| name | gstack-browse-reference |
| description | Hard-won working patterns for the gstack `browse` CLI (headed-Chromium automation used for logged-in verification and scraping). Load when driving gstack browse — correct binary path, the `js` vs `eval` gotcha, key names, and `@e##` element refs over CSS. Saves ~1,200 tokens/session of re-discovery. |
| allowed-tools | Bash, Read |
gstack browse CLI — working reference
The gstack browse binary drives a headed Chromium with a persistent profile (used for logged-in automation and verification). These are the patterns that actually work; the obvious alternatives fail.
Binary path is fixed — call it directly
B="$HOME/.claude/skills/gstack/browse/dist/browse"
"$B" snapshot -i
Skip the git rev-parse discovery dance — the path never moves.
Use js, never eval
"$B" js "document.querySelector('#price').textContent"
$B eval "..." fails with "File not found". For any in-page DOM script, use the js subcommand.
Key names
- Press Enter with
$B press "Enter" — not "Return" (fails).
Prefer @e## element refs over CSS selectors
Run $B snapshot -i to get an interactive snapshot with stable refs like @e42, then act on them:
"$B" snapshot -i
"$B" click @e42
CSS selectors often match multiple elements or time out. @e## refs are unambiguous.
Heavy-use stability
Headed Chromium drops its session / navigates to about:blank after ~150s of heavy use. Keep in-page loops under ~120s and reconnect (browse connect) on crash — the persistent-profile login is restored on reconnect.
