, permissions, and test data; and - evidence destination. Without trustworthy provenance, observations may guide investigation but cannot become release evidence. Never seed synthetic data into production. ## Pass A — black-box exploration Before reading existing test code, read only the approved behaviour, feature intent, and user-facing surface. Exercise a bounded set of paths chosen by risk: - the P0 happy path; - invalid, empty, boundary, and permission states; - interruption, retry, recovery, or state transitions where applicable; and - interactions with the nearest external boundar