Nothing to hand? Load the — a class that adds a listener on every search — the , which disposes most of what it creates but not all, or the , which should come back with nothing to fix.
Paste the code
A class, a widget, a controller, a service, a test file, or all of it together. The instant prescan reads it for free while you type and lists what it mechanically found: raw addEventListener calls, on* property handlers, listeners bound to window or document, _register() inside a method that runs more than once, a DisposableStore never tied to a model, intervals with no clearInterval, observers with no disconnect, AbortControllers never aborted, test suites with no leak guard — plus the disposable classes and the listener counts in your paste.
The AI audits it
A clean/patch/leaking verdict; severity-ranked findings, each quoting the code it concerns, saying how fast the leak grows, and carrying corrected code; a twelve-item disposable checklist scored pass, fail or not-observable; and the six-step walk from the source skill — DOM listeners, one-time events, repeated method calls, model-tied stores, resource pools, test guards. Fixes are written for the runtime you picked, so a plain browser paste gets removeEventListener and AbortController rather than editor-only helpers. Every prescan hit is confirmed or explicitly set aside, including the false positives.
Take the fixed code and go
Your own code with the lifetimes corrected — same classes, same behaviour, listeners tracked, stores tied to what they follow, timers cleared. Every prescan hit and every finding whose quoted code can be found carries its line number and jumps to it in your paste, so nothing has to be hunted down by hand. See exactly what changed as a line diff, take it as a file or as a unified .patch written against the real path from your paste so git apply lands it, and read the free re-scan that says which detected leaks the rewrite actually clears. Copy review comment gives you the PR-thread version — verdict, the high and medium findings with line numbers, the summary, no rewrite body. Then fix it your way and run it again, or press Re-audit the rewrite: either way the second run is compared against the first, item by item — what got fixed, what broke, which findings survived. Audits are saved to your SkillSafe account and searchable there, so they follow you to your next device; signed out they stay on this one until you sign in.
Derived from the @microsoft/memory-leak-audit skill (microsoft/vscode, MIT).