Add AgentRuntime and strategy-based agent execution

[JohnRichard4096]

Summary by Sourcery

Introduce a strategy-based agent execution architecture with AgentRuntime and AgentStrategy, refactor the built-in agent workflow into a reusable AmritaAgentStrategy integrated with ChatObject, and update configuration, templates, and docs to support the new agent runtime and simplified create_agent API.

New Features:

• Add AgentRuntime and create_agent factory to provide a reusable, high-level agent runtime API over ChatObject and presets.
• Introduce AgentStrategy, StrategyContext, and AmritaAgentStrategy to support pluggable, strategy-based agent execution modes including an agent-mixed strategy.
• Add Jinja2-based prompt templating and default system prompt template for dynamic train message rendering.
• Expose minimal_init helper for lightweight initialization flows.

Bug Fixes:

• Tighten message list validation in libchat to reject invalid message item types and avoid silent acceptance of bad inputs.

Enhancements:

• Refactor the built-in agent core into a strategy implementation that uses ToolsManager and StrategyContext, with clearer separation of concerns and improved error/limit handling.
• Update MemoryLimiter to operate on Message objects, support configurable abstract instructions, and better handle token limiting, dropped messages, and rollback behavior.
• Extend ChatObject to accept Message-based system prompts, integrate strategy execution into the main run flow, and support callback-based streaming and queue overflow handling.
• Refine dependency resolution in the event matcher and deprecation metadata for legacy APIs.
• Adjust default configuration by moving agent-related options into a dedicated BuiltinAgentConfig and tuning LLM memory abstraction defaults.

Build:

• Bump project version to 0.6.0, add Jinja2 as a dependency, and remove the deprecated deprecated package from runtime dependencies.

Documentation:

• Revise architecture documentation (EN/ZH) to describe the new entry layer with AgentRuntime, strategy-based execution flow, and layered design, including updated mermaid diagrams.
• Update minimal example docs (EN/ZH) to use create_agent instead of manual ChatObject and preset setup, and clarify ChatObject’s role for advanced use cases.
• Add concept docs for Agent Strategy (EN/ZH) and API reference pages for AgentRuntime, AgentStrategy, AmritaAgentStrategy, and StrategyContext, plus expand API index and navigation.
• Clarify that the debug_log decorator is deprecated in favor of standard logging and update contributing guidelines with an AI-generated content policy.

Tests:

• Rewrite and expand agent tests to cover the new strategy system, StrategyContext behavior, and AmritaAgentStrategy tool-calling scenarios and edge cases.
• Add advanced tests for MemoryLimiter’s abstraction, token limiting, dropped messages, and exception-driven rollback.
• Add advanced ChatObject tests for queue overflow, callbacks, full_response aggregation, message preparation, and callback reconfiguration errors.
• Introduce libchat tests for text generation, message validation, adapter reflection, tools_caller, call_completion, and get_last_response behavior.
• Tighten ToolsManager-related tests to enforce singleton cleanliness and decorator behavior across tests.

Chores:

• Add a uv.lock file for dependency locking and update CONTRIBUTING with AI-generated content guidelines.

[sourcery-ai]

Sorry @JohnRichard4096, your pull request is larger than the review limit of 150000 diff characters

[cloudflare-workers-and-pages]

Deploying amritacore with    Cloudflare Pages

Latest commit:	b3d57c0
Status:	✅  Deploy successful!
Preview URL:	https://ed8eb057.amritacore.pages.dev
Branch Preview URL:	https://feat-agent-matcher.amritacore.pages.dev

View logs

[JohnRichard4096]

@sourcery-ai title

[JohnRichard4096]

@sourcery-ai summary

[JohnRichard4096]

@sourcery-ai review

[sourcery-ai]

Sorry @JohnRichard4096, your pull request is larger than the review limit of 150000 diff characters

[SourceryAI]

Sorry @JohnRichard4096, your pull request is larger than the review limit of 150000 diff characters