Hermes Agent

Hermes Agent - Complete Project Guide (A-Z)

Purpose of this document: A single, comprehensive reference that explains everything about the Hermes Agent project — its architecture, source code, features, release history, and design patterns — so that any AI or developer can fully understand the system.

Table of Contents

1. Project Overview
2. Key Features Summary
3. Installation & Getting Started
4. Project Structure
5. Core Architecture
   • 5.1 AIAgent Class (run_agent.py)
   • 5.2 Tool Orchestration (model_tools.py)
   • 5.3 Toolset System (toolsets.py)
   • 5.4 Tool Registry (tools/registry.py)
   • 5.5 Session Database (hermes_state.py)
   • 5.6 Constants & Home Directory (hermes_constants.py)
6. CLI System
   • 6.1 Interactive CLI (cli.py)
   • 6.2 CLI Entry Point (hermes_cli/main.py)
   • 6.3 Configuration System (hermes_cli/config.py)
   • 6.4 Slash Command Registry (hermes_cli/commands.py)
   • 6.5 Setup Wizard (hermes_cli/setup.py)
   • 6.6 Model Catalog (hermes_cli/models.py)
   • 6.7 Skin/Theme Engine (hermes_cli/skin_engine.py)
7. Tool System
   • 7.1 Terminal Tool (tools/terminal_tool.py)
   • 7.2 File Tools (tools/file_tools.py)
   • 7.3 Web Tools (tools/web_tools.py)
   • 7.4 Browser Tool (tools/browser_tool.py)
   • 7.5 Delegate Tool (tools/delegate_tool.py)
   • 7.6 MCP Tool (tools/mcp_tool.py)
   • 7.7 Approval System (tools/approval.py)
   • 7.8 Terminal Backends (tools/environments/)
8. Agent Internals
   • 8.1 Prompt Builder (agent/prompt_builder.py)
   • 8.2 Context Compressor (agent/context_compressor.py)
   • 8.3 Prompt Caching (agent/prompt_caching.py)
   • 8.4 Auxiliary Client (agent/auxiliary_client.py)
   • 8.5 Display & Spinner (agent/display.py)
   • 8.6 Skill Commands (agent/skill_commands.py)
9. Messaging Gateway
   • 9.1 GatewayRunner (gateway/run.py)
   • 9.2 Session Store (gateway/session.py)
   • 9.3 Platform Adapters (gateway/platforms/)
10. Cron Scheduling
11. Skills System
12. Plugin System
13. Memory System
14. ACP Server (IDE Integration)
15. API Server
16. MCP Server Mode
17. RL Training Environments
18. Profiles (Multi-Instance)
19. Security Model
20. Provider & Model System
21. Streaming & Reasoning
22. Release History
23. File Dependency Chain
24. Key Design Patterns
25. Configuration Reference
26. Known Pitfalls

1. Project Overview

Hermes Agent is a self-improving AI agent built by Nous Research. It is an open-source (MIT licensed), Python-based project that provides:

• A full interactive terminal UI (CLI) for conversing with LLMs
• A messaging gateway supporting 16+ platforms (Telegram, Discord, Slack, WhatsApp, Signal, Matrix, Email, etc.)
• A closed learning loop — the agent creates skills from experience, improves them during use, nudges itself to persist knowledge, searches past conversations, and builds a deepening model of who you are
• 40+ built-in tools — terminal execution, file manipulation, web search, browser automation, code execution, image generation, TTS/STT, and more
• Any LLM provider — OpenRouter (200+ models), Nous Portal (400+ models), OpenAI, Anthropic, Hugging Face, GitHub Copilot, z.ai/GLM, Kimi/Moonshot, MiniMax, Alibaba/DashScope, custom endpoints
• Six terminal backends — local, Docker, SSH, Modal (serverless), Daytona (serverless), Singularity (HPC)
• Scheduled automations via built-in cron scheduler
• IDE integration via ACP (Agent Communication Protocol) for VS Code, Zed, JetBrains
• MCP integration — both client (connect to any MCP server) and server (expose Hermes to MCP clients)
• RL training via Atropos environments for training the next generation of tool-calling models

Tech Stack:

• Python 3.11+ (core agent, tools, gateway, cron)
• Node.js (browser automation via agent-browser)
• SQLite with WAL mode and FTS5 (session storage, full-text search)
• OpenAI-compatible API (primary inference interface)
• Anthropic SDK (native Anthropic support)
• Rich + prompt_toolkit (CLI rendering)

Repository: github.com/NousResearch/hermes-agent Version: 0.7.0 (as of April 2026) License: MIT

2. Key Features Summary

3. Installation & Getting Started

# One-line install (Linux, macOS, WSL2)
curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash

# After install
source ~/.bashrc    # or: source ~/.zshrc
hermes              # start chatting

# Key commands
hermes model        # Choose LLM provider and model
hermes tools        # Configure which tools are enabled
hermes config set   # Set individual config values
hermes gateway      # Start the messaging gateway
hermes setup        # Run the full setup wizard
hermes update       # Update to latest version
hermes doctor       # Diagnose any issues

For development:

git clone https://github.com/NousResearch/hermes-agent.git
cd hermes-agent
curl -LsSf https://astral.sh/uv/install.sh | sh
uv venv venv --python 3.11
source venv/bin/activate
uv pip install -e ".[all,dev]"
python -m pytest tests/ -q    # ~3000 tests

4. Project Structure

hermes-agent/
├── run_agent.py              # AIAgent class — core conversation loop
├── model_tools.py            # Tool orchestration, _discover_tools(), handle_function_call()
├── toolsets.py               # Toolset definitions, _HERMES_CORE_TOOLS list
├── toolset_distributions.py  # Toolset sampling distributions for RL
├── cli.py                    # HermesCLI class — interactive CLI orchestrator
├── hermes_state.py           # SessionDB — SQLite session store (FTS5 search)
├── hermes_constants.py       # Shared constants, get_hermes_home()
├── hermes_time.py            # Timezone handling
├── utils.py                  # Shared utility functions
├── batch_runner.py           # Parallel batch processing
├── trajectory_compressor.py  # Trajectory compression for RL training
├── mcp_serve.py              # MCP server mode entry point
├── mini_swe_runner.py        # Minimal SWE benchmark runner
├── rl_cli.py                 # RL CLI commands
│
├── agent/                    # Agent internals
│   ├── prompt_builder.py         # System prompt assembly
│   ├── context_compressor.py     # Auto context compression
│   ├── prompt_caching.py         # Anthropic prompt caching
│   ├── auxiliary_client.py       # Auxiliary LLM client (vision, summarization)
│   ├── model_metadata.py         # Model context lengths, token estimation
│   ├── models_dev.py             # models.dev registry integration
│   ├── display.py                # KawaiiSpinner, tool preview formatting
│   ├── skill_commands.py         # Skill slash commands (shared CLI/gateway)
│   └── trajectory.py             # Trajectory saving helpers
│
├── hermes_cli/               # CLI subcommands and setup
│   ├── main.py               # Entry point — all `hermes` subcommands
│   ├── config.py             # DEFAULT_CONFIG, OPTIONAL_ENV_VARS, migration
│   ├── commands.py           # Slash command definitions + SlashCommandCompleter
│   ├── callbacks.py          # Terminal callbacks (clarify, sudo, approval)
│   ├── setup.py              # Interactive setup wizard
│   ├── skin_engine.py        # Skin/theme engine
│   ├── skills_config.py      # `hermes skills` — skill management
│   ├── tools_config.py       # `hermes tools` — tool management
│   ├── skills_hub.py         # Skills Hub integration
│   ├── models.py             # Model catalog, provider model lists
│   ├── model_switch.py       # Shared /model switch pipeline
│   └── auth.py               # Provider credential resolution
│
├── tools/                    # Tool implementations (one file per tool)
│   ├── registry.py           # Central tool registry
│   ├── approval.py           # Dangerous command detection
│   ├── terminal_tool.py      # Terminal/shell execution
│   ├── process_registry.py   # Background process management
│   ├── file_tools.py         # File read/write/search/patch
│   ├── web_tools.py          # Web search/extract
│   ├── browser_tool.py       # Browser automation
│   ├── code_execution_tool.py # execute_code sandbox
│   ├── delegate_tool.py      # Subagent delegation
│   ├── mcp_tool.py           # MCP client integration
│   ├── skills_tool.py        # Skill management tool
│   ├── todo_tool.py          # Todo/task tracking tool
│   ├── memory_tool.py        # Memory read/write tool
│   ├── tts_tool.py           # Text-to-speech
│   ├── vision_tool.py        # Image analysis
│   ├── image_gen_tool.py     # Image generation
│   └── environments/         # Terminal backends
│       ├── base.py               # BaseEnvironment ABC
│       ├── local.py              # Local execution
│       ├── docker.py             # Docker containers
│       ├── ssh.py                # SSH remote execution
│       ├── modal.py              # Modal serverless
│       ├── managed_modal.py      # Nous-hosted Modal
│       ├── daytona.py            # Daytona serverless
│       ├── singularity.py        # Singularity HPC containers
│       └── persistent_shell.py   # Persistent shell mixin
│
├── gateway/                  # Messaging platform gateway
│   ├── run.py                # GatewayRunner — main message loop
│   ├── session.py            # SessionStore — conversation persistence
│   ├── status.py             # Gateway status, token locks
│   └── platforms/            # 16 platform adapters
│       ├── base.py               # BasePlatformAdapter ABC
│       ├── telegram.py           # Telegram (polling + webhook)
│       ├── discord.py            # Discord
│       ├── slack.py              # Slack
│       ├── whatsapp.py           # WhatsApp
│       ├── matrix.py             # Matrix (E2EE)
│       ├── signal.py             # Signal
│       ├── email.py              # Email (IMAP/SMTP)
│       ├── homeassistant.py      # Home Assistant
│       ├── sms.py                # SMS (Twilio)
│       ├── mattermost.py         # Mattermost
│       ├── dingtalk.py           # DingTalk
│       ├── feishu.py             # Feishu/Lark
│       ├── wecom.py              # WeCom (Enterprise WeChat)
│       ├── webhook.py            # Generic webhook
│       └── api_server.py         # OpenAI-compatible API server
│
├── acp_adapter/              # ACP server (IDE integration)
│   ├── server.py             # HermesACPAgent class
│   ├── session.py            # SessionManager
│   ├── events.py             # Streaming callbacks
│   ├── permissions.py        # Approval callbacks
│   └── entry.py              # Entry point
│
├── cron/                     # Scheduler
│   ├── scheduler.py          # tick() — job execution engine
│   └── jobs.py               # Job storage and CRUD
│
├── plugins/                  # Plugin system
│   └── memory/               # 8 memory provider plugins
│       ├── openviking/
│       ├── mem0/
│       ├── hindsight/
│       ├── holographic/
│       ├── honcho/
│       ├── retaindb/
│       └── byterover/
│
├── environments/             # RL training environments (Atropos)
│   ├── hermes_base_env.py    # Abstract base RL environment
│   ├── agent_loop.py         # HermesAgentLoop — rollout execution
│   ├── tool_context.py       # ToolContext — sandbox for RL
│   ├── web_research_env.py   # Web research tasks
│   └── agentic_opd_env.py    # Observation-Prediction-Demo env
│
├── skills/                   # 70+ bundled skills across 28 categories
├── optional-skills/          # Additional optional skills
├── tests/                    # ~3000 pytest tests
├── scripts/                  # Install, update, packaging scripts
├── docker/                   # Docker build files
├── docs/                     # Documentation source (Docusaurus)
├── website/                  # Landing page
├── desktop/                  # Desktop app (Electron, separate repo)
├── tinker-atropos/           # RL submodule
│
├── pyproject.toml            # Python package config
├── AGENTS.md                 # Developer guide for AI assistants
├── RELEASE_v0.2.0.md → v0.7.0.md  # Release notes
└── cli-config.yaml.example   # Example config

User config directory: ~/.hermes/

~/.hermes/
├── config.yaml           # User settings
├── .env                  # API keys and secrets
├── MEMORY.md             # Persistent agent memory
├── USER.md               # User profile
├── SOUL.md               # Agent personality/identity
├── sessions.db           # SQLite session database
├── skills/               # User-installed skills
├── skins/                # Custom CLI themes
├── plugins/              # User plugins
├── cron/                 # Cron jobs and output
│   ├── jobs.json
│   └── output/
├── cache/                # Image/audio cache
├── plans/                # Generated plans
├── profiles/             # Multi-instance profiles
└── mcp/                  # MCP server configs

5. Core Architecture

5.1 AIAgent Class (run_agent.py)

The AIAgent class is the heart of the system — the core conversation loop that orchestrates LLM calls, tool execution, context management, and response delivery.

Constructor (~60 parameters):

class AIAgent:
    def __init__(self,
        model: str = "anthropic/claude-opus-4.6",
        max_iterations: int = 90,
        enabled_toolsets: list = None,
        disabled_toolsets: list = None,
        quiet_mode: bool = False,
        save_trajectories: bool = False,
        platform: str = None,            # "cli", "telegram", etc.
        session_id: str = None,
        session_db: SessionDB = None,
        skip_context_files: bool = False,
        skip_memory: bool = False,
        base_url: str = None,
        api_key: str = None,
        provider: str = None,
        api_mode: str = "chat_completions",  # or "anthropic_messages" or "codex_responses"
        tool_progress_callback = None,
        stream_delta_callback = None,
        thinking_callback = None,
        status_callback = None,
        iteration_budget: IterationBudget = None,
        credential_pool = None,
        checkpoints_enabled: bool = False,
        # ... plus provider, routing, callback params
    )

Main Methods:

The Core Agent Loop (inside run_conversation()):

while api_call_count < self.max_iterations and self.iteration_budget.remaining > 0:
    response = client.chat.completions.create(
        model=model, messages=messages, tools=tool_schemas
    )
    if response.tool_calls:
        for tool_call in response.tool_calls:
            result = handle_function_call(tool_call.name, tool_call.args, task_id)
            messages.append(tool_result_message(result))
        api_call_count += 1
    else:
        return response.content  # Final text response

Key Behaviors:

• Three API modes: chat_completions (OpenAI-compatible), anthropic_messages (Anthropic SDK), codex_responses (OpenAI Codex)
• Parallel tool execution: Independent tool calls run concurrently via ThreadPoolExecutor (unless they share file paths or are in the never-parallel list)
• Interrupt support: Background threads allow interrupt detection without blocking on HTTP
• Error recovery: Automatic fallback chain (primary → fallback model), retry with exponential backoff, context compression on token overflow
• Budget pressure: Warnings at 70% (caution) and 90% (urgent) of iteration budget
• Oversized results: Tool results >100K chars are saved to temp files with a preview
• Stale connection detection: 90s timeout for streaming, 60s read timeout

IterationBudget (thread-safe):

class IterationBudget:
    def consume() -> bool    # Check and consume one iteration
    def refund()             # Give back iteration (for execute_code turns)
    @property remaining      # Remaining iterations

5.2 Tool Orchestration (model_tools.py)

Bridges the agent and tool registry — handles discovery, schema generation, and dispatch.

Key Functions:

Tool Discovery Order:

1. Static tools (web_tools, terminal_tool, file_tools, browser_tool, etc.)
2. Optional tools (fal_client for image gen, honcho, etc.) — graceful fallback if missing
3. Plugin-registered tools
4. MCP server tools (dynamic, via tools/list_changed notifications)

Special Tool Handling:

• Agent-level tools (todo, memory, session_search, delegate_task): Intercepted by run_agent.py before handle_function_call()
• execute_code: Passes enabled_tools for sandbox tool list
• Dynamic schema adjustments: browser_navigate strips web_search reference if tools unavailable

Async Bridging:

• Persistent event loops (not asyncio.run()) to prevent "Event loop is closed" errors
• Main thread uses shared loop; worker threads get per-thread loops
• _run_async() detects running loop and spins up disposable thread if needed

5.3 Toolset System (toolsets.py)

Provides flexible tool grouping and composition.

Core Toolsets:

Composite Toolsets:

• hermes-cli — All core tools for CLI platform
• hermes-telegram, hermes-discord, etc. — Platform-specific tool sets
• hermes-gateway — Union of all platform tools
• debugging — terminal + file + web
• safe — Everything except terminal

Resolution:

resolve_toolset(name, visited=None) → List[str]
# Recursively resolves toolset to tool names
# Handles composition (includes) and cycle detection
# Special aliases: "all" or "*" = all tools

5.4 Tool Registry (tools/registry.py)

Singleton managing all tool schemas and handlers. Circular-import safe — has no tool dependencies.

ToolEntry (per-tool metadata):

@dataclass(slots=True)
class ToolEntry:
    name: str
    toolset: str
    schema: dict           # OpenAI-format tool definition
    handler: Callable      # Sync or async handler function
    check_fn: Callable     # Returns True if tool is available
    requires_env: list     # Required environment variables
    is_async: bool
    description: str
    emoji: str

Key Methods:

registry.register(name, toolset, schema, handler, check_fn, requires_env)
registry.get_definitions(tool_names, quiet)  # Returns filtered schemas
registry.dispatch(name, args, **kwargs)      # Execute with async bridging
registry.deregister(name)                    # Remove (for MCP tool refresh)
registry.check_tool_availability()           # Returns (available, unavailable)

5.5 Session Database (hermes_state.py)

SQLite-based persistent session storage with FTS5 full-text search.

Schema (v6):

-- Sessions table
sessions (
    id TEXT PRIMARY KEY,
    source TEXT, user_id TEXT, model TEXT, model_config TEXT,
    system_prompt TEXT, parent_session_id TEXT,
    started_at TEXT, ended_at TEXT, end_reason TEXT,
    message_count INTEGER, tool_call_count INTEGER,
    input_tokens INTEGER, output_tokens INTEGER,
    cache_read_tokens INTEGER, cache_write_tokens INTEGER, reasoning_tokens INTEGER,
    estimated_cost_usd REAL, actual_cost_usd REAL,
    title TEXT  -- UNIQUE INDEX
)

-- Messages table
messages (
    id INTEGER PRIMARY KEY AUTOINCREMENT,
    session_id TEXT, role TEXT, content TEXT,
    tool_call_id TEXT, tool_calls TEXT,  -- JSON
    tool_name TEXT, timestamp TEXT,
    token_count INTEGER, finish_reason TEXT,
    reasoning TEXT, reasoning_details TEXT, codex_reasoning_items TEXT
)

-- FTS5 virtual table (auto-synced via triggers)
messages_fts (content)

Concurrency Model:

• WAL (Write-Ahead Logging) for concurrent readers + single writer
• BEGIN IMMEDIATE for write transactions (lock at start, not commit)
• Jitter retry on lock: 20-150ms random backoff, max 15 retries
• Periodic WAL checkpoint every 50 writes

Key Operations:

• create_session(), end_session(), reopen_session()
• add_message(), get_messages()
• search_sessions(query) — FTS5 full-text search
• update_token_counts() — Supports both incremental (CLI) and absolute (gateway) modes

5.6 Constants & Home Directory (hermes_constants.py)

Import-safe constants module with no circular dependencies.

get_hermes_home() → Path          # HERMES_HOME env var or ~/.hermes
display_hermes_home() → str       # User-friendly display: "~/.hermes"
get_optional_skills_dir() → Path  # HERMES_OPTIONAL_SKILLS env var
parse_reasoning_effort(str) → Dict  # "high" → {"enabled": True, "effort": "high"}

# Key constants
OPENROUTER_BASE_URL = "https://openrouter.ai/api/v1"
NOUS_API_BASE_URL = "https://inference-api.nousresearch.com/v1"
AI_GATEWAY_BASE_URL = "https://ai-gateway.vercel.sh/v1"
VALID_REASONING_EFFORTS = ("xhigh", "high", "medium", "low", "minimal")

6. CLI System

6.1 Interactive CLI (cli.py)

The HermesCLI class provides the interactive terminal interface.

Features:

• Rich for banner/panels, prompt_toolkit for input with autocomplete
• KawaiiSpinner — animated faces during API calls, ┊ activity feed for tool results
• Multiline editing with Shift+Enter
• Slash-command autocomplete
• Session history with up/down arrow navigation
• Clipboard image paste (Alt+V / Ctrl+V)
• Status bar showing model, provider, and token counts
• Inline diff previews for file write/patch operations

Configuration Loading:

load_cli_config() → dict
# Loads from ~/.hermes/config.yaml (or ./cli-config.yaml fallback)
# Merges with hardcoded defaults
# Expands ${ENV_VAR} references
# Maps terminal config → env vars

6.2 CLI Entry Point (hermes_cli/main.py)

All hermes subcommands are dispatched from here:

hermes                    # Default: interactive chat
hermes chat               # Explicit interactive mode
hermes gateway start|stop|status|install|uninstall
hermes setup              # Setup wizard
hermes model              # Select model/provider
hermes tools              # Configure tools
hermes skills             # Manage skills
hermes config set|get     # Direct config manipulation
hermes cron list|delete   # Cron job management
hermes doctor             # Diagnose issues
hermes sessions browse    # Session picker
hermes profile create|list|switch|delete|export|import
hermes mcp serve|add|remove  # MCP management
hermes acp                # Start ACP server
hermes update|uninstall|version

Profile System:

• _apply_profile_override() runs BEFORE any imports to set HERMES_HOME
• Pre-parses --profile/-p from argv
• Allows fully isolated agent instances with separate config, memory, sessions, skills

6.3 Configuration System (hermes_cli/config.py)

Key Configuration Sections:

model: "anthropic/claude-opus-4.6" # or dict with provider/base_url/api_key
providers: {} # Provider-specific configs
fallback_providers: [] # Ordered failover list
credential_pool: {} # Multiple API keys per provider

agent:
  max_turns: 90
  gateway_timeout: 1800
  tool_use_enforcement: "auto"

terminal:
  backend: "local" # local|docker|modal|daytona|ssh|singularity
  timeout: 180
  persistent_shell: true
  docker_image: "nikolaik/python-nodejs:..."

compression:
  enabled: true
  threshold: 0.50 # Compress when 50% of context used
  target_ratio: 0.20 # Summary = 20% of compressed content
  protect_last_n: 20

auxiliary:
  vision: { provider, model }
  web_extract: { provider, model }
  compression: { provider, model }

memory:
  memory_enabled: true
  provider: "" # "" | "honcho" | "mem0" | etc.
  memory_char_limit: 2200

display:
  personality: "kawaii"
  show_reasoning: false
  inline_diffs: true
  skin: "default"
  streaming: true

tts:
  provider: "edge" # edge|elevenlabs|openai|neutts

stt:
  enabled: true
  provider: "local" # local|groq|openai

privacy:
  redact_pii: false

mcp_servers: {} # MCP server configurations

skills:
  external_dirs: [] # Additional skill directories

approvals:
  mode: "smart" # smart|always|off

Config Files:

• ~/.hermes/config.yaml — User settings (authoritative)
• ~/.hermes/.env — API keys and secrets
• Config version migration system (currently v5)

6.4 Slash Command Registry (hermes_cli/commands.py)

All slash commands defined centrally in COMMAND_REGISTRY:

CommandDef(name, description, category, aliases, args_hint, cli_only, gateway_only)

Derived automatically by:

• CLI process_command() — dispatch on canonical name
• Gateway dispatch + help
• Telegram BotCommand menu
• Slack /hermes subcommands
• Autocomplete + help text

Key Commands:

6.5 Setup Wizard (hermes_cli/setup.py)

Modular interactive wizard with independent sections:

1. Model & Provider — Select AI provider, enter API keys, choose model
2. Terminal Backend — Choose execution environment
3. Agent Settings — Max iterations, compression, session policies
4. Messaging Platforms — Configure Telegram, Discord, Slack, etc.
5. Tools — TTS, STT, web search, image generation, browser

Features:

• Live credential validation
• Real-time model list fetching from provider APIs
• Automatic OpenClaw migration detection
• Atomic config file writes

6.6 Model Catalog (hermes_cli/models.py)

Provider-specific model lists:

_PROVIDER_MODELS = {
    "nous": ["anthropic/claude-opus-4.6", "anthropic/claude-sonnet-4.6", ...],  # 25+
    "openrouter": ["anthropic/claude-opus-4.6", "google/gemini-3-flash", ...],  # 30+
    "anthropic": ["claude-opus-4-6", "claude-sonnet-4-6", ...],
    "openai": ["gpt-5", "gpt-5.4-mini", "gpt-4.1", "gpt-4o", ...],
    "copilot": ["gpt-5.4", "gpt-5.4-mini", "gpt-5.3-codex", ...],
    "huggingface": [...],
    "minimax": [...],
    "kimi-coding": [...],
    "alibaba": [...],
    "deepseek": [...],
    # ... more providers
}

Features:

• Dynamic fetching via provider /models endpoints
• Curated lists used when live probe returns fewer models
• Fuzzy matching for typo correction
• Validation against provider catalog

6.7 Skin/Theme Engine (hermes_cli/skin_engine.py)

Data-driven CLI visual customization — no code changes needed.

Customizable Elements:

Built-in Skins: default, ares, mono, slate, poseidon, sisyphus, charizard

User Skins: Drop ~/.hermes/skins/<name>.yaml and activate with /skin <name>

7. Tool System

7.1 Terminal Tool (tools/terminal_tool.py)

Shell command execution across multiple backends.

def terminal_tool(
    command: str,
    background: bool = False,
    timeout: Optional[int] = None,
    task_id: Optional[str] = None,
    force: bool = False,        # Skip approval for dangerous commands
    workdir: Optional[str] = None,
    check_interval: Optional[int] = None,  # Background task polling
    pty: bool = False,
) -> str  # JSON result

Features:

• Multi-backend: Selects based on TERMINAL_ENV (local/docker/ssh/modal/daytona/singularity)
• Per-task_id sandboxes with thread-safe creation locks
• Dangerous command routing through approval system
• Background task support with file-based IPC
• Interrupt handling — polls is_interrupted() during execution
• Auto-cleanup daemon thread for idle environments (>300s)
• Disk usage warnings at configurable threshold

7.2 File Tools (tools/file_tools.py)

Safe file operations with size guards and sensitive path protection.

• read_file_tool(path, offset, limit) — Read with pagination (default 100K char limit)
• write_file_tool(path, content) — Write with approval for sensitive paths
• edit_file_tool(path, old_text, new_text) — String replacement editing
• list_files_tool(path) — Directory listing
• search_files_tool(pattern, path) — Glob/regex file search

Safety:

• Device path blocklist (/dev/zero, /dev/stdin, etc.)
• Read dedup tracking — returns stub on re-read if mtime unchanged
• Sensitive path blocking: /etc/, /boot/, ~/.ssh without approval
• Prompt injection protection for known dangerous paths

7.3 Web Tools (tools/web_tools.py)

Web search and content extraction.

• web_search_tool(query, limit) — Search via configurable backend
• web_extract_tool(urls, format) — Extract content from URLs
• `web_crawl_tool(url, ins

…(truncated)