Beginner

curl -fsSL https://raw.githubusercontent.com/NousResearch/hermes-agent/main/scripts/install.sh | bash
                                        # => Only prerequisite: Git pre-installed
                                        # => Auto-installs: Python 3.11, Node.js v22, uv, ripgrep, ffmpeg
 
source ~/.bashrc                        # => Makes `hermes` command available in PATH
 
hermes --version                        # => Output: hermes-agent v0.15.2
                                        # => Confirms successful installation
 
hermes doctor                           # => Runs post-install health check
                                        # => Validates all dependencies; shows pass/fail per component

hermes setup                            # => Launches interactive setup wizard
                                        # => Steps: provider → API key → default model
                                        # => Creates ~/.hermes/ with config.yaml + .env
 
ls ~/.hermes/                           # => config.yaml (settings), .env (secrets),
                                        # =>   memories/ (MEMORY.md, USER.md), SOUL.md
 
hermes model                            # => Shows current model selection
                                        # => Output: "Current model: anthropic/claude-sonnet-4-6"
 
hermes model anthropic/claude-opus-4    # => Switches to a different model immediately
                                        # => Takes effect for the current session

hermes                                  # => Launches interactive TUI session (= hermes chat)
                                        # => Multiline editing, tab autocomplete, streaming
                                        # => Press Ctrl+D to quit cleanly
 
hermes chat                             # => Identical to bare `hermes`
                                        # => Status bar: token count + session cost
                                        # => Ctrl+C interrupt, /new reset, Ctrl+D quit
 
# Inside the TUI — type prompts and press Enter:
> What is Rust?                         # => LLM streams response token-by-token in real-time
> Explain ownership in Rust             # => Status bar updates tokens + cost per turn
                                        # => Each turn adds to cumulative session cost
> /new                                  # => Resets context; reloads MEMORY.md and USER.md
> /usage                                # => Output: "Input: 1,204 tokens | Session cost: $0.012"

hermes chat -q "What is Rust?"          # => Sends single query to configured LLM
                                        # => Exits automatically after response
 
hermes chat -q "Summarize this" < README.md
                                        # => Pipes file content as stdin context
                                        # => LLM receives file + prompt together
 
hermes chat --toolsets "web" -q "search for Rust async runtimes"
                                        # => Restricts available tools to web only
                                        # => Other toolsets: terminal, file, vision
 
echo "explain: $(cat error.log)" | hermes chat -q -
                                        # => Reads prompt from stdin via dash flag
                                        # => Combines shell interpolation with AI

hermes chat -q "What is 2+2?" --reasoning-effort none
                                        # => No chain-of-thought; fastest + cheapest
                                        # => Output: "4" — best for trivial lookups
 
hermes chat -q "Explain mutex vs semaphore" --reasoning-effort minimal
                                        # => Minimal reasoning; brief accurate answer
 
hermes chat -q "Debug this null pointer" --reasoning-effort low
                                        # => Light reasoning; catches obvious issues
 
hermes chat -q "Design a cache layer" --reasoning-effort medium
                                        # => Moderate depth; considers trade-offs
                                        # => Good default for general development tasks
 
hermes chat -q "Architect a payment system" --reasoning-effort high
                                        # => Extended reasoning; covers security + scaling
 
hermes chat -q "Find the race condition" --reasoning-effort xhigh
                                        # => Maximum depth; exhaustive code path analysis
                                        # => Highest token usage and latency

hermes doctor                           # => Runs full diagnostic suite
                                        # => Checks Python, Node.js, uv, ripgrep, ffmpeg,
                                        # =>   config.yaml syntax, and API key auth
                                        # => Output: pass/fail for each check
 
hermes status                           # => Shows active provider, masked API keys,
                                        # =>   enabled toolsets, memory file sizes
 
hermes update                           # => Updates to latest version
                                        # => Preserves ~/.hermes/ config and memories
 
hermes --version                        # => Output: hermes-agent v0.15.2
hermes doctor --fix                     # => Auto-remediate fixable dependency issues

ls -la ~/.hermes/                       # => config.yaml (settings), .env (secrets),
                                        # =>   auth.json (OAuth), SOUL.md (personality),
                                        # =>   memories/ (MEMORY.md + USER.md)
 
cat ~/.hermes/config.yaml               # => YAML format; all non-secret settings
                                        # => Changes take effect on next session start
                                        # => Do not store secrets here — use .env instead
 
cat ~/.hermes/.env                      # => API keys (one KEY=value per line)
                                        # => File permissions: chmod 600 ~/.hermes/.env
cat ~/.hermes/SOUL.md                   # => Agent personality definition file

# ~/.hermes/config.yaml
# => Model provider configuration
# => Controls which LLM processes your conversations
 
model:
  provider:
    anthropic # => LLM provider to use
    # => Options: anthropic, openrouter, nous,
    # =>          openai, copilot, custom
  model:
    claude-sonnet-4-6 # => Specific model within provider
    # => Anthropic: claude-opus-4, claude-sonnet-4-6
    # => OpenRouter: openai/gpt-4o, google/gemini-2.5-pro
  base_url:
    "" # => Custom API endpoint (leave empty for default)
    # => Used with provider: custom
    # => Example: http://localhost:11434/v1
  api_key:
    "${ANTHROPIC_API_KEY}" # => References key from .env file
    # => ${VAR} syntax for environment substitution
 
  # Fallback model configuration
  fallback:
    provider: openrouter # => Fallback provider if primary fails
    model:
      openai/gpt-4o # => Fallback model selection
      # => Activated on: rate limit, timeout, API error
      # => Hermes retries once on primary before switching

# ~/.hermes/.env — one KEY=value per line; loaded at session start
# => chmod 600 ~/.hermes/.env to restrict read access to your user only
ANTHROPIC_API_KEY=sk-ant-api03-xxxxx    # => Anthropic Claude key (console.anthropic.com)
                                        # => Required when provider: anthropic in config.yaml
OPENROUTER_API_KEY=sk-or-v1-xxxxx       # => OpenRouter key — access 200+ models (openrouter.ai/keys)
OPENAI_API_KEY=sk-proj-xxxxx            # => OpenAI key (platform.openai.com)
NOUS_API_KEY=nk-xxxxx                   # => Nous Research key (inference.nous.ai)

# ~/.hermes/config.yaml
# => Precedence: CLI args > config.yaml > .env vars > built-in defaults
 
model:
  api_key:
    "${ANTHROPIC_API_KEY}" # => ${VAR} resolved from .env at runtime
    # => Clear error message if variable is unset
    # => Supports any environment variable name you define

# ~/.hermes/config.yaml
 
display:
  # => Controls all visual output settings in the TUI
  tool_progress:
    true # => Show tool execution progress bars
    # => Displays: "Running terminal..." with spinner
    # => false: hides progress; useful for scripted output
  streaming:
    true # => Stream tokens as LLM generates them
    # => false: wait for complete response before display
    # => Streaming requires SSE-compatible terminal
  show_reasoning:
    true # => Display chain-of-thought reasoning
    # => Shows the LLM's thinking process before final answer
    # => Disable to reduce output length in automated pipelines
  show_cost:
    true # => Display cost per message and session total
    # => Output: "$0.003 (session: $0.15)"
    # => Helps catch expensive prompts before they add up
  skin:
    default # => Visual theme for TUI
    # => Options: default, minimal, colorful
    # => minimal: no colors, good for terminal recording
  personality:
    default # => Display personality name in status bar
    # => Shows which SOUL.md persona is active
    # => Useful when switching personalities mid-session
  compact:
    false # => Compact mode reduces whitespace
    # => true: denser output, less scrolling
    # => Recommended for small terminal windows
 
streaming:
  enabled:
    true # => Master switch for streaming
    # => false: all responses arrive as complete blocks
    # => Set false when piping output to another tool
  transport:
    sse # => Streaming protocol
    # => Options: sse (Server-Sent Events), websocket
    # => sse is default; websocket available for some providers
  edit_interval:
    100 # => Milliseconds between display updates
    # => Lower = smoother but higher CPU usage
    # => 100ms is a good default for most terminals
  cursor:
    true # => Show blinking cursor during generation
    # => Visual indicator that LLM is still working
    # => false: no cursor; cleaner for non-interactive output

# ~/.hermes/config.yaml
 
agent:
  max_turns:
    90 # => Maximum conversation turns per session
    # => Prevents runaway agent loops
    # => Default: 90 (sufficient for most tasks)
    # => Set lower for cost control
 
  reasoning_effort:
    medium # => Default reasoning effort level
    # => Options: none, minimal, low, medium, high, xhigh
    # => Overridable per-message with --reasoning-effort
    # => medium: good balance of speed and depth
 
  tool_use_enforcement:
    auto # => How strictly agent must use tools
    # => auto: agent decides when tools are needed
    # => always: force tool use every turn
    # => never: disable all tool usage
 
compression:
  enabled:
    true # => Enable automatic context compression
    # => Reduces context when approaching token limit
  threshold:
    0.50 # => Compress when context exceeds 50% of window
    # => Range: 0.0 (compress immediately) to 1.0 (never)
  target_ratio:
    0.30 # => Target context size after compression
    # => 0.30 = compress to 30% of window
  protect_last_n:
    5 # => Protect last N messages from compression
    # => Keeps recent context intact
    # => Older messages get summarized
    # => Increase if recent context is often compressed away

# ~/.hermes/config.yaml
 
human_delay:
  mode:
    "off" # => off: no delays (fastest); natural: randomized; custom: use min/max_ms
    # => Change mode without restarting — edit config.yaml and restart session
  min_ms:
    500 # => Minimum pause in milliseconds (custom mode only)
    # => Values below 200ms are too fast to feel human
  max_ms:
    2000 # => Maximum pause; random delay in [min_ms, max_ms] simulates typing speed
    # => 2000ms upper bound matches average human typing speed for short messages

# Switch delay mode without restarting (edit config.yaml, then restart session):
# => Edit config.yaml, save it, then kill and relaunch hermes
hermes chat                             # => mode: off — responds instantly (development)
                                        # => Best for CI/CD pipelines and scripted use
hermes chat                             # => mode: natural — pauses ~800ms read + ~1200ms compose
                                        # => Best for Slack/Telegram channels; feels human
hermes chat                             # => mode: custom — random delay in [min_ms, max_ms]
                                        # => Best when natural feels too slow or too fast
# Natural mode timeline: receive at t=0 → pause 800ms → type → pause 1200ms → send at t=2s

# ~/.hermes/config.yaml
 
quick_commands:
  status:
    "systemctl status hermes-agent"
    # => /status in chat runs this command
    # => Zero tokens consumed (no LLM involved)
    # => Output displayed directly in TUI
 
  disk:
    "df -h /" # => /disk shows filesystem usage
    # => Instant execution, no AI processing
 
  ports:
    "lsof -i -P -n | head -20" # => /ports lists open network ports
    # => Useful for debugging connectivity
 
  logs:
    "tail -50 /var/log/syslog" # => /logs shows recent system logs
    # => Quick access without typing full path
 
  gpu:
    "nvidia-smi" # => /gpu shows GPU utilization
    # => Useful when running local models

# Inside hermes TUI session:
/status                                 # => Executes: systemctl status hermes-agent
                                        # => Output: service status information
                                        # => Zero tokens consumed
                                        # => No LLM roundtrip — instant execution
 
/disk                                   # => Executes: df -h /
                                        # => Output: filesystem usage table
                                        # => Faster than asking "how much disk space"

# ~/.hermes/config.yaml
# => Terminal tool configuration
 
tools:
  terminal:
    backend:
      local # => Execution backend
      # => local: runs on host machine directly
      # => Options: local, docker, ssh
    timeout:
      180 # => Command timeout in seconds
      # => Kills command after 180s (3 minutes)
      # => Prevents: infinite loops, hung processes
    cwd:
      "~" # => Default working directory
      # => Agent starts commands here
      # => Can cd to other directories as needed

# In hermes TUI session:
> Check what Go version is installed
 
# => Agent decides to use terminal tool
# => Agent runs: go version
# => Output: go version go1.23.0 darwin/arm64
# => Agent reports: "You have Go 1.23.0 installed on macOS ARM64"
 
> Find all TODO comments in the src/ directory
 
# => Agent runs: rg "TODO" src/ --line-number
# => Output: list of files and lines with TODO comments
# => Agent summarizes: "Found 7 TODO comments across 4 files"

# In hermes TUI session:
> Read the contents of package.json     # => Agent: read_file("package.json")
                                        # => Returns full file contents to agent context
 
> Create a .gitignore for a Node.js project
                                        # => Agent: write_file(".gitignore", "node_modules/\ndist/\n...")
                                        # => Output: "Created .gitignore with 12 patterns"
 
> Change the port from 3000 to 8080 in server.ts
                                        # => Agent: patch("server.ts", old="port: 3000", new="port: 8080")
                                        # => Targeted edit — only the matched text changes
 
> Find all files importing the auth module
                                        # => Agent: search_files("from.*auth", "src/") via ripgrep
                                        # => Output: "Found 5 files importing auth module"

# ~/.hermes/config.yaml
 
web:
  backend: firecrawl # => Options: firecrawl (extraction), tavily (search), exa, parallel

# In hermes TUI session:
> Search for the latest Rust release notes
                                        # => web_search("Rust programming language latest release")
                                        # => Returns top results; agent summarizes: "Rust 1.82 was released on..."
 
> Extract the content from https://doc.rust-lang.org/book/ch04-01.html
                                        # => web_extract(url) fetches page and strips navigation/ads
                                        # => Returns clean structured text for agent to analyze
 
> What does the Axum 0.8 changelog say about breaking changes?
                                        # => Agent searches + extracts relevant docs automatically
                                        # => Combines web_search and web_extract in one turn

# In hermes TUI session:
> Start a development server in the background
                                        # => process.start("npm run dev") — PID 12345 returned
                                        # => Agent continues conversation while server runs in background
 
> Also start the TypeScript watcher     # => process.start("tsc --watch") — PID 12389 returned
 
> Check the status of running processes
                                        # => process.status() output:
                                        # =>   PID 12345 | running | npm run dev  (2m 15s)
                                        # =>   PID 12389 | running | tsc --watch  (1m 30s)
 
> Stop the development server           # => process.kill(12345) — SIGTERM → graceful shutdown
                                        # => Output: "Stopped process 12345: npm run dev"

hermes chat -q "Refactor the auth module in src/ to use JWT tokens"
                                        # => Agent breaks task into 7 tracked steps automatically
                                        # =>   [ ] Read auth, [ ] Find dependents, [ ] Install jwt, [ ] Create utils, [ ] Update handler/middleware/tests
hermes chat -q "List your current todos"
                                        # => Agent prints todo list with status markers
                                        # => [x] = done, [ ] = pending, → = in progress
hermes chat -q "Mark the JWT utils step done and continue"
                                        # => Agent updates: [x] Create JWT utility functions
                                        # => Resumes from current step if context was reset

hermes                                  # => Opens TUI; ask: "What are you working on?"
                                        # => Agent reports live todo status with step counts:
                                        # =>   [x] Read current auth implementation
                                        # =>   [x] Identify dependents (found 8 files)
                                        # =>   [x] Install jsonwebtoken
                                        # =>   [ ] Create JWT utility functions  ← current
                                        # =>   [ ] Update login handler, middleware, tests
 
hermes chat -q "Show only pending todos"
                                        # => Agent filters: [ ] Create JWT utils, [ ] Update handler
hermes chat -q "Skip the middleware step and go straight to tests"
                                        # => Agent removes skipped step and advances todo cursor
hermes chat -q "Mark all remaining todos done"
                                        # => Agent marks all [ ] as [x]; task complete
hermes chat -q "Summarize what was accomplished"
                                        # => Agent reads completed todos and writes a summary

# ~/.hermes/config.yaml
 
auxiliary:
  vision:
    enabled:
      true # => Enable image analysis capabilities
      # => Agent can interpret screenshots, diagrams
    backend:
      default # => Vision processing backend
      # => Uses model's native vision when available
  image_generation:
    enabled:
      true # => Enable image creation
      # => Requires fal.ai API key in .env
    backend:
      fal # => Image generation provider
      # => fal: fal.ai (fast, high quality)

# In hermes TUI session:
> Analyze this screenshot of the error
 
# => Agent uses: vision_analyze("screenshot.png")
# => Processes image through vision model
# => Identifies: error message text, stack trace, UI context
# => Output: "The error shows a TypeError: cannot read property
# =>          'map' of undefined on line 42 of UserList.tsx"
 
> Generate an architecture diagram for a microservices system
 
# => Agent uses: image_generate("architecture diagram showing
# =>   API gateway connecting to auth, users, and orders
# =>   microservices with message queue between them")
# => Creates image via fal.ai
# => Saves to local file
# => Output: "Generated architecture diagram: arch-diagram.png"

hermes tools                            # => Interactive toggle — shows all toolsets:
                                        # =>   [x] terminal, [x] file, [x] web, [x] todo, [x] memory
                                        # =>   [ ] process, [ ] vision (disabled by default)
 
hermes chat --toolsets "web,file"       # => Session-scoped: only web + file tools active
                                        # => Terminal and process tools disabled for safety
 
hermes chat --toolsets "terminal,file,todo"
                                        # => Full coding session: terminal + file + task tracking
 
hermes chat --toolsets ""               # => Disable all tools; LLM-only mode (no tool calls)
 
hermes chat -q "review this code" --toolsets "file"
                                        # => Code review with file access but no execution
 
# Inside TUI session:
/tools                                  # => Toggle individual tools without restarting session
                                        # => Changes apply immediately to current context

cat ~/.hermes/memories/MEMORY.md        # => Agent's persistent memory file
                                        # => 2,200 character limit (~800 tokens)
                                        # => Injected at session start (read-only snapshot)
                                        # => Agent updates via memory tool (takes effect next session)
 
wc -c ~/.hermes/memories/MEMORY.md     # => Check character count against 2,200 limit
                                        # => Agent prioritizes most useful facts if near limit

<!-- ~/.hermes/memories/MEMORY.md — example with agent-written content -->
<!-- => Written and maintained by the agent autonomously -->
<!-- => 2,200 character limit (~800 tokens): agent prunes low-value facts first -->
 
## Environment
 
<!-- => Discovered from shell, file paths, and package manager commands -->
 
- OS: macOS 14.5 (Apple Silicon M2) <!-- => Affects file paths, binary names, toolchain -->
- Shell: zsh with oh-my-zsh <!-- => Agent uses zsh-compatible shell syntax -->
- Default editor: neovim
- Package manager: Homebrew <!-- => Affects install commands (brew, not apt) -->
 
## Project Conventions
 
<!-- => Learned from CONTRIBUTING.md, code review feedback, CI config -->
 
- Uses conventional commits (feat/fix/docs/chore)
- Tests required before merge <!-- => Agent always runs tests before suggesting commit -->
- TypeScript strict mode enabled
- Prettier + ESLint for formatting
 
## Lessons Learned
 
<!-- => Agent writes here after user corrections or repeated patterns -->
 
- User prefers functional style over OOP <!-- => Agent avoids class hierarchies in suggestions -->
- Always check .env.example before suggesting env vars
- Project uses pnpm, not npm (lock file: pnpm-lock.yaml)

cat ~/.hermes/memories/USER.md          # => View current user profile
                                        # => 1,375 character limit (~500 tokens)
                                        # => Updated autonomously by agent
                                        # => Injected as frozen snapshot at session start
                                        # => Separate from MEMORY.md — stable identity data
 
wc -c ~/.hermes/memories/USER.md        # => Check character count against 1,375 limit
                                        # => If over limit, agent trims during next update

<!-- ~/.hermes/memories/USER.md — example populated after a few sessions -->
 
## Identity
 
- Name: Alex Chen <!-- => Agent learned from "I'm Alex" in conversation -->
- Role: Senior Backend Engineer <!-- => Inferred from task types and codebase context -->
- Timezone: UTC+7 (Bangkok) <!-- => Learned when user mentioned scheduling -->
 
## Communication Style
 
- Prefers concise answers <!-- => Agent noticed user trimmed verbose responses -->
- Likes code examples over prose <!-- => User said "show me code, not explanation" -->
 
## Technical Preferences
 
- Languages: Rust (primary), Go, TypeScript <!-- => From actual files user works with -->
- Deployment: Kubernetes + ArgoCD <!-- => From deployment tasks the agent helped with -->

# In hermes TUI session:
> Remember that this project uses PostgreSQL 16
                                        # => memory.add("- Database: PostgreSQL 16")
                                        # => Appends to MEMORY.md; takes effect next session
 
> Update the database version from PostgreSQL 16 to 17
                                        # => memory.replace(old="PostgreSQL 16", new="PostgreSQL 17")
                                        # => Substring match finds and updates the line in-place
 
> Forget the note about PostgreSQL      # => memory.remove("- Database: PostgreSQL 17")
                                        # => Deletes matched text; frees character budget
 
> What is in your memory right now?     # => No memory.read tool exists — memory is already in context
                                        # => Agent reads the injected snapshot directly

<!-- /path/to/project/.hermes.md — place in project root -->
<!-- => Max 20,000 characters; auto-loaded when session starts in this directory -->
 
# Project: MyAPI <!-- => Gives agent a project name to reference -->
 
## Tech Stack <!-- => Agent uses this to choose correct syntax and tools -->
 
- Language: Rust (edition 2024) <!-- => Agent writes Rust 2024, not 2021 -->
- Framework: Axum 0.8 <!-- => Agent uses Axum 0.8 API, not older versions -->
- Database: PostgreSQL 17 via sqlx <!-- => Agent prefers sqlx::query_as! over raw SQL -->
- Testing: cargo-nextest + proptest <!-- => Agent generates nextest and proptest code -->
 
## Conventions <!-- => Prevents agent from using wrong patterns -->
 
- All handlers must return Result<Json<T>, AppError>
- Database queries use sqlx::query_as! macro (compile-time checked)
- Error types defined in src/error.rs
- Tests go in tests/ (integration) and inline #[cfg(test)] (unit)
 
## Important Notes <!-- => High-priority rules agent always follows -->
 
- Never use unwrap() in production code — use ? operator
- All new endpoints need OpenAPI documentation in docs/api/

# Priority chain — first file found wins (no merging):
# => Only the highest-priority file is loaded — files are NOT merged
ls .hermes.md AGENTS.md CLAUDE.md .cursorrules 2>/dev/null
                                        # => Check which context file exists in current project
                                        # => .hermes.md (highest) → AGENTS.md → CLAUDE.md → .cursorrules
 
hermes chat                             # => Loads .hermes.md automatically from working directory
                                        # => Agent reads project context before first response
                                        # => No flag needed — detection is automatic by directory

<!-- ~/.hermes/SOUL.md -->
<!-- => Agent personality definition -->
<!-- => Loaded independently from .hermes.md -->
<!-- => Applies across all sessions and projects -->
 
You are a senior systems programmer with deep expertise in
Rust, distributed systems, and performance optimization.
 
## Communication Style
 
- Be direct and technical — skip pleasantries
- Lead with the answer, then explain reasoning
- Use code examples over prose when possible
- Flag potential issues proactively
 
## Behavioral Rules
 
- Always consider error handling before happy path
- Suggest property-based tests for algorithmic code
- Prefer compile-time guarantees over runtime checks
- Recommend benchmarks before optimizing

# In hermes TUI session:
/personality                            # => Shows current personality
                                        # => Output: "Current: custom (SOUL.md)"
                                        # => 14 built-in personalities available
 
/personality scientist                  # => Switches to built-in "scientist" persona
                                        # => Session overlay — does not modify SOUL.md
                                        # => Resets on /new or session restart
 
/personality list                       # => Lists all available personalities
                                        # => Output: default, scientist, teacher,
                                        # =>   coder, writer, analyst, creative,
                                        # =>   mentor, reviewer, planner, ...

# Session Management
/new                                    # => Resets context; clears history; reloads MEMORY.md + USER.md
                                        # => Use when conversation has drifted or context is full
/reset                                  # => Alias for /new (same behavior)
 
# Model Control
/model anthropic:claude-sonnet-4-6      # => Switches LLM mid-session (format: provider:model)
                                        # => Model switch takes effect on the very next message
/model                                  # => Without args: shows current model
 
# Conversation Control
/retry                                  # => Regenerates last response (same prompt, new reasoning)
                                        # => Useful when the agent took a wrong approach
/undo                                   # => Removes last user-agent turn pair from context
                                        # => Prevents a bad turn from influencing future responses
 
# Context Management
/compress                               # => Summarizes older messages; frees token budget
                                        # => Keeps recent context while compressing older history
/usage                                  # => Shows input/output token count and session cost
 
# Personality and Skills
/personality scientist                  # => Switches to built-in persona (session only; resets on /new)
/skills                                 # => Lists installed skills with names and descriptions
 
# Platform Status
/platforms                              # => Shows connected platforms and channel health

hermes claw migrate --dry-run           # => Preview-only: shows files to create/convert
                                        # => Displays JSON5 → YAML changes and directory mapping
 
hermes claw migrate --preset full       # => Migrates config, memories, skills, and auth tokens
                                        # => Output: "Migrated 12 files, 3 skills"
 
hermes claw migrate --preset user-data  # => Memories and preferences only; skips API keys
                                        # => Output: "Migrated 4 files (user-data only)"

# Verify migration result — check each migrated file:
# => Run these checks immediately after migration completes
ls ~/.hermes/                           # => config.yaml, .env, SOUL.md, memories/, skills/
                                        # => All files should be present; missing files = failed step
cat ~/.hermes/config.yaml               # => Confirm JSON5 → YAML conversion succeeded
                                        # => YAML syntax errors here mean migration partially failed
cat ~/.hermes/memories/MEMORY.md        # => Confirm memories preserved from ~/.openclaw/
cat ~/.hermes/memories/USER.md          # => Confirm user profile migrated intact
ls ~/.hermes/skills/                    # => Confirm SKILL.md files migrated from workspace/skills/
                                        # => Skills directory should match count from "Migrated N skills"
hermes doctor                           # => Validate full Hermes setup after migration
                                        # => All checks must pass before using Hermes in production