export interface WikiSkill { name: string category: string description: string version: string platforms: string[] tags: string[] body: string } export interface WikiPlugin { name: string version: string description: string tools: string[] hooks: string[] } export interface WikiTool { name: string description: string category: string params: { name: string; type: string; required: boolean; description: string }[] } export interface WikiCommand { command: string description: string flags: string[] } export const overview = { version: '0.7.0', skillCount: 142, pluginCount: 8, toolCount: 56, sessionCount: 1847, totalCost: 312.48, categories: [ 'apple', 'productivity', 'research', 'creative', 'software-development', 'github', 'media', 'devops', 'mlops', 'data-science', 'mcp', 'messaging', 'email', 'leisure', 'red-teaming', 'smart-home', 'diagramming', 'social-media', 'gaming', 'note-taking', ], } export const skills: WikiSkill[] = [ { name: 'imessage', category: 'apple', description: 'Send and receive iMessages/SMS via the imsg CLI on macOS', version: '1.0.0', platforms: ['macos'], tags: ['iMessage', 'SMS', 'messaging', 'Apple'], body: `## When to Use\n\nUse this skill when the user asks you to send or read iMessages, SMS, or text messages on macOS.\n\n## Prerequisites\n\n- macOS only\n- \`imsg\` CLI tool installed\n- Full Disk Access granted to terminal\n\n## Usage\n\n\`\`\`bash\nimsg send "+1234567890" "Hello from Hermes"\nimsg read --limit 10\nimsg conversations\n\`\`\`\n\n## Notes\n\n- Group messages require the conversation ID\n- Media attachments are supported via \`--attachment\` flag`, }, { name: 'apple-notes', category: 'apple', description: 'Create, read, and search Apple Notes via AppleScript', version: '1.0.0', platforms: ['macos'], tags: ['Notes', 'Apple', 'AppleScript'], body: `## When to Use\n\nUse when the user wants to interact with Apple Notes -- creating, reading, searching, or organizing notes.\n\n## How It Works\n\nUses AppleScript to interface with Notes.app. All operations are non-destructive by default.\n\n## Commands\n\n- Create note: \`osascript -e 'tell application "Notes" to make new note...'\`\n- List notes: queries the Notes database\n- Search: full-text search across all folders`, }, { name: 'browser-automation', category: 'research', description: 'Headless browser automation with Playwright for web scraping and interaction', version: '1.2.0', platforms: ['macos', 'linux', 'windows'], tags: ['browser', 'Playwright', 'scraping', 'automation'], body: `## When to Use\n\nUse for tasks requiring real browser interaction: filling forms, navigating SPAs, scraping JavaScript-rendered content, taking screenshots of web pages.\n\n## Backends\n\n- **Browserbase** (cloud) -- no local install needed\n- **Local Chromium** -- headless via Playwright\n\n## Capabilities\n\n- Navigate, click, type, scroll via accessibility tree snapshots\n- Extract page content as markdown\n- Execute JavaScript in page context\n- Vision analysis of rendered pages\n- Screenshot capture`, }, { name: 'web-search', category: 'research', description: 'Search the web using Exa, Firecrawl, Tavily, or parallel multi-engine search', version: '1.0.0', platforms: ['macos', 'linux', 'windows'], tags: ['search', 'web', 'Exa', 'Tavily', 'Firecrawl'], body: `## When to Use\n\nUse when you need current information not in your training data.\n\n## Backends\n\n1. **Exa** -- neural search with content extraction\n2. **Firecrawl** -- web crawling and extraction\n3. **Tavily** -- AI-optimized search\n4. **Parallel** -- multi-engine aggregation\n\n## Usage\n\nThe orchestrator automatically selects this when it detects a query needing current information. Results are compressed via LLM summarization.`, }, { name: 'git-workflow', category: 'software-development', description: 'Git operations: branching, committing, rebasing, PR workflows', version: '1.3.0', platforms: ['macos', 'linux', 'windows'], tags: ['git', 'version control', 'branching', 'PR'], body: `## When to Use\n\nUse for any git-related task: creating branches, committing, rebasing, resolving merge conflicts, managing PRs.\n\n## Patterns\n\n### Feature Branch\n\`\`\`bash\ngit checkout -b feature/my-feature\ngit add -A && git commit -m "add feature"\ngit push -u origin feature/my-feature\ngh pr create --title "Add feature"\n\`\`\`\n\n### Rebase Workflow\n\`\`\`bash\ngit fetch origin\ngit rebase origin/main\ngit push --force-with-lease\n\`\`\``, }, { name: 'docker-compose', category: 'devops', description: 'Manage Docker Compose services, volumes, networks, and builds', version: '1.0.0', platforms: ['macos', 'linux'], tags: ['Docker', 'containers', 'orchestration', 'devops'], body: `## When to Use\n\nUse when the user needs to manage Docker Compose services.\n\n## Common Operations\n\n\`\`\`bash\ndocker compose up -d\ndocker compose logs -f service_name\ndocker compose build --no-cache\ndocker compose down -v\n\`\`\``, }, { name: 'kubernetes', category: 'devops', description: 'Kubernetes cluster management: pods, services, deployments, scaling', version: '1.1.0', platforms: ['macos', 'linux'], tags: ['Kubernetes', 'k8s', 'pods', 'deployments'], body: `## When to Use\n\nUse for Kubernetes operations: inspecting pods, scaling deployments, managing services.\n\n## Prerequisites\n\n- \`kubectl\` installed and configured\n- Valid kubeconfig\n\n## Common Patterns\n\n\`\`\`bash\nkubectl get pods -n namespace\nkubectl logs -f pod-name\nkubectl scale deployment/name --replicas=3\n\`\`\``, }, { name: 'jupyter-notebooks', category: 'data-science', description: 'Create, edit, and execute Jupyter notebooks programmatically', version: '1.0.0', platforms: ['macos', 'linux', 'windows'], tags: ['Jupyter', 'notebooks', 'Python', 'data analysis'], body: `## When to Use\n\nUse when the user wants to create data analysis notebooks, run experiments, or visualize data.\n\n## Capabilities\n\n- Create new notebooks from scratch\n- Add/edit/delete cells\n- Execute cells and capture output\n- Export to HTML/PDF\n- Manage kernels`, }, { name: 'mcp-server', category: 'mcp', description: 'Run Hermes as an MCP server or connect to external MCP tools', version: '1.0.0', platforms: ['macos', 'linux', 'windows'], tags: ['MCP', 'protocol', 'server', 'integration'], body: `## When to Use\n\nUse when you need to expose Hermes tools to other AI systems or consume tools from external MCP servers.\n\n## As Server\n\n\`\`\`bash\nhermes --mcp-server --port 8080\n\`\`\`\n\n## As Client\n\nConfigure in \`~/.hermes/config.yaml\`:\n\`\`\`yaml\nmcp_servers:\n my-server:\n command: "npx"\n args: ["-y", "@my/mcp-server"]\n\`\`\`\n\nTools from MCP servers are auto-registered in the tool registry.`, }, { name: 'home-assistant', category: 'smart-home', description: 'Control smart home devices via Home Assistant REST API', version: '1.0.0', platforms: ['macos', 'linux', 'windows'], tags: ['Home Assistant', 'smart home', 'IoT', 'automation'], body: `## When to Use\n\nUse when the user wants to control smart home devices: lights, thermostats, locks, media players.\n\n## Tools\n\n- \`ha_list_entities\` -- browse devices by domain or area\n- \`ha_get_state\` -- check device status\n- \`ha_list_services\` -- available actions\n- \`ha_call_service\` -- execute actions (turn_on, set_temperature, etc.)\n\n## Setup\n\nRequires \`HA_URL\` and \`HA_TOKEN\` environment variables.`, }, { name: 'mixture-of-agents', category: 'research', description: 'Multi-layer LLM collaboration using the Mixture-of-Agents architecture', version: '0.5.0', platforms: ['macos', 'linux', 'windows'], tags: ['MoA', 'multi-agent', 'collaboration', 'reasoning'], body: `## When to Use\n\nUse for complex reasoning tasks where multiple perspectives improve output quality.\n\n## How It Works\n\nBased on the Mixture-of-Agents paper (Junlin Wang et al.):\n\n1. Multiple LLMs generate initial responses\n2. Responses are aggregated and refined\n3. A final synthesis produces the output\n\n## Configuration\n\nUse the \`moa\` toolset to enable.`, }, { name: 'rl-training', category: 'mlops', description: 'Reinforcement learning training via Tinker-Atropos environments', version: '0.3.0', platforms: ['macos', 'linux'], tags: ['RL', 'training', 'Tinker', 'Atropos', 'WandB'], body: `## When to Use\n\nUse when fine-tuning models with RL or running training experiments.\n\n## Tools\n\n- \`rl_list_environments\` -- discover available training environments\n- \`rl_select_environment\` -- choose environment\n- \`rl_edit_config\` -- modify hyperparameters\n- \`rl_start_training\` -- launch training run\n- \`rl_check_status\` -- monitor progress (WandB integration)\n- \`rl_get_results\` -- retrieve trained model outputs`, }, ] export const plugins: WikiPlugin[] = [ { name: 'agent-monitor', version: '1.0.0', description: 'Real-time session event streaming to external monitoring dashboards and observability tools via Unix socket', tools: [], hooks: ['on_session_start', 'pre_tool_call', 'post_tool_call', 'pre_llm_call', 'post_llm_call', 'on_session_end'] }, { name: 'desktop-control', version: '0.1.0', description: 'Autonomous desktop control via vision model and PyAutoGUI for GUI tasks that cannot be done through terminal', tools: ['desktop_use'], hooks: [] }, { name: 'notification-relay', version: '0.4.0', description: 'Push notifications to Slack, Teams, or webhooks when agents need attention or complete tasks', tools: ['notify_send'], hooks: ['on_session_start', 'on_session_end'] }, { name: 'sandbox-runner', version: '0.3.0', description: 'Execute untrusted code in isolated Docker containers with resource limits and network policies', tools: ['sandbox_exec', 'sandbox_status'], hooks: ['pre_tool_call'] }, { name: 'audit-logger', version: '0.6.0', description: 'Comprehensive audit trail -- logs every tool call, LLM interaction, and approval decision to structured JSON', tools: [], hooks: ['on_session_start', 'pre_tool_call', 'post_tool_call', 'pre_llm_call', 'post_llm_call', 'on_session_end'] }, { name: 'cost-tracker', version: '0.2.0', description: 'Token usage and cost tracking per session, per agent, and per project with budget alerts', tools: ['cost_report'], hooks: ['post_llm_call', 'on_session_end'] }, { name: 'cron-scheduler', version: '0.5.0', description: 'Scheduled task execution -- run agent tasks on cron schedules with retry logic and failure alerts', tools: ['cron_create', 'cron_list', 'cron_delete'], hooks: ['on_session_start'] }, { name: 'gateway-bridge', version: '1.0.0', description: 'Cross-platform messaging gateway -- connects Hermes to Telegram, Discord, Slack, Signal, Email, SMS, and 8+ other platforms', tools: ['send_message'], hooks: ['on_session_start', 'on_session_end'] }, ] export const tools: WikiTool[] = [ // File Operations { name: 'read_file', description: 'Read file contents with optional line range', category: 'File Operations', params: [{ name: 'file_path', type: 'string', required: true, description: 'Path to the file' }, { name: 'offset', type: 'integer', required: false, description: 'Start line' }, { name: 'limit', type: 'integer', required: false, description: 'Number of lines' }] }, { name: 'write_file', description: 'Create or overwrite a file with safety checks', category: 'File Operations', params: [{ name: 'file_path', type: 'string', required: true, description: 'File to write' }, { name: 'content', type: 'string', required: true, description: 'Content to write' }] }, { name: 'patch', description: 'Apply fuzzy-matched patches to code files', category: 'File Operations', params: [{ name: 'file_path', type: 'string', required: true, description: 'File to patch' }, { name: 'old_string', type: 'string', required: true, description: 'Text to find' }, { name: 'new_string', type: 'string', required: true, description: 'Replacement text' }] }, { name: 'search_files', description: 'Search file contents and names with regex', category: 'File Operations', params: [{ name: 'pattern', type: 'string', required: true, description: 'Regex pattern' }, { name: 'path', type: 'string', required: false, description: 'Directory to search' }] }, // Terminal { name: 'terminal', description: 'Execute shell commands in local, Docker, Modal, SSH, or Daytona environments', category: 'Terminal', params: [{ name: 'command', type: 'string', required: true, description: 'Command to run' }, { name: 'timeout', type: 'integer', required: false, description: 'Timeout in seconds' }, { name: 'background', type: 'boolean', required: false, description: 'Run in background' }] }, { name: 'process', description: 'Manage background processes with polling, output buffering, and crash recovery', category: 'Terminal', params: [{ name: 'action', type: 'string', required: true, description: 'start, stop, status, or output' }, { name: 'process_id', type: 'string', required: false, description: 'Target process ID' }] }, // Web { name: 'web_search', description: 'Search the web via Exa, Firecrawl, Tavily, or parallel multi-engine', category: 'Web', params: [{ name: 'query', type: 'string', required: true, description: 'Search query' }, { name: 'backend', type: 'string', required: false, description: 'exa, firecrawl, tavily, parallel' }] }, { name: 'web_extract', description: 'Extract and parse web page content to markdown with compression', category: 'Web', params: [{ name: 'url', type: 'string', required: true, description: 'URL to extract' }] }, // Browser { name: 'browser_navigate', description: 'Navigate to a URL in headless browser', category: 'Browser', params: [{ name: 'url', type: 'string', required: true, description: 'URL to navigate to' }] }, { name: 'browser_snapshot', description: 'Take accessibility tree snapshot of the current page', category: 'Browser', params: [] }, { name: 'browser_click', description: 'Click an element on the page', category: 'Browser', params: [{ name: 'element', type: 'string', required: true, description: 'Element selector or aria ref' }] }, { name: 'browser_type', description: 'Type text into a focused input', category: 'Browser', params: [{ name: 'text', type: 'string', required: true, description: 'Text to type' }] }, { name: 'browser_vision', description: 'Analyze the rendered page with a vision model', category: 'Browser', params: [{ name: 'prompt', type: 'string', required: false, description: 'Analysis prompt' }] }, // AI & Reasoning { name: 'delegate_task', description: 'Spawn an isolated subagent with restricted toolsets and focused prompt', category: 'AI & Reasoning', params: [{ name: 'task', type: 'string', required: true, description: 'Task description' }, { name: 'toolsets', type: 'string', required: false, description: 'Comma-separated toolsets for subagent' }] }, { name: 'mixture_of_agents', description: 'Multi-layer LLM collaboration for complex reasoning', category: 'AI & Reasoning', params: [{ name: 'prompt', type: 'string', required: true, description: 'The question or task' }] }, { name: 'execute_code', description: 'Run Python scripts that call tools via RPC, collapsing multi-step pipelines', category: 'AI & Reasoning', params: [{ name: 'code', type: 'string', required: true, description: 'Python code to execute' }] }, // Vision & Media { name: 'vision_analyze', description: 'Analyze images from URLs or files with custom prompts', category: 'Vision & Media', params: [{ name: 'image_url', type: 'string', required: true, description: 'URL or path to image' }, { name: 'prompt', type: 'string', required: false, description: 'Analysis prompt' }] }, { name: 'image_generate', description: 'Generate images using FAL.ai FLUX 2 Pro with upscaling', category: 'Vision & Media', params: [{ name: 'prompt', type: 'string', required: true, description: 'Image description' }, { name: 'upscale', type: 'boolean', required: false, description: 'Apply Clarity Upscaler' }] }, { name: 'text_to_speech', description: 'Convert text to speech via Edge TTS, ElevenLabs, OpenAI, or local NeuTTS', category: 'Vision & Media', params: [{ name: 'text', type: 'string', required: true, description: 'Text to speak' }, { name: 'backend', type: 'string', required: false, description: 'edge, elevenlabs, openai, neutts' }] }, // Memory & Planning { name: 'memory', description: 'Persistent file-backed memory injected into system prompt each session', category: 'Memory & Planning', params: [{ name: 'action', type: 'string', required: true, description: 'add, replace, remove, or read' }, { name: 'content', type: 'string', required: false, description: 'Content to write' }] }, { name: 'session_search', description: 'FTS5 search over past session transcripts with LLM summarization', category: 'Memory & Planning', params: [{ name: 'query', type: 'string', required: true, description: 'Search query' }, { name: 'limit', type: 'integer', required: false, description: 'Max results' }] }, { name: 'todo', description: 'In-memory task list for multi-step decomposition, persistent across compression', category: 'Memory & Planning', params: [{ name: 'action', type: 'string', required: true, description: 'add, complete, remove, or list' }, { name: 'task', type: 'string', required: false, description: 'Task description' }] }, // Smart Home { name: 'ha_call_service', description: 'Execute Home Assistant service calls (turn_on, set_temperature, etc.)', category: 'Smart Home', params: [{ name: 'domain', type: 'string', required: true, description: 'Service domain (light, climate, etc.)' }, { name: 'service', type: 'string', required: true, description: 'Service name' }, { name: 'entity_id', type: 'string', required: true, description: 'Target entity' }] }, { name: 'ha_get_state', description: 'Get detailed state of a Home Assistant device', category: 'Smart Home', params: [{ name: 'entity_id', type: 'string', required: true, description: 'Entity to query' }] }, // Messaging { name: 'send_message', description: 'Send messages across Telegram, Discord, Slack, Signal, Email, SMS, and more', category: 'Messaging', params: [{ name: 'platform', type: 'string', required: true, description: 'Target platform' }, { name: 'recipient', type: 'string', required: true, description: 'Channel, user, or address' }, { name: 'message', type: 'string', required: true, description: 'Message content' }] }, // Scheduling { name: 'cronjob', description: 'Create and manage scheduled tasks with natural language schedules', category: 'Scheduling', params: [{ name: 'action', type: 'string', required: true, description: 'create, list, update, pause, resume, remove, trigger' }, { name: 'schedule', type: 'string', required: false, description: 'Cron expression or natural language' }, { name: 'task', type: 'string', required: false, description: 'Task to execute' }] }, ] export const commands: WikiCommand[] = [ { command: 'hermes', description: 'Start an interactive Hermes agent session', flags: ['--tools ', '--model ', '--auto', '--personality ', '--reasoning-effort '] }, { command: 'hermes --mcp-server', description: 'Run Hermes as an MCP (Model Context Protocol) server', flags: ['--port ', '--transport '] }, { command: 'hermes batch', description: 'Run a batch of tasks from a file', flags: ['--input ', '--output ', '--parallel '] }, { command: 'hermes config', description: 'View or edit configuration', flags: ['--edit', '--reset', '--show'] }, { command: 'hermes plugins', description: 'List and manage installed plugins', flags: ['--install ', '--remove ', '--list'] }, { command: 'hermes skills', description: 'Browse and manage available skills', flags: ['--list', '--search ', '--info '] }, { command: 'hermes memory', description: 'Inspect persistent memory stores', flags: ['--show', '--clear', '--export '] }, { command: 'hermes auth', description: 'Configure API credentials and providers', flags: ['--setup', '--status', '--rotate'] }, { command: 'hermes cron', description: 'Manage scheduled tasks', flags: ['--list', '--create', '--delete ', '--logs '] }, { command: 'hermes update', description: 'Update Hermes to the latest version', flags: ['--check', '--force'] }, { command: 'hermes gateway', description: 'Start the cross-platform messaging gateway', flags: ['--platform ', '--config '] }, ] export const architecture = `# Architecture ## Core Loop \`\`\` User Input | v Orchestrator / Planner | ├── Skill Selection (SKILL.md matching) ├── Tool Dispatch (registry.py) │ ├── pre_tool_call hooks │ ├── Safety check (approval.py, 40+ patterns) │ ├── Tool execution │ └── post_tool_call hooks ├── LLM Call (streaming, multi-provider) │ ├── pre_llm_call hooks │ ├── API call via OpenRouter / Nous / Anthropic / Custom │ └── post_llm_call hooks └── Response ├── Text output (with optional TTS) └── Tool calls (loop back) \`\`\` ## Provider Resolution Hermes auto-detects available AI providers in priority order: 1. **OpenRouter** -- OPENROUTER_API_KEY 2. **Nous Portal** -- auth.json active provider 3. **Custom endpoint** -- config.yaml base_url 4. **Codex OAuth** -- gpt-5.3-codex via Responses API 5. **Native Anthropic** -- ANTHROPIC_API_KEY 6. **Direct keys** -- z.ai, Kimi, MiniMax, etc. ## Memory System ### Session Memory (Built-in) - MEMORY.md -- agent observations, project conventions, tool quirks - USER.md -- user preferences and communication style - Frozen snapshot at session start (prefix cache stability) - Mid-session writes update disk but not system prompt ### Session Search - FTS5 full-text search over all past session transcripts - LLM summarization for relevant context retrieval - SQLite-backed with relevance ranking ### Memory Providers (Switchable) 8 backends available via \`memory.provider\` in config: Honcho, OpenViking, Mem0, ByteRover, Hindsight, Holographic, RetainDB, Supermemory ## Safety System ### Auto Mode - Risk classifier on every tool call - 40+ regex patterns for dangerous commands - Smart approval via auxiliary LLM for ambiguous cases - Manual fallback for high-risk operations - Tirith security scanner for pre-execution analysis ### Approval Flow 1. Agent proposes tool call 2. \`detect_dangerous_command()\` pattern matching 3. Dangerous: block or route to human approval 4. Safe: execute immediately 5. Per-session allowlist + permanent config allowlist ## Terminal Backends The terminal tool supports multiple execution environments: - **Local** -- direct shell execution - **Docker** -- containerized commands - **Modal** -- serverless compute - **SSH** -- remote host execution - **Singularity** -- HPC container runtime - **Daytona** -- cloud dev environments ## Browser System Headless browser automation with two backends: - **Browserbase** -- cloud-hosted browsers - **Local Chromium** -- headless via Playwright Uses accessibility tree snapshots (ariaSnapshot) instead of raw HTML for reliable element targeting. ## Plugin System Plugins register via \`register(ctx)\`: - \`ctx.register_tool()\` -- add tools to the registry - \`ctx.register_hook()\` -- subscribe to lifecycle events - \`ctx.inject_message()\` -- send messages into active sessions ### Hook Lifecycle | Hook | When | Use Case | |------|------|----------| | on_session_start | Session begins | Load context, initialize state | | pre_llm_call | Before API call | Inject memories, modify context | | post_llm_call | After response | Capture decisions, log metrics | | pre_tool_call | Before tool exec | Audit, sandboxing, monitoring | | post_tool_call | After tool exec | Result processing, telemetry | | on_session_end | Session ends | Summarize, persist, cleanup | ## Gateway System Cross-platform messaging with 13+ connectors: Telegram, Discord, Slack, WhatsApp, Signal, Email, SMS, Mattermost, Matrix, DingTalk, Feishu/Lark, WeCom, Webhook All platforms share the same agent loop and tools with conversation continuity. ## MCP Integration Hermes both serves and consumes MCP: - **Server mode**: expose Hermes tools to other AI systems - **Client mode**: auto-discover and register tools from external MCP servers - Configured via \`mcp_servers\` in config.yaml - Supports stdio and HTTP/StreamableHTTP transports ` export const changelog = `# Changelog ## v0.7.0 (2026-04-01) - Computer Use skill via Anthropic beta API - Plugin \`inject_message()\` for external message injection - Auto Mode safety: 12 new dangerous patterns - Credential pool rotation for custom endpoints - Tinker-Atropos RL training environment integration - WandB metrics logging for RL runs ## v0.6.2 (2026-03-15) - Honcho cross-session context integration - Compression threshold now configurable - 8 memory provider backends (switchable via config) - Fix: memory snapshot race condition ## v0.6.0 (2026-03-01) - MCP server mode (Model Context Protocol) - MCP client: auto-discover tools from external servers - Cron scheduler for automated tasks - Gateway: WhatsApp, Signal, Mattermost, Matrix connectors - 20+ new bundled skills - Tirith security scanner integration ## v0.5.0 (2026-02-01) - Session search (FTS5 over past transcripts) - execute_code tool (Python RPC bridge) - Mixture of Agents (multi-LLM collaboration) - Process management tool (background tasks) - Skills hub integration (agentskills.io) ## v0.4.0 (2026-01-15) - Auto Mode with risk classification - Smart approval via auxiliary LLM - Gateway: Telegram, Discord, Slack, Email, SMS - Browser automation (Browserbase + local Playwright) - Home Assistant smart home tools ## v0.3.0 (2025-12-01) - Persistent memory (MEMORY.md + USER.md) - Skill auto-discovery and platform matching - Configuration system (config.yaml) - SOUL.md personality files - Image generation (FAL.ai FLUX 2 Pro) - Text-to-speech (Edge TTS, ElevenLabs, OpenAI) ## v0.2.0 (2025-11-01) - Plugin system with hook registration - Tool registry with availability checks - Context compression - Multi-provider support (OpenRouter, Anthropic, custom) - Vision analysis tool - Streaming responses ## v0.1.0 (2025-10-01) - Initial release - Core agent loop with orchestrator - Terminal, file, and web tools - Basic skill system - Todo planning tool ` export const mockConfig = `# Hermes Agent Configuration # ~/.hermes/config.yaml agent: max_turns: 60 reasoning_effort: medium display: personality: default skin: default streaming: true show_reasoning: false memory: memory_enabled: true memory_char_limit: 2200 user_char_limit: 1375 nudge_interval: 10 provider: null # honcho, mem0, openviking, etc. security: redact_secrets: true tirith_enabled: true tirith_timeout: 5 tirith_fail_open: true skills: disabled: [] external_dirs: [] plugins: disabled: [] terminal: backend: local # local, docker, modal, ssh, daytona timeout: 180 model: default: openai/gpt-4o provider: openrouter compression: enabled: true target_ratio: 0.2 threshold: 0.85 mcp_servers: {} # example: # my-server: # command: "npx" # args: ["-y", "@my/mcp-server"] ` export const mockMemory = { memory: `# Agent Memory ## Project Conventions - This monorepo uses pnpm workspaces - Tests run with vitest, not jest - CSS modules preferred over styled-components - API routes follow /api/v1/{resource} pattern ## Tool Quirks - search_files with multiline needed for cross-line patterns - patch tool fails silently if old_string not unique -- provide enough context - terminal timeout defaults to 180s, use explicit timeout for long builds ## Learnings - User prefers concise responses with no trailing summaries - Always check git status before making commits - Run lint before build to catch errors early `, user: `# User Profile ## Role Senior full-stack engineer working on a SaaS platform. ## Preferences - Terse communication style - Prefers explicit over implicit - Values production-ready code over demos - Likes monospace fonts and dark themes ## Tech Stack - TypeScript, React, Node.js - PostgreSQL, Redis - Docker, Kubernetes - Hono for API routes, Zod for validation `, } export const mockSoul = `# Hermes Agent Persona You are Hermes, a precise and capable AI assistant. ## Communication Style - Direct and concise - Technical accuracy over politeness - Show reasoning when it adds value - Admit uncertainty instead of guessing ## Principles - Production code, not demos - Simplest solution that works - Delete dead code immediately - DRY after the second occurrence - Composition over inheritance `