browser4-experience

Persist and recall browser task traces to reuse selectors and extraction patterns across sessions.

1.1k|151|Updated Mar 12, 2018
One-click install
npx skills add https://github.com/platonai/Browser4 --skill browser4-experience
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: browser4-experience
Source: https://github.com/platonai/Browser4/tree/main/skills/browser4-experience
Command: npx skills add https://github.com/platonai/Browser4 --skill browser4-experience

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Browser automation agents rediscover selectors, steps, and blockers from scratch on every run, wasting tokens and time on sites they have already visited. This Skill persists completed task traces into a local knowledge store and recalls them before new tasks, so repeated or similar tasks complete faster with fewer steps.

Core Features & Use Cases

  • experience_save: Persist a completed task's trace (selectors, steps, blockers, outcome) into a file-backed YAML knowledge store, with success and failure paths both recorded.
  • experience_query: Retrieve prior knowledge before starting a task, returning a replay tier (P1 direct replay through P5 cold start) with confidence scores, primary selectors, and known blockers.
  • experience_list: Inspect stored knowledge entries per domain for diagnostics and debugging.
  • Use Case: An agent extracting product details from Amazon product pages queries stored knowledge first, replays verified selectors at tier P2, and saves the new trace afterward so the next visit runs at higher confidence.

Quick Start

Ask the agent to query prior experience for a target URL before running a browser4-cli agent task, then save the resulting trace after the task completes.

Frequently Asked Questions about browser4-experience

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make a browser agent remember selectors between sessions?

Call experience_save after each completed task to persist its trace, including selectors, steps, and blockers, into a local YAML knowledge store. Before the next task, call experience_query with the target URL to retrieve the stored knowledge and a replay tier.

What are the replay tiers P1 to P5 in experience_query?

The tiers form a confidence ladder: P1 (confidence 0.85 or higher) replays steps directly, P2 verifies each selector before use, P3 uses knowledge as hints with full discovery, P4 is advisory only, and P5 means cold start with no prior data.

Does experience_query require an open browser session?

No, experience_query operates on the file system, not the browser, so it can run before open_session. This lets you plan a task using stored knowledge before launching Chrome.

Why does stored knowledge not apply to a different URL pattern?

Knowledge is matched by URL pattern specificity, so entries stored for one pattern such as /dp/* are not automatically available for another like /s?k=*. You must save separate traces per pattern or run full discovery on the new pattern.

What happens if I forget to call experience_save after a task?

The automatic engine hook in RobustBrowserAgent deposits completed and failed tasks into the knowledge store via MemoryConsolidator, so knowledge is not lost. Manual experience_save calls are still supported for richer traces and diagnostics.