pi-browser-use 0.3.1 → 0.3.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "pi-browser-use",
-  "version": "0.3.1",
+  "version": "0.3.3",
   "description": "Opinionated browser-use for the Pi coding agent, powered by chrome-devtools-mcp (not Playwright). Fresh headless by default, authenticated persistent profile opt-in, CLI-first policy bundled.",
   "keywords": [
     "pi-package",

package/skills/browser-policy/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: browser-policy
-description: Browser-use policy for Pi agents. Use before any browser_* tool call: prefer CLIs and APIs over browser automation, default to fresh headless sessions, escalate to the authenticated profile only on login walls, and never steal user focus.
+description: "Browser-use policy for Pi agents. Use before any browser_* tool call. Prefer CLIs and APIs over browser automation, default to fresh headless sessions, escalate to the authenticated profile only on login walls, and never steal user focus."
 ---
 
 # Browser Policy

package/skills/playwright-handoff/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: playwright-handoff
-description: Decide when browser automation has hit its limits and hand off to a repo Playwright run instead. Use when a task needs request mocking, assertions, traces, repeatability, or CI parity that browser_* tools cannot provide.
+description: "Decide when browser automation has hit its limits and hand off to a repo Playwright run instead. Use when a task needs request mocking, assertions, traces, repeatability, or CI parity that browser_* tools cannot provide."
 ---
 
 # Playwright Handoff
@@ -69,6 +69,10 @@ PLAYWRIGHT_HEADLESS=1 bunx playwright test --config=playwright.config.ts
 
 Keep headed runs for human-watched perf validation only.
 
+## Python alternative
+
+For Python-first repos, the global `webapp-testing` skill covers the same ground with `sync_playwright()` scripts and a multi-server lifecycle helper — same handoff rule applies regardless of language.
+
 ## Reporting back
 
 Summarize pass/fail plus the failing assertion or trace path. Delete nothing from the repo — the only artifacts are `/tmp/*.spec.ts` and the shared browser cache.

package/skills/triage-console/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: triage-console
-description: Standard flow for diagnosing a broken page. Use when a site misbehaves, renders wrong, or throws — snapshot, console errors, failed network requests, then screenshot.
+description: "Standard flow for diagnosing a broken page. Use when a site misbehaves, renders wrong, or throws — snapshot, console errors, failed network requests, then screenshot."
 ---
 
 # Console Triage

package/skills/visual-qa/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: visual-qa
-description: Screenshot discipline for visual checks of WebGL/canvas scenes and layouts. Use when asked whether something "looks right", for before/after comparisons, or for viewport coverage.
+description: "Screenshot discipline for visual checks of WebGL/canvas scenes and layouts. Use when asked whether something \"looks right\", for before/after comparisons, or for viewport coverage."
 ---
 
 # Visual QA
@@ -22,7 +22,7 @@ Save, don't inline — `browser_save_artifact` returns paths. Inline `browser_ta
 
 ## When the tree is blind (canvas/WebGL)
 
-1. `browser_analyze_screenshot` (requires `visionModel` config) for coordinates or element identification the tree cannot provide.
+1. `browser_analyze_screenshot` (requires `visionModel` config) for coordinates or element identification the tree cannot provide. Without it and with a vision-less model, fall back to the global `vision` skill (external Gemini describe script) instead of guessing pixels.
 2. Feed coordinates to `browser_click_at` — never guess pixels yourself.
 3. If the scene needs deterministic frames (animations, thumbnails), stop: that is `playwright-handoff` territory (`waitForFunction`, fixed timeouts, traces).