Memlab Capture Heap Snapshots
Capture JavaScript heap snapshots (.heapsnapshot files) for memory analysis from Chrome...
Install
npx skills add meta/memlab --skill memlab-capture-heap-snapshots -g -a claude-code -y- Repository
- meta/memlab5.1k
- Category
- Testing
- Updated
- 4d ago
- License
- MIT
- Context cost
- Lightabout 1,234 tokens
- Runs scripts
- No, instructions only
What it does
Capture JavaScript heap snapshots (.heapsnapshot files) for memory analysis from Chrome or Edge, Puppeteer, Playwright, Node.js, Electron or React Native (Hermes), including a series of snapshots that shows what grows over repeated interactions. Use when the user needs a heap snapshot, a heap dump of a Node.js process or one taken near out-of-memory, wants to measure memory growth or a suspected leak, or asks how to take snapshots in a test so memlab can analyze them.
Capture JavaScript heap snapshots
A heap snapshot is a full dump of a JavaScript heap: every object, the
references between them, and their sizes. memlab, Chrome DevTools and the
memlab MCP server all read the same .heapsnapshot format.
Decide what to capture first
- "What is using memory?" One snapshot, taken when memory is high.
- "Is something leaking?" Several snapshots from the same page or process: no reload and no restart in between, with the suspect interaction repeated between captures. A single snapshot can't show growth.
- Take snapshots from a build that matches the problem. Development builds (React development mode, hot reload, source-map tooling) hold extra memory a production build doesn't.
Browser (Chrome, Edge)
By hand: DevTools → Memory → Heap snapshot → Take snapshot, then right-click the snapshot → Save.... DevTools runs garbage collection before each snapshot. Use a profile without extensions (or a guest window): extensions add their own objects to the page's heap.
From an agent, with the memlab MCP server: start the browser with
--remote-debugging-port=9222 --user-data-dir=/tmp/chrome-memlab, then call
memlab_capture with out_dir and a label. Chrome 136 and later ignore
--remote-debugging-port on the default profile, so the separate
--user-data-dir is required; a fresh profile also has no extensions. It also captures the page's dedicated workers, which a page snapshot
never includes. list_targets: true lists the pages it can attach to, and
url_filter picks one. A saved capture can later be referred to as
checkpoint:<label> wherever a snapshot path is accepted.
From Puppeteer or Playwright test code (Chromium only): use a CDP session. See references/capture-code.md for a copy-paste helper that runs garbage collection and streams the snapshot to a file.
