Claude Skills by SSHdotCodes

Verify a UI change by driving a real browser at real viewport sizes and emitting a machine-gradable report. Use when a Bandaid probe reports "no current browser report", when a goal's acceptance criteria mention layout, responsiveness, mobile, overflow, or a page rendering correctly, or when the user asks whether a change actually looks right in a browser rather than whether its unit tests pass. Produces .bandaid/artifacts/browser/report.json, which the browser probe grades by exit code. Not ...

Put a backend under sustained load and grade the result against a budget written beforehand. Use when a Bandaid load probe abstains for want of a budgets file, when a goal's criteria mention throughput, latency, p95/p99, concurrency, or holding up under load, or when a change touches a request path and the user wants to know it still performs. Produces .bandaid/load-budgets.json and, on a run, a latency report the probe grades by exit code. Not for microbenchmarks and not for capacity planning.

Check what this work introduced for security defects — credentials written into the repository, new dependency advisories, and injection or authorization gaps reachable from untrusted input in the diff. Use when a goal touches auth, input handling, secrets, dependencies, or a request path, when a Bandaid security probe reports a finding, or when the user asks whether a change is safe to ship. Grades what the diff added, not what the repository already contained. Not a substitute for a real au...

Hunt for the same class of defect elsewhere in the repository after one instance has been found and fixed, where every finding must ship a command that fails right now. Use when a bug has just been fixed and the same shape could exist in other call sites, when the user asks whether a defect appears anywhere else, or when a goal's criteria mention a repo-wide fix. Produces .bandaid/artifacts/sweep/findings.json with executable reproductions. Not for a general "find bugs" pass, which is expensi...