By name: browser-interaction description: "Fetch and inspect JavaScript-heavy pages with Playwright in a Daytona browser-capable snapshot."
Browser Interaction (Daytona)
Use when plain fetch / urllib returns empty or incomplete HTML (SPAs, dynamic docs).
When to use what
| Approach | Use when |
|---|---|
Python urllib / requests |
Static HTML, APIs, raw text downloads |
document_text REPL variable |
fleet-rlm already fetched the URL into RLM |
| Playwright (browser snapshot) | Client-rendered pages, need visible text or screenshots |
fleet-rlm selects a browser-capable Daytona snapshot when this skill is active.
Playwright patterns (sandbox)
# After navigation, always print bounded excerpts for the RLM loop.
print(page.title())
print(page.inner_text("body")[:4000])
- Wait for network idle or a known selector before extracting text.
- Capture screenshots to the volume
artifacts/path for evidence. - Bound extracted text; use
llm_queryon excerpts, not full page dumps.
Safety
- Do not exfiltrate credentials, cookies, or environment variables.
- Do not automate logins unless the user explicitly provided test credentials in-scope.
- Cap page size and iteration count; prefer section samples over full DOM serializations.
Volume access
Bundled instructions are seeded to {volume}/skills/system/browser-interaction.md.
Call load_skill("browser-interaction") from the REPL when the volume is mounted.