Spider Agent Types

Pure data types and constants for spider_agent automation with zero heavy runtime dependencies.

Overview

This crate extracts all type definitions, system prompts, and helper utilities from spider_agent into a lightweight, dependency-minimal package. Use it when you need automation types without the full agent runtime (no tokio, reqwest, dashmap, or chromey).

Installation

[dependencies]
spider_agent_types = "0.1"

What's Included

Action Types — ActionType (30+ variants), ActionResult, ActionRecord
Chain Execution — ChainStep, ChainCondition, ChainBuilder for sequential action chains
Concurrent Chains — DependencyGraph, DependentStep for parallel execution with dependency ordering
Confidence Tracking — ConfidenceTracker, ConfidenceRetryStrategy for smarter retry decisions
Configuration — RemoteMultimodalConfig, AutomationConfig, ModelProfile, ModelCapabilities
Content Analysis — ContentAnalysis with aho-corasick for fast HTML structure detection
HTML Diff — PageStateDiff, HtmlDiffResult for 50-70% token reduction between rounds
Memory Operations — AutomationMemory, MemoryOperation for session state
Page Observation — PageObservation, InteractiveElement, FormInfo
Planning — PlanningModeConfig, ExecutionPlan, PlannedStep with checkpoints
System Prompts — DEFAULT_SYSTEM_PROMPT, CHROME_AI_SYSTEM_PROMPT, and 6 more
Schema Generation — generate_schema(), infer_schema(), SchemaCache
Selector Cache — SelectorCache with LRU eviction and reliability scoring
Self-Healing — SelfHealingConfig, HealingRequest, HealedSelectorCache
Synthesis — SynthesisConfig, MultiPageContext for multi-page analysis
Tool Calling — ToolCallingMode, ToolDefinition, parse_tool_calls() for OpenAI-compatible function calling
Helpers — JSON extraction, LLM response parsing, FNV hashing, UTF-8 truncation

Quick Start

use spider_agent_types::{
    ActionType, AutomationConfig, RemoteMultimodalConfig,
    ToolCallingMode, HtmlDiffMode, ContentAnalysis,
};

// Create automation config
let config = AutomationConfig::new("Extract product data");

// Analyze HTML content
let analysis = ContentAnalysis::analyze("<html>...</html>");
println!("Needs screenshot: {}", analysis.needs_screenshot);
println!("Text ratio: {:.1}%", analysis.text_ratio * 100.0);

// Parse LLM responses
use spider_agent_types::{extract_last_json_object, extract_assistant_content};
let json = extract_last_json_object(r#"Here is the result: {"name": "test"}"#);

Model Registry

Built-in model profiles for 40+ models with capabilities and pricing:

use spider_agent_types::{model_profile, supports_vision, MODEL_INFO};

if let Some(profile) = model_profile("gpt-4o") {
    println!("Context: {} tokens", profile.context_window);
    println!("Vision: {}", supports_vision("gpt-4o"));
}

Dependencies

Crate	Purpose
`serde` + `serde_json`	Serialization
`aho-corasick`	Fast multi-pattern matching for content analysis
`llm_models_spider`	Model capabilities detection

License

MIT

spider_agent_types 2.45.7