📄 agent/system_prompt.py (第 1-22 行)

"""System-prompt assembly for :class:AIAgent.

The agent's system prompt is built once per session and reused across all
turns — only context compression triggers a rebuild.  This keeps the
upstream prefix cache warm.

Three tiers are joined with \\n\\n:

`stable   — identity, tool guidance, skills prompt,

  environment hints, platform hints, model-family guidance.

context  — caller-supplied system_message plus context files
  (AGENTS.md / .cursorrules / etc.) discovered under TERMINAL_CWD.
volatile — memory snapshot, USER.md profile, external memory
  provider block, timestamp/session/model/provider line.

Pure helpers that read the agent's state.  AIAgent keeps thin forwarders."""

扩展现有 tool → CLI 命令 + skill → 服务门控 tool → 插件 → MCP 服务器 → 新核心 tool（最后手段）

📄 run_agent.py (第 1-21 行)

#!/usr/bin/env python3
"""
AI Agent Runner with Tool Calling

This module provides a clean, standalone agent that can execute AI models
with tool calling capabilities. It handles the conversation loop, tool execution,
and response management.

Features:
Automatic tool calling loop until completion

Configurable model parameters

Error handling and recovery

Message history management

Support for multiple model providers


Usage:

    from run_agent import AIAgent

    

    agent = AIAgent(base_url="http://localhost:30000/v1", model="claude-opus-4-20250514")

    response = agent.run_conversation("Tell me about the latest Python updates")

"""

📄 agent/conversation_loop.py (第 1-15 行)

"""The agent conversation loop — extracted from run_agent.AIAgent.

This is the biggest single chunk pulled out of run_agent.py: the
roughly 3,900-line :func:run_conversation body that drives one user
turn through the agent (model call, tool dispatch, retries, fallbacks,
compression, post-turn hooks, background memory/skill review nudges).

The function takes the parent AIAgent instance as its first
argument (agent) and accesses its state via attribute lookup.
_ra().AIAgent.run_conversation is now a thin forwarder."""

📄 agent/system_prompt.py (第 62-79 行)

def build_system_prompt_parts(agent: Any, system_message: Optional[str] = None) -> Dict[str, str]:
    """Assemble the system prompt as three ordered parts.

    Returns a dict with three keys:
      * stable   — identity, tool guidance, skills prompt,
        environment hints, platform hints, model-family operational
        guidance.
      * context  — context files (AGENTS.md, .cursorrules, etc.)
        and caller-supplied system_message.
      * volatile — memory snapshot, user profile, external
        memory provider block, timestamp line.

    Joined into a single string by :func:build_system_prompt and
    cached on agent._cached_system_prompt for the lifetime of the
    AIAgent.  Hermes never re-renders parts of this string mid-
    session — that's the only way to keep upstream prompt caches
    warm across turns."""

📄 tools/registry.py (第 1-15 行)

"""Central registry for all hermes-agent tools.

Each tool file calls registry.register() at module level to declare its
schema, handler, toolset membership, and availability check.  model_tools.py`
queries the registry instead of maintaining its own parallel data structures.

Import chain (circular-import safe):
    tools/registry.py  (no imports from model_tools or tool files)
           ^
    tools/*.py  (import from tools.registry at module level)
           ^
    model_tools.py  (imports tools.registry + all tool modules)
           ^
    run_agent.py, cli.py, batch_runner.py, etc."""

📄 toolsets.py (第 31-76 行)

# Shared tool list for CLI and all messaging platform toolsets.
Edit this once to update all platforms simultaneously.

_HERMES_CORE_TOOLS = [
    # Web
    "web_search", "web_extract",
    # Terminal + process management
    "terminal", "process",
    # File manipulation
    "read_file", "write_file", "patch", "search_files",
    # Vision + image generation
    "vision_analyze", "image_generate",
    # Skills
    "skills_list", "skill_view", "skill_manage",
    # Browser automation
    "browser_navigate", "browser_snapshot", "browser_click",
    "browser_type", "browser_scroll", "browser_back",
    # ... 更多工具
    # Kanban multi-agent coordination
    "kanban_show", "kanban_list",
    "kanban_complete", "kanban_block", "kanban_heartbeat",
    # Computer use (macOS)
    "computer_use",
]

📄 hermes_state.py (第 1-15 行)

#!/usr/bin/env python3
"""
SQLite State Store for Hermes Agent.

Provides persistent session storage with FTS5 full-text search, replacing
the per-session JSONL file approach. Stores session metadata, full message
history, and model configuration for CLI and gateway sessions.

Key design decisions:
WAL mode for concurrent readers + one writer (gateway multi-platform)

FTS5 virtual table for fast text search across all session messages

Compression-triggered session splitting via parent_session_id chains

Batch runner and RL trajectories are NOT stored here (separate systems)

Session source tagging ('cli', 'telegram', 'discord', etc.) for filtering

"""