Skip to main contentCrate skilllite_agent
Source - agent_loop
- Core agent loop: LLM ↔ tool execution cycle.
- chat
- CLI chat entry-points: single-shot and interactive REPL.
- chat_session
- Chat session: persistent conversation with transcript and memory.
- dependency_resolver
- LLM-based dependency resolution for skilllite-agent.
- evolution
- Evolution integration: implements EvolutionLlm for agent’s LlmClient.
- extensions
- Agent tool extensions and unified registry.
- goal_boundaries
- Goal boundary extraction (A5): scope, exclusions, completion conditions.
- high_risk
- A11: 高危工具确认 — 可配置 write_key_path、run_command、network 等需发消息确认
- llm
- LLM HTTP client for chat completions with tool calling.
- long_text
- Long text processing: truncation and chunked selection/summarization.
- planning_guard
- Lightweight guards around task planning results.
- planning_rules
- Planning rules for task generation.
- prompt
- Prompt construction for the agent.
- rpc
- Agent Chat RPC: JSON-Lines event stream protocol over stdio.
- run_checkpoint
- A13: Run mode checkpoint — save/restore state for long-running tasks.
- skills
- Skills invocation: wraps sandbox run/exec for the agent layer.
- soul
- Agent constitution: Law + Beliefs + Soul.
- task_planner
- Task Planner: task planning and tracking for agentic loops.
- tool_hint_resolver
- ToolHintResolver: single source of truth for tool_hint → tool name mapping.
- types
- Shared types for the agent module.