Add memory system to BrowserPerceptiveAgent

[Copilot]

Adds short-term and long-term memory to enable the agent to maintain context across multi-step interactions.

Architecture

Three-tier memory system:

• ShortTermMemory: Sliding window (50 entries) for recent context
• LongTermMemory: Importance-filtered (threshold 0.7) persistent storage (200 entries)
• CompositeMemory: Unified interface with deduplication and relevance-ranked retrieval

LLM integration:

• Added memory field to ObserveElement schema for LLM to provide contextual observations
• PromptBuilder injects relevant memories into prompts based on current instruction
• Automatic importance scoring (0.3-1.0) based on action success and task completion

Checkpoint persistence:

• MemorySnapshot serialization for checkpoint save/restore
• Memories automatically persisted and restored across sessions

Configuration

val config = AgentConfig(
    enableMemory = true,                    // Toggle memory system
    shortTermMemorySize = 50,               // Sliding window size
    longTermMemorySize = 200,               // Max long-term entries
    longTermMemoryThreshold = 0.7,          // Min importance for long-term storage
    memoryRetrievalLimit = 10               // Max memories in prompts
)

Usage

val agent = BrowserPerceptiveAgent(driver, session, config)

// Memories auto-collected from LLM responses and auto-injected into prompts
agent.resolve("Search for products and compare prices")

// Manual memory operations
agent.memory.add("User prefers items under $100", importance = 0.9)
val relevant = agent.memory.retrieve(query = "price", limit = 5)

Files Changed

• Core: BrowserPerceptiveAgent.kt, PromptBuilder.kt, PerceptiveAgent.kt, CheckpointManager.kt
• New: AgentMemory.kt, ShortTermMemory.kt, LongTermMemory.kt, CompositeMemory.kt
• Tests: Comprehensive coverage for all memory components (100% pass rate)
• Docs: English and Chinese documentation with architecture, examples, and best practices

Original prompt

为 BrowserPerceptiveAgent 增加记忆支持。

---

✨ Let Copilot coding agent set things up for you — coding agent works faster and does higher quality work when set up for your repo.