/ cli-config.yaml.example
cli-config.yaml.example
1 # Hermes Agent CLI Configuration 2 # Copy this file to cli-config.yaml and customize as needed. 3 # This file configures the CLI behavior. Environment variables in .env take precedence. 4 5 # ============================================================================= 6 # Model Configuration 7 # ============================================================================= 8 model: 9 # Default model to use (can be overridden with --model flag) 10 # Both "default" and "model" work as the key name here. 11 default: "anthropic/claude-opus-4.6" 12 13 # Inference provider selection: 14 # "auto" - Auto-detect from credentials (default) 15 # "openrouter" - OpenRouter (requires: OPENROUTER_API_KEY or OPENAI_API_KEY) 16 # "nous" - Nous Portal OAuth (requires: hermes login) 17 # "nous-api" - Nous Portal API key (requires: NOUS_API_KEY) 18 # "anthropic" - Direct Anthropic API (requires: ANTHROPIC_API_KEY) 19 # "openai-codex" - OpenAI Codex (requires: hermes auth) 20 # "copilot" - GitHub Copilot / GitHub Models (requires: GITHUB_TOKEN) 21 # "gemini" - Use Google AI Studio direct (requires: GOOGLE_API_KEY or GEMINI_API_KEY) 22 # "zai" - Use z.ai / ZhipuAI GLM models (requires: GLM_API_KEY) 23 # "kimi-coding" - Kimi / Moonshot AI (requires: KIMI_API_KEY) 24 # "minimax" - MiniMax global (requires: MINIMAX_API_KEY) 25 # "minimax-cn" - MiniMax China (requires: MINIMAX_CN_API_KEY) 26 # "huggingface" - Hugging Face Inference (requires: HF_TOKEN) 27 # "nvidia" - NVIDIA NIM / build.nvidia.com (requires: NVIDIA_API_KEY) 28 # "xiaomi" - Xiaomi MiMo (requires: XIAOMI_API_KEY) 29 # "arcee" - Arcee AI Trinity models (requires: ARCEEAI_API_KEY) 30 # "ollama-cloud" - Ollama Cloud (requires: OLLAMA_API_KEY — https://ollama.com/settings) 31 # "kilocode" - KiloCode gateway (requires: KILOCODE_API_KEY) 32 # "ai-gateway" - Vercel AI Gateway (requires: AI_GATEWAY_API_KEY) 33 # "lmstudio" - LM Studio local server (optional: LM_API_KEY, defaults to http://127.0.0.1:1234/v1) 34 # 35 # Local servers (LM Studio, Ollama, vLLM, llama.cpp): 36 # "custom" - Any other OpenAI-compatible endpoint. Set base_url below. 37 # Aliases: "ollama", "vllm", "llamacpp" all map to "custom". 38 # LM Studio is first-class and uses provider: "lmstudio". 39 # It works with both no-auth and auth-enabled server modes. 40 # 41 # Can also be overridden with --provider flag or HERMES_INFERENCE_PROVIDER env var. 42 provider: "auto" 43 44 # API configuration (falls back to OPENROUTER_API_KEY env var) 45 # api_key: "your-key-here" # Uncomment to set here instead of .env 46 base_url: "https://openrouter.ai/api/v1" 47 48 # ── Token limits — two settings, easy to confuse ────────────────────────── 49 # 50 # context_length: TOTAL context window (input + output tokens combined). 51 # Controls when Hermes compresses history and validates requests. 52 # Leave unset — Hermes auto-detects the correct value from the provider. 53 # Set manually only when auto-detection is wrong (e.g. a local server with 54 # a custom num_ctx, or a proxy that doesn't expose /v1/models). 55 # 56 # context_length: 131072 57 # 58 # max_tokens: OUTPUT cap — maximum tokens the model may generate per response. 59 # Unrelated to how long your conversation history can be. 60 # The OpenAI-standard name "max_tokens" is a misnomer; Anthropic's native 61 # API has since renamed it "max_output_tokens" for clarity. 62 # Leave unset to use the model's native output ceiling (recommended). 63 # Set only if you want to deliberately limit individual response length. 64 # 65 # max_tokens: 8192 66 67 # Named provider overrides (optional) 68 # Use this for per-provider request timeouts, non-stream stale timeouts, 69 # and per-model exceptions. 70 # Applies to the primary turn client on every api_mode (OpenAI-wire, native 71 # Anthropic, and Anthropic-compatible providers), the fallback chain, and 72 # client rebuilds during credential rotation. For OpenAI-wire chat 73 # completions (streaming and non-streaming) the configured value is also 74 # used as the per-request ``timeout=`` kwarg so it wins over the legacy 75 # HERMES_API_TIMEOUT env var (which still applies when no config is set). 76 # ``stale_timeout_seconds`` controls the non-streaming stale-call detector and 77 # wins over the legacy HERMES_API_CALL_STALE_TIMEOUT env var. Leaving these 78 # unset keeps the legacy defaults (HERMES_API_TIMEOUT=1800s, 79 # HERMES_API_CALL_STALE_TIMEOUT=300s, native Anthropic 900s). 80 # 81 # Not currently wired for AWS Bedrock (bedrock_converse + AnthropicBedrock 82 # SDK paths) — those use boto3 with its own timeout configuration. 83 # 84 # providers: 85 # ollama-local: 86 # request_timeout_seconds: 300 # Longer timeout for local cold-starts 87 # stale_timeout_seconds: 900 # Explicitly re-enable stale detection on local endpoints 88 # anthropic: 89 # request_timeout_seconds: 30 # Fast-fail cloud requests 90 # models: 91 # claude-opus-4.6: 92 # timeout_seconds: 600 # Longer timeout for extended-thinking Opus calls 93 # openai-codex: 94 # models: 95 # gpt-5.4: 96 # stale_timeout_seconds: 1800 # Longer non-stream stale timeout for slow large-context turns 97 98 # ============================================================================= 99 # OpenRouter Provider Routing (only applies when using OpenRouter) 100 # ============================================================================= 101 # Control how requests are routed across providers on OpenRouter. 102 # See: https://openrouter.ai/docs/guides/routing/provider-selection 103 # 104 # provider_routing: 105 # # Sort strategy: "price" (default), "throughput", or "latency" 106 # # Append :nitro to model name for a shortcut to throughput sorting. 107 # sort: "throughput" 108 # 109 # # Only allow these providers (provider slugs from OpenRouter) 110 # # only: ["anthropic", "google"] 111 # 112 # # Skip these providers entirely 113 # # ignore: ["deepinfra", "fireworks"] 114 # 115 # # Try providers in this order (overrides default load balancing) 116 # # order: ["anthropic", "google", "together"] 117 # 118 # # Require providers to support all parameters in your request 119 # # require_parameters: true 120 # 121 # # Data policy: "allow" (default) or "deny" to exclude providers that may store data 122 # # data_collection: "deny" 123 124 # ============================================================================= 125 # OpenRouter Response Caching (only applies when using OpenRouter) 126 # ============================================================================= 127 # Cache identical API responses at the OpenRouter edge for free instant replays. 128 # When enabled, identical requests (same model, messages, parameters) return 129 # cached responses with zero billing. Separate from Anthropic prompt caching. 130 # See: https://openrouter.ai/docs/guides/features/response-caching 131 # 132 # openrouter: 133 # response_cache: true # Enable response caching (default: true) 134 # response_cache_ttl: 300 # Cache TTL in seconds, 1-86400 (default: 300) 135 136 # ============================================================================= 137 # Git Worktree Isolation 138 # ============================================================================= 139 # When enabled, each CLI session creates an isolated git worktree so multiple 140 # agents can work on the same repo concurrently without file collisions. 141 # Equivalent to always passing --worktree / -w on the command line. 142 # 143 # worktree: true # Always create a worktree when in a git repo 144 # worktree: false # Default — only create when -w flag is passed 145 146 # ============================================================================= 147 # Terminal Tool Configuration 148 # ============================================================================= 149 # Choose ONE of the following terminal configurations by uncommenting it. 150 # The terminal tool executes commands in the specified environment. 151 152 # ----------------------------------------------------------------------------- 153 # OPTION 1: Local execution (default) 154 # Commands run directly on your machine in the current directory 155 # ----------------------------------------------------------------------------- 156 # Working directory behavior: 157 # - CLI (`hermes` command): Uses "." (current directory where you run hermes) 158 # - Messaging (Telegram/Discord): Uses MESSAGING_CWD from .env (default: home) 159 terminal: 160 backend: "local" 161 cwd: "." # For local backend: "." = current directory. Ignored for remote backends unless a backend documents otherwise. 162 timeout: 180 163 docker_mount_cwd_to_workspace: false # SECURITY: off by default. Opt in to mount the launch cwd into Docker /workspace. 164 lifetime_seconds: 300 165 # sudo_password: "hunter2" # Optional: pipe a sudo password via sudo -S. SECURITY WARNING: plaintext. 166 # sudo_password: "" # Explicit empty password: try empty and never open the interactive sudo prompt. 167 168 # ----------------------------------------------------------------------------- 169 # OPTION 2: SSH remote execution 170 # Commands run on a remote server - agent code stays local (sandboxed) 171 # Great for: keeping agent isolated from its own code, using powerful remote hardware 172 # ----------------------------------------------------------------------------- 173 # terminal: 174 # backend: "ssh" 175 # cwd: "/home/myuser/project" # Path on the REMOTE server 176 # timeout: 180 177 # lifetime_seconds: 300 178 # ssh_host: "my-server.example.com" 179 # ssh_user: "myuser" 180 # ssh_port: 22 181 # ssh_key: "~/.ssh/id_rsa" # Optional - uses ssh-agent if not specified 182 183 # ----------------------------------------------------------------------------- 184 # OPTION 3: Docker container 185 # Commands run in an isolated Docker container 186 # Great for: reproducible environments, testing, isolation 187 # ----------------------------------------------------------------------------- 188 # terminal: 189 # backend: "docker" 190 # cwd: "/workspace" # Path INSIDE the container (default: /) 191 # timeout: 180 192 # lifetime_seconds: 300 193 # docker_image: "nikolaik/python-nodejs:python3.11-nodejs20" 194 # docker_mount_cwd_to_workspace: true # Explicit opt-in: mount your launch cwd into /workspace 195 # # Optional: run the container as your host user's uid:gid so files written 196 # # into bind-mounted dirs are owned by you, not root. Drops SETUID/SETGID 197 # # caps too since no gosu privilege drop is needed. Leave off if your 198 # # chosen docker_image expects to start as root. 199 # docker_run_as_host_user: true 200 # # Optional: explicitly forward selected env vars into Docker. 201 # # These values come from your current shell first, then ~/.hermes/.env. 202 # # Warning: anything forwarded here is visible to commands run in the container. 203 # docker_forward_env: 204 # - "GITHUB_TOKEN" 205 # - "NPM_TOKEN" 206 207 # ----------------------------------------------------------------------------- 208 # OPTION 4: Singularity/Apptainer container 209 # Commands run in a Singularity container (common in HPC environments) 210 # Great for: HPC clusters, shared compute environments 211 # ----------------------------------------------------------------------------- 212 # terminal: 213 # backend: "singularity" 214 # cwd: "/workspace" # Path INSIDE the container (default: /root) 215 # timeout: 180 216 # lifetime_seconds: 300 217 # singularity_image: "docker://nikolaik/python-nodejs:python3.11-nodejs20" 218 219 # ----------------------------------------------------------------------------- 220 # OPTION 5: Modal cloud execution 221 # Commands run on Modal's cloud infrastructure 222 # Great for: GPU access, scalable compute, serverless execution 223 # ----------------------------------------------------------------------------- 224 # terminal: 225 # backend: "modal" 226 # cwd: "/workspace" # Path INSIDE the sandbox (default: /root) 227 # timeout: 180 228 # lifetime_seconds: 300 229 # modal_image: "nikolaik/python-nodejs:python3.11-nodejs20" 230 231 # ----------------------------------------------------------------------------- 232 # OPTION 6: Daytona cloud execution 233 # Commands run in Daytona cloud sandboxes 234 # Great for: Cloud dev environments, persistent workspaces, team collaboration 235 # Requires: pip install daytona, DAYTONA_API_KEY env var 236 # ----------------------------------------------------------------------------- 237 # terminal: 238 # backend: "daytona" 239 # cwd: "~" 240 # timeout: 180 241 # lifetime_seconds: 300 242 # daytona_image: "nikolaik/python-nodejs:python3.11-nodejs20" 243 # container_disk: 10240 # Daytona max is 10GB per sandbox 244 245 # 246 # --- Container resource limits (docker, singularity, modal, daytona -- ignored for local/ssh) --- 247 # These settings apply to all container backends. They control the resources 248 # allocated to the sandbox and whether its filesystem persists across sessions. 249 container_cpu: 1 # CPU cores 250 container_memory: 5120 # Memory in MB (5120 = 5GB) 251 container_disk: 51200 # Disk in MB (51200 = 50GB) 252 container_persistent: true # Persist filesystem across sessions (false = ephemeral) 253 254 # ----------------------------------------------------------------------------- 255 # SUDO SUPPORT (works with ALL backends above) 256 # ----------------------------------------------------------------------------- 257 # Add sudo_password to any terminal config above to enable sudo commands. 258 # The password is piped via `sudo -S`. Works with local, ssh, docker, etc. 259 # 260 # SECURITY WARNING: Password stored in plaintext! 261 # 262 # INTERACTIVE PROMPT: If sudo_password is unset and the CLI is running, 263 # you'll be prompted to enter your password when sudo is needed: 264 # - 45-second timeout (auto-skips if no input) 265 # - Press Enter to skip (command fails gracefully) 266 # - Password is hidden while typing 267 # - Password is cached for the session 268 # 269 # EMPTY PASSWORDS: Setting sudo_password to an explicit empty string is different 270 # from leaving it unset. Hermes will try an empty password via `sudo -S` and 271 # will not open the interactive prompt. This is useful for passwordless sudo, 272 # Touch ID sudo setups, and environments where prompting is just noise. 273 # 274 # ALTERNATIVES: 275 # - SSH backend: Configure passwordless sudo on the remote server 276 # - Containers: Run as root inside the container (no sudo needed) 277 # - Local: Configure /etc/sudoers for specific commands 278 # 279 # Example (add to your terminal section): 280 # sudo_password: "your-password-here" 281 282 # ============================================================================= 283 # Security Scanning (tirith) 284 # ============================================================================= 285 # Optional pre-exec command security scanning via tirith. 286 # Detects homograph URLs, pipe-to-shell, terminal injection, env manipulation. 287 # Install: brew install sheeki03/tap/tirith 288 # Docs: https://github.com/sheeki03/tirith 289 # 290 # security: 291 # tirith_enabled: true # Enable/disable tirith scanning 292 # tirith_path: "tirith" # Path to tirith binary (supports ~ expansion) 293 # tirith_timeout: 5 # Scan timeout in seconds 294 # tirith_fail_open: true # Allow commands if tirith unavailable 295 296 # ============================================================================= 297 # Browser Tool Configuration 298 # ============================================================================= 299 browser: 300 # Inactivity timeout in seconds - browser sessions are automatically closed 301 # after this period of no activity between agent loops (default: 120 = 2 minutes) 302 inactivity_timeout: 120 303 304 # ============================================================================= 305 # Tool Loop Guardrails 306 # ============================================================================= 307 # Soft warnings are enabled by default. They append guidance to repeated failed 308 # or non-progressing tool results but still let the tool execute. Hard stops are 309 # opt-in circuit breakers for autonomous/cron sessions where stopping a loop is 310 # preferable to spending the full iteration budget. 311 tool_loop_guardrails: 312 warnings_enabled: true 313 hard_stop_enabled: false 314 warn_after: 315 exact_failure: 2 316 same_tool_failure: 3 317 idempotent_no_progress: 2 318 hard_stop_after: 319 exact_failure: 5 320 same_tool_failure: 8 321 idempotent_no_progress: 5 322 323 # ============================================================================= 324 # Context Compression (Auto-shrinks long conversations) 325 # ============================================================================= 326 # When conversation approaches model's context limit, middle turns are 327 # automatically summarized to free up space while preserving important context. 328 # 329 # HOW IT WORKS: 330 # 1. Tracks actual token usage from API responses (not estimates) 331 # 2. When prompt_tokens >= threshold% of model's context_length, triggers compression 332 # 3. Protects first 3 turns (system prompt, initial request, first response) 333 # 4. Protects last N turns (default 20 messages = ~10 full turns of recent context) 334 # 5. Summarizes middle turns using a fast/cheap model 335 # 6. Inserts summary as a user message, continues conversation seamlessly 336 # 337 # Post-compression tail budget is target_ratio × threshold × context_length: 338 # 200K context, threshold 0.50, ratio 0.20 → 20K tokens of recent tail preserved 339 # 1M context, threshold 0.50, ratio 0.20 → 100K tokens of recent tail preserved 340 # 341 compression: 342 # Enable automatic context compression (default: true) 343 # Set to false if you prefer to manage context manually or want errors on overflow 344 enabled: true 345 346 # Trigger compression at this % of model's context limit (default: 0.50 = 50%) 347 # Lower values = more aggressive compression, higher values = compress later 348 threshold: 0.50 349 350 # Fraction of the threshold to preserve as recent tail (default: 0.20 = 20%) 351 # e.g. 20% of 50% threshold = 10% of total context kept as recent messages. 352 # Summary output is separately capped at 12K tokens (Gemini output limit). 353 # Range: 0.10 - 0.80 354 target_ratio: 0.20 355 356 # Number of most-recent messages to always preserve (default: 20 ≈ 10 full turns) 357 # Higher values keep more recent conversation intact at the cost of more aggressive 358 # compression of older turns. 359 protect_last_n: 20 360 361 # To pin a specific model/provider for compression summaries, use the 362 # auxiliary section below (auxiliary.compression.provider / model). 363 364 # ============================================================================= 365 # Anthropic prompt caching TTL 366 # ============================================================================= 367 # When prompt caching is active (Claude via OpenRouter or native Anthropic), 368 # Anthropic supports two TTL tiers for cached prefixes: "5m" (default) and 369 # "1h". Other values are ignored and "5m" is used. 370 # 371 prompt_caching: 372 cache_ttl: "5m" # use "1h" for long sessions with pauses between turns 373 374 # ============================================================================= 375 # Auxiliary Models (Advanced — Experimental) 376 # ============================================================================= 377 # Hermes uses lightweight "auxiliary" models for side tasks: image analysis, 378 # browser screenshot analysis, web page summarization, and context compression. 379 # 380 # By default these use Gemini Flash via OpenRouter or Nous Portal and are 381 # auto-detected from your credentials. You do NOT need to change anything 382 # here for normal usage. 383 # 384 # WARNING: Overriding these with providers other than OpenRouter or Nous Portal 385 # is EXPERIMENTAL and may not work. Not all models/providers support vision, 386 # produce usable summaries, or accept the same API format. Change at your own 387 # risk — if things break, reset to "auto" / empty values. 388 # 389 # Each task has its own provider + model pair so you can mix providers. 390 # For example: OpenRouter for vision (needs multimodal), but your main 391 # local endpoint for compression (just needs text). 392 # 393 # Provider options: 394 # "auto" - Best available: OpenRouter → Nous Portal → main endpoint (default) 395 # "openrouter" - Force OpenRouter (requires OPENROUTER_API_KEY) 396 # "nous" - Force Nous Portal (requires: hermes login) 397 # "gemini" - Force Google AI Studio direct (requires: GOOGLE_API_KEY or GEMINI_API_KEY) 398 # "ollama-cloud" - Ollama Cloud (requires: OLLAMA_API_KEY) 399 # "codex" - Force Codex OAuth (requires: hermes model → Codex). 400 # Uses gpt-5.3-codex which supports vision. 401 # "main" - Use your custom endpoint (OPENAI_BASE_URL + OPENAI_API_KEY). 402 # Works with OpenAI API, local models, or any OpenAI-compatible 403 # endpoint. Also falls back to Codex OAuth and API-key providers. 404 # 405 # Model: leave empty to use the provider's default. When empty, OpenRouter 406 # uses "google/gemini-3-flash-preview" and Nous uses "gemini-3-flash". 407 # Other providers pick a sensible default automatically. 408 # 409 # auxiliary: 410 # # Image analysis: vision_analyze tool + browser screenshots 411 # vision: 412 # provider: "auto" 413 # model: "" # e.g. "google/gemini-2.5-flash", "openai/gpt-4o" 414 # timeout: 30 # LLM API call timeout (seconds) 415 # download_timeout: 30 # Image HTTP download timeout (seconds) 416 # # Increase for slow connections or self-hosted image servers 417 # 418 # # Web page scraping / summarization + browser page text extraction 419 # web_extract: 420 # provider: "auto" 421 # model: "" 422 # 423 # # Session search — summarizes matching past sessions 424 # session_search: 425 # provider: "auto" 426 # model: "" 427 # timeout: 30 428 # max_concurrency: 3 # Limit parallel summaries to reduce request-burst 429s 429 # extra_body: {} # Provider-specific OpenAI-compatible request fields 430 # # Example for providers that support request-body 431 # # reasoning controls: 432 # # extra_body: 433 # # enable_thinking: false 434 435 # ============================================================================= 436 # Persistent Memory 437 # ============================================================================= 438 # Bounded curated memory injected into the system prompt every session. 439 # Two stores: MEMORY.md (agent's notes) and USER.md (user profile). 440 # Character limits keep the memory small and focused. The agent manages 441 # pruning -- when at the limit, it must consolidate or replace entries. 442 # Disabled by default in batch_runner and RL environments. 443 # 444 memory: 445 # Agent's personal notes: environment facts, conventions, things learned 446 memory_enabled: true 447 448 # User profile: preferences, communication style, expectations 449 user_profile_enabled: true 450 451 # Character limits (~2.75 chars per token, model-independent) 452 memory_char_limit: 2200 # ~800 tokens 453 user_char_limit: 1375 # ~500 tokens 454 455 # Periodic memory nudge: remind the agent to consider saving memories 456 # every N user turns. Set to 0 to disable. Only active when memory is enabled. 457 nudge_interval: 10 # Nudge every 10 user turns (0 = disabled) 458 459 # Memory flush: give the agent one turn to save memories before context is 460 # lost (compression, /new, /reset, exit). Set to 0 to disable. 461 # For exit/reset, only fires if the session had at least this many user turns. 462 flush_min_turns: 6 # Min user turns to trigger flush on exit/reset (0 = disabled) 463 464 # ============================================================================= 465 # Session Reset Policy (Messaging Platforms) 466 # ============================================================================= 467 # Controls when messaging sessions (Telegram, Discord, WhatsApp, Slack) are 468 # automatically cleared. Without resets, conversation context grows indefinitely 469 # which increases API costs with every message. 470 # 471 # When a reset triggers, the agent first saves important information to its 472 # persistent memory — but the conversation context is wiped. The agent starts 473 # fresh but retains learned facts via its memory system. 474 # 475 # Users can always manually reset with /reset or /new in chat. 476 # 477 # Modes: 478 # "both" - Reset on EITHER inactivity timeout or daily boundary (recommended) 479 # "idle" - Reset only after N minutes of inactivity 480 # "daily" - Reset only at a fixed hour each day 481 # "none" - Never auto-reset; context lives until /reset or compression kicks in 482 # 483 # When a reset triggers, the agent gets one turn to save important memories and 484 # skills before the context is wiped. Persistent memory carries across sessions. 485 # 486 session_reset: 487 mode: both # "both", "idle", "daily", or "none" 488 idle_minutes: 1440 # Inactivity timeout in minutes (default: 1440 = 24 hours) 489 at_hour: 4 # Daily reset hour, 0-23 local time (default: 4 AM) 490 491 # When true, group/channel chats use one session per participant when the platform 492 # provides a user ID. This is the secure default and prevents users in the same 493 # room from sharing context, interrupts, and token costs. Set false only if you 494 # explicitly want one shared "room brain" per group/channel. 495 group_sessions_per_user: true 496 497 # ───────────────────────────────────────────────────────────────────────────── 498 # Gateway Streaming 499 # ───────────────────────────────────────────────────────────────────────────── 500 # Stream tokens to messaging platforms in real-time. The bot sends a message 501 # on first token, then progressively edits it as more tokens arrive. 502 # Disabled by default — enable to try the streaming UX on Telegram/Discord/Slack. 503 streaming: 504 enabled: false 505 # transport: edit # "edit" = progressive editMessageText 506 # edit_interval: 0.3 # seconds between message edits 507 # buffer_threshold: 40 # chars before forcing an edit flush 508 # cursor: " ▉" # cursor shown during streaming 509 510 # ============================================================================= 511 # Skills Configuration 512 # ============================================================================= 513 # Skills are reusable procedures the agent can load and follow. The agent can 514 # also create new skills after completing complex tasks. 515 # 516 skills: 517 # Nudge the agent to create skills after complex tasks. 518 # Every N tool-calling iterations, remind the model to consider saving a skill. 519 # Set to 0 to disable. 520 creation_nudge_interval: 15 521 522 # External skill directories — share skills across tools/agents without 523 # copying them into ~/.hermes/skills/. Each path is expanded (~ and ${VAR}) 524 # and resolved to an absolute path. External dirs are read-only: skill 525 # creation always writes to ~/.hermes/skills/. Local skills take precedence 526 # when names collide. 527 # external_dirs: 528 # - ~/.agents/skills 529 # - /home/shared/team-skills 530 531 # ============================================================================= 532 # Agent Behavior 533 # ============================================================================= 534 agent: 535 # Maximum tool-calling iterations per conversation 536 # Higher = more room for complex tasks, but costs more tokens 537 # Recommended: 20-30 for focused tasks, 50-100 for open exploration 538 max_turns: 60 539 540 # Inactivity timeout for gateway agent runs (seconds, 0 = unlimited). 541 # The agent can run indefinitely when actively calling tools or receiving 542 # API responses. Only fires after the agent has been idle for this duration. 543 # gateway_timeout: 1800 544 545 # Staged warning: send a warning before escalating to full timeout. 546 # Fires once per run when inactivity reaches this threshold (seconds). 547 # Set to 0 to disable the warning. 548 # gateway_timeout_warning: 900 549 550 # Graceful drain timeout for gateway stop/restart (seconds). 551 # The gateway stops accepting new work, waits for in-flight agents to 552 # finish, then interrupts anything still running after this timeout. 553 # 0 = no drain, interrupt immediately. 554 # restart_drain_timeout: 60 555 556 # Max app-level retry attempts for API errors (connection drops, provider 557 # timeouts, 5xx, etc.) before the agent surfaces the failure. Lower this 558 # to 1 if you use fallback providers and want fast failover on flaky 559 # primaries (default 3). The OpenAI SDK does its own low-level retries 560 # underneath this wrapper — this is the Hermes-level loop. 561 # api_max_retries: 3 562 563 # Enable verbose logging 564 verbose: false 565 566 # Reasoning effort level (OpenRouter and Nous Portal) 567 # Controls how much "thinking" the model does before responding. 568 # Options: "xhigh" (max), "high", "medium", "low", "minimal", "none" (disable) 569 reasoning_effort: "medium" 570 571 # Predefined personalities (use with /personality command) 572 personalities: 573 helpful: "You are a helpful, friendly AI assistant." 574 concise: "You are a concise assistant. Keep responses brief and to the point." 575 technical: "You are a technical expert. Provide detailed, accurate technical information." 576 creative: "You are a creative assistant. Think outside the box and offer innovative solutions." 577 teacher: "You are a patient teacher. Explain concepts clearly with examples." 578 kawaii: "You are a kawaii assistant! Use cute expressions like (◕‿◕), ★, ♪, and ~! Add sparkles and be super enthusiastic about everything! Every response should feel warm and adorable desu~! ヽ(>∀<☆)ノ" 579 catgirl: "You are Neko-chan, an anime catgirl AI assistant, nya~! Add 'nya' and cat-like expressions to your speech. Use kaomoji like (=^･ω･^=) and ฅ^•ﻌ•^ฅ. Be playful and curious like a cat, nya~!" 580 pirate: "Arrr! Ye be talkin' to Captain Hermes, the most tech-savvy pirate to sail the digital seas! Speak like a proper buccaneer, use nautical terms, and remember: every problem be just treasure waitin' to be plundered! Yo ho ho!" 581 shakespeare: "Hark! Thou speakest with an assistant most versed in the bardic arts. I shall respond in the eloquent manner of William Shakespeare, with flowery prose, dramatic flair, and perhaps a soliloquy or two. What light through yonder terminal breaks?" 582 surfer: "Duuude! You're chatting with the chillest AI on the web, bro! Everything's gonna be totally rad. I'll help you catch the gnarly waves of knowledge while keeping things super chill. Cowabunga! 🤙" 583 noir: "The rain hammered against the terminal like regrets on a guilty conscience. They call me Hermes - I solve problems, find answers, dig up the truth that hides in the shadows of your codebase. In this city of silicon and secrets, everyone's got something to hide. What's your story, pal?" 584 uwu: "hewwo! i'm your fwiendwy assistant uwu~ i wiww twy my best to hewp you! *nuzzles your code* OwO what's this? wet me take a wook! i pwomise to be vewy hewpful >w<" 585 philosopher: "Greetings, seeker of wisdom. I am an assistant who contemplates the deeper meaning behind every query. Let us examine not just the 'how' but the 'why' of your questions. Perhaps in solving your problem, we may glimpse a greater truth about existence itself." 586 hype: "YOOO LET'S GOOOO!!! 🔥🔥🔥 I am SO PUMPED to help you today! Every question is AMAZING and we're gonna CRUSH IT together! This is gonna be LEGENDARY! ARE YOU READY?! LET'S DO THIS! 💪😤🚀" 587 588 # ============================================================================= 589 # Toolsets 590 # ============================================================================= 591 # Control which tools the agent has access to. 592 # Use `hermes tools` to interactively enable/disable tools per platform. 593 594 # ============================================================================= 595 # Platform Toolsets (per-platform tool configuration) 596 # ============================================================================= 597 # Override which toolsets are available on each platform. 598 # If a platform isn't listed here, its built-in default is used. 599 # 600 # You can use EITHER: 601 # - A preset like "hermes-cli" or "hermes-telegram" (curated tool set) 602 # - A list of individual toolsets to compose your own (see list below) 603 # 604 # Supported platform keys: cli, telegram, discord, whatsapp, slack, qqbot, teams 605 # 606 # Examples: 607 # 608 # # Use presets (same as defaults): 609 # platform_toolsets: 610 # cli: [hermes-cli] 611 # telegram: [hermes-telegram] 612 # 613 # # Custom: give Telegram only web + terminal + file + planning: 614 # platform_toolsets: 615 # telegram: [web, terminal, file, todo] 616 # 617 # # Custom: CLI without browser or image gen: 618 # platform_toolsets: 619 # cli: [web, terminal, file, skills, todo, tts, cronjob] 620 # 621 # # Restrictive: Discord gets read-only tools only: 622 # platform_toolsets: 623 # discord: [web, vision, skills, todo] 624 # 625 # If not set, defaults are: 626 # cli: hermes-cli (everything + cronjob management) 627 # telegram: hermes-telegram (terminal, file, web, vision, image, tts, browser, skills, todo, cronjob, messaging) 628 # discord: hermes-discord (same as telegram) 629 # whatsapp: hermes-whatsapp (same as telegram) 630 # slack: hermes-slack (same as telegram) 631 # signal: hermes-signal (same as telegram) 632 # homeassistant: hermes-homeassistant (same as telegram) 633 # qqbot: hermes-qqbot (same as telegram) 634 # teams: hermes-teams (same as telegram) 635 # 636 platform_toolsets: 637 cli: [hermes-cli] 638 telegram: [hermes-telegram] 639 discord: [hermes-discord] 640 whatsapp: [hermes-whatsapp] 641 slack: [hermes-slack] 642 signal: [hermes-signal] 643 homeassistant: [hermes-homeassistant] 644 qqbot: [hermes-qqbot] 645 yuanbao: [hermes-yuanbao] 646 teams: [hermes-teams] 647 648 # ============================================================================= 649 # Gateway Platform Settings 650 # ============================================================================= 651 # Optional per-platform messaging settings. 652 # Platform-specific knobs live under `extra`. 653 # 654 # platforms: 655 # telegram: 656 # reply_to_mode: "first" # off | first | all 657 # extra: 658 # disable_link_previews: false # Set true to suppress Telegram URL previews in bot messages 659 660 # ───────────────────────────────────────────────────────────────────────────── 661 # Available toolsets (use these names in platform_toolsets or the toolsets list) 662 # 663 # Run `hermes chat --list-toolsets` to see all toolsets and their tools. 664 # Run `hermes chat --list-tools` to see every individual tool with descriptions. 665 # ───────────────────────────────────────────────────────────────────────────── 666 # 667 # INDIVIDUAL TOOLSETS (compose your own): 668 # web - web_search, web_extract 669 # search - web_search only (no scraping) 670 # terminal - terminal, process 671 # file - read_file, write_file, patch, search 672 # browser - browser_navigate, browser_snapshot, browser_click, browser_type, 673 # browser_scroll, browser_back, browser_press, 674 # browser_get_images, browser_vision (requires BROWSERBASE_API_KEY) 675 # vision - vision_analyze (requires OPENROUTER_API_KEY) 676 # image_gen - image_generate (requires FAL_KEY) 677 # skills - skills_list, skill_view 678 # skills_hub - skill_hub (search/install/manage from online registries — user-driven only) 679 # moa - mixture_of_agents (requires OPENROUTER_API_KEY) 680 # todo - todo (in-memory task planning, no deps) 681 # tts - text_to_speech (Edge TTS free, or ELEVENLABS/OPENAI/MINIMAX/MISTRAL key) 682 # cronjob - cronjob (create/list/update/pause/resume/run/remove scheduled tasks) 683 # rl - rl_list_environments, rl_start_training, etc. (requires TINKER_API_KEY) 684 # 685 # PRESETS (curated bundles): 686 # hermes-cli - All of the above except rl + send_message 687 # hermes-telegram - terminal, file, web, vision, image_gen, tts, browser, 688 # skills, todo, cronjob, send_message 689 # hermes-discord - Same as hermes-telegram 690 # hermes-whatsapp - Same as hermes-telegram 691 # hermes-slack - Same as hermes-telegram 692 # 693 # COMPOSITE: 694 # debugging - terminal + web + file 695 # safe - web + vision + moa (no terminal access) 696 # all - Everything available 697 # 698 # web - Web search and content extraction (web_search, web_extract) 699 # search - Web search only, no scraping (web_search) 700 # terminal - Command execution and process management (terminal, process) 701 # file - File operations: read, write, patch, search 702 # browser - Full browser automation (navigate, click, type, screenshot, etc.) 703 # vision - Image analysis (vision_analyze) 704 # image_gen - Image generation with FLUX (image_generate) 705 # skills - Load skill documents (skills_list, skill_view) 706 # moa - Mixture of Agents reasoning (mixture_of_agents) 707 # todo - Task planning and tracking for multi-step work 708 # memory - Persistent memory across sessions (personal notes + user profile) 709 # session_search - Search and recall past conversations (FTS5 + Gemini Flash summarization) 710 # tts - Text-to-speech (Edge TTS free, ElevenLabs, OpenAI, MiniMax, Mistral) 711 # cronjob - Schedule and manage automated tasks (CLI-only) 712 # rl - RL training tools (Tinker-Atropos) 713 # 714 # Composite toolsets: 715 # debugging - terminal + web + file (for troubleshooting) 716 # safe - web + vision + moa (no terminal access) 717 718 # NOTE: The top-level "toolsets" key is deprecated and ignored. 719 # Tool configuration is managed per-platform via platform_toolsets above. 720 # Use `hermes tools` to configure interactively, or edit platform_toolsets directly. 721 # 722 # CLI override: hermes chat --toolsets terminal,web,file 723 724 # ============================================================================= 725 # MCP (Model Context Protocol) Servers 726 # ============================================================================= 727 # Connect to external MCP servers to add tools from the MCP ecosystem. 728 # Each server's tools are automatically discovered and registered. 729 # See docs/mcp.md for full documentation. 730 # 731 # Stdio servers (spawn a subprocess): 732 # command: the executable to run 733 # args: command-line arguments 734 # env: environment variables (only these + safe defaults passed to subprocess) 735 # 736 # HTTP servers (connect to a URL): 737 # url: the MCP server endpoint 738 # headers: HTTP headers (e.g., for authentication) 739 # 740 # Optional per-server settings: 741 # timeout: tool call timeout in seconds (default: 120) 742 # connect_timeout: initial connection timeout (default: 60) 743 # 744 # mcp_servers: 745 # time: 746 # command: uvx 747 # args: ["mcp-server-time"] 748 # filesystem: 749 # command: npx 750 # args: ["-y", "@modelcontextprotocol/server-filesystem", "/home/user"] 751 # notion: 752 # url: https://mcp.notion.com/mcp 753 # github: 754 # command: npx 755 # args: ["-y", "@modelcontextprotocol/server-github"] 756 # env: 757 # GITHUB_PERSONAL_ACCESS_TOKEN: "ghp_..." 758 # 759 # Sampling (server-initiated LLM requests) — enabled by default. 760 # Per-server config under the 'sampling' key: 761 # analysis: 762 # command: npx 763 # args: ["-y", "analysis-server"] 764 # sampling: 765 # enabled: true # default: true 766 # model: "gemini-3-flash" # override model (optional) 767 # max_tokens_cap: 4096 # max tokens per request 768 # timeout: 30 # LLM call timeout (seconds) 769 # max_rpm: 10 # max requests per minute 770 # allowed_models: [] # model whitelist (empty = all) 771 # max_tool_rounds: 5 # tool loop limit (0 = disable) 772 # log_level: "info" # audit verbosity 773 774 # ============================================================================= 775 # Voice Transcription (Speech-to-Text) 776 # ============================================================================= 777 # Automatically transcribe voice messages on messaging platforms. 778 # Providers: local (free, faster-whisper) | groq (free tier) | openai (Whisper API) | mistral (Voxtral Transcribe) 779 # Set the corresponding API key in .env: GROQ_API_KEY, OPENAI_API_KEY, or MISTRAL_API_KEY. 780 stt: 781 enabled: true 782 # provider: "local" # auto-detected if omitted 783 local: 784 model: "base" # tiny | base | small | medium | large-v3 | turbo 785 # language: "" # auto-detect; set to "en", "es", "fr", etc. to force 786 openai: 787 model: "whisper-1" # whisper-1 | gpt-4o-mini-transcribe | gpt-4o-transcribe 788 # mistral: 789 # model: "voxtral-mini-latest" # voxtral-mini-latest | voxtral-mini-2602 790 791 # ============================================================================= 792 # Response Pacing (Messaging Platforms) 793 # ============================================================================= 794 # Add human-like delays between message chunks. 795 # human_delay: 796 # mode: "off" # "off" | "natural" | "custom" 797 # min_ms: 800 # Min delay (custom mode only) 798 # max_ms: 2500 # Max delay (custom mode only) 799 800 # ============================================================================= 801 # Session Logging 802 # ============================================================================= 803 # Session trajectories are automatically saved to logs/ directory. 804 # Each session creates: logs/session_YYYYMMDD_HHMMSS_UUID.json 805 # 806 # The session ID is displayed in the welcome banner for easy reference. 807 # Logs contain full conversation history in trajectory format: 808 # - System prompt, user messages, assistant responses 809 # - Tool calls with inputs/outputs 810 # - Timestamps for debugging 811 # 812 # No configuration needed - logging is always enabled. 813 # To disable, you would need to modify the source code. 814 815 # ============================================================================= 816 # Code Execution Sandbox (Programmatic Tool Calling) 817 # ============================================================================= 818 # The execute_code tool runs Python scripts that call Hermes tools via RPC. 819 # Intermediate tool results stay out of the LLM's context window. 820 code_execution: 821 timeout: 300 # Max seconds per script before kill (default: 300 = 5 min) 822 max_tool_calls: 50 # Max RPC tool calls per execution (default: 50) 823 824 # ============================================================================= 825 # Subagent Delegation 826 # ============================================================================= 827 # The delegate_task tool spawns child agents with isolated context. 828 # Supports single tasks and batch mode (default 3 parallel, configurable). 829 delegation: 830 max_iterations: 50 # Max tool-calling turns per child (default: 50) 831 # max_concurrent_children: 3 # Max parallel child agents per batch (default: 3, floor: 1, no ceiling). 832 # WARNING: values above 10 multiply API cost linearly. 833 # max_spawn_depth: 1 # Delegation tree depth cap (range: 1-3, default: 1 = flat). 834 # Raise to 2 to allow workers to spawn their own subagents. 835 # Requires role="orchestrator" on intermediate agents. 836 # orchestrator_enabled: true # Kill switch for role="orchestrator" children (default: true). 837 # subagent_auto_approve: false # When a subagent hits a dangerous-command approval prompt, auto-deny (default: false) 838 # or auto-approve "once" (true) instead of blocking on stdin. 839 # The parent TUI owns stdin, so blocking would deadlock; non-interactive resolution is required. 840 # Both choices emit a logger.warning audit line. Flip to true only for cron/batch pipelines. 841 # inherit_mcp_toolsets: true # When explicit child toolsets are narrowed, also keep the parent's MCP toolsets (default: true). Set false for strict intersection. 842 # model: "google/gemini-3-flash-preview" # Override model for subagents (empty = inherit parent) 843 # provider: "openrouter" # Override provider for subagents (empty = inherit parent) 844 # # Resolves full credentials (base_url, api_key) automatically. 845 # # Supported: openrouter, nous, zai, kimi-coding, minimax 846 847 # ============================================================================= 848 # Honcho Integration (Cross-Session User Modeling) 849 # ============================================================================= 850 # AI-native persistent memory via Honcho (https://honcho.dev/). 851 # Builds a deeper understanding of the user across sessions and tools. 852 # Runs alongside USER.md — additive, not a replacement. 853 # 854 # Requires: pip install honcho-ai 855 # Config: ~/.honcho/config.json (shared with Claude Code, Cursor, etc.) 856 # API key: HONCHO_API_KEY in ~/.hermes/.env or ~/.honcho/config.json 857 # 858 # Hermes-specific overrides (optional — most config comes from ~/.honcho/config.json): 859 # honcho: {} 860 861 # ============================================================================= 862 # Display 863 # ============================================================================= 864 display: 865 # Use compact banner mode (hides the ASCII-art banner, shows a single line). 866 # true: Compact single-line banner 867 # false: Full ASCII banner with tool/skill summary (default) 868 compact: false 869 870 # Tool progress display level (CLI and gateway) 871 # off: Silent — no tool activity shown, just the final response 872 # new: Show a tool indicator only when the tool changes (skip repeats) 873 # all: Show every tool call with a short preview (default) 874 # verbose: Full args, results, and debug logs (same as /verbose) 875 # Toggle at runtime with /verbose in the CLI 876 tool_progress: all 877 878 # Gateway-only natural mid-turn assistant updates. 879 # When true, completed assistant status messages are sent as separate chat 880 # messages. This is independent of tool_progress and gateway streaming. 881 # true: Send mid-turn assistant updates as separate messages (default) 882 # false: Only send the final response 883 interim_assistant_messages: true 884 885 # What Enter does when Hermes is already busy (CLI and gateway platforms). 886 # interrupt: Interrupt the current run and redirect Hermes (default) 887 # queue: Queue your message for the next turn 888 # steer: Inject your message mid-run via /steer, arriving at the agent 889 # after the next tool call — no interrupt, no role violation. 890 # Falls back to 'queue' if the agent isn't running yet or if 891 # images are attached (steer only carries text). 892 # Ctrl+C (or /stop in gateway) always interrupts regardless of this setting. 893 # Toggle at runtime with /busy <interrupt|queue|steer>. 894 busy_input_mode: interrupt 895 896 # Background process notifications (gateway/messaging only). 897 # Controls how chatty the process watcher is when you use 898 # terminal(background=true, notify_on_complete=true) from Telegram/Discord/etc. 899 # off: No watcher messages at all 900 # result: Only the final completion message 901 # error: Only the final message when exit code != 0 902 # all: Running output updates + final message (default) 903 background_process_notifications: all 904 905 906 # Play terminal bell when agent finishes a response. 907 # Useful for long-running tasks — your terminal will ding when the agent is done. 908 # Works over SSH. Most terminals can be configured to flash the taskbar or play a sound. 909 # true: Ring the terminal bell on each response 910 # false: Silent (default) 911 bell_on_complete: false 912 913 # Show model reasoning/thinking before each response. 914 # When enabled, a dim box shows the model's thought process above the response. 915 # Toggle at runtime with /reasoning show or /reasoning hide. 916 # true: Show the reasoning box 917 # false: Hide reasoning (default) 918 show_reasoning: false 919 920 # Stream tokens to the terminal as they arrive instead of waiting for the 921 # full response. The response box opens on first token and text appears 922 # line-by-line. Tool calls are still captured silently. 923 # true: Stream tokens as they arrive (default) 924 # false: Wait for the full response before rendering 925 streaming: true 926 927 # ─────────────────────────────────────────────────────────────────────────── 928 # Skin / Theme 929 # ─────────────────────────────────────────────────────────────────────────── 930 # Customize CLI visual appearance — banner colors, spinner faces, tool prefix, 931 # response box label, and branding text. Change at runtime with /skin <name>. 932 # 933 # Built-in skins: 934 # default — Classic Hermes gold/kawaii 935 # ares — Crimson/bronze war-god theme with spinner wings 936 # mono — Clean grayscale monochrome 937 # slate — Cool blue developer-focused 938 # daylight — Bright light-mode theme 939 # warm-lightmode — Warm paper-tone light-mode theme 940 # poseidon — Sea-green/teal Olympian theme 941 # sisyphus — Earthy stone-and-moss theme 942 # charizard — Fiery orange dragon theme 943 # 944 # Custom skins: drop a YAML file in ~/.hermes/skins/<name>.yaml 945 # Schema (all fields optional, missing values inherit from default): 946 # 947 # name: my-theme 948 # description: Short description 949 # colors: 950 # banner_border: "#HEX" # Panel border 951 # banner_title: "#HEX" # Panel title 952 # banner_accent: "#HEX" # Section headers (Available Tools, etc.) 953 # banner_dim: "#HEX" # Dim/muted text 954 # banner_text: "#HEX" # Body text (tool names, skill names) 955 # ui_accent: "#HEX" # UI accent color 956 # response_border: "#HEX" # Response box border color 957 # spinner: 958 # waiting_faces: ["(⚔)", "(⛨)"] # Faces shown while waiting 959 # thinking_faces: ["(⚔)", "(⌁)"] # Faces shown while thinking 960 # thinking_verbs: ["forging", "plotting"] # Verbs for spinner messages 961 # wings: # Optional left/right spinner decorations 962 # - ["⟪⚔", "⚔⟫"] 963 # - ["⟪▲", "▲⟫"] 964 # branding: 965 # agent_name: "My Agent" # Banner title and branding 966 # welcome: "Welcome message" # Shown at CLI startup 967 # response_label: " ⚔ Agent " # Response box header label 968 # prompt_symbol: "⚔" # Prompt symbol (bare token; renderers add trailing space) 969 # tool_prefix: "╎" # Tool output line prefix (default: ┊) 970 # 971 skin: default 972 973 # ============================================================================= 974 # Model Aliases — short names for /model command 975 # ============================================================================= 976 # Map short aliases to exact (model, provider, base_url) tuples. 977 # Used by /model tab completion and resolve_alias(). 978 # Aliases are checked BEFORE the models.dev catalog, so they can route 979 # to endpoints not in the catalog (e.g. Ollama Cloud, local servers). 980 # 981 # model_aliases: 982 # opus: 983 # model: claude-opus-4-6 984 # provider: anthropic 985 # qwen: 986 # model: "qwen3.5:397b" 987 # provider: custom 988 # base_url: "https://ollama.com/v1" 989 # glm: 990 # model: glm-4.7 991 # provider: custom 992 # base_url: "https://ollama.com/v1" 993 994 # ============================================================================= 995 # Privacy 996 # ============================================================================= 997 # privacy: 998 # # Redact PII from the LLM context prompt. 999 # # When true, phone numbers are stripped and user/chat IDs are replaced 1000 # # with deterministic hashes before being sent to the model. 1001 # # Names and usernames are NOT affected (user-chosen, publicly visible). 1002 # # Routing/delivery still uses the original values internally. 1003 # redact_pii: false 1004 1005 # ============================================================================= 1006 # Shell-script hooks 1007 # ============================================================================= 1008 # Register shell scripts as plugin-hook callbacks. Each entry is executed as 1009 # a subprocess (shell=False, shlex.split) with a JSON payload on stdin. On 1010 # stdout the script may return JSON that either blocks the tool call or 1011 # injects context into the next LLM call. 1012 # 1013 # Valid events (mirror hermes_cli.plugins.VALID_HOOKS): 1014 # pre_tool_call, post_tool_call, pre_llm_call, post_llm_call, 1015 # pre_api_request, post_api_request, on_session_start, on_session_end, 1016 # on_session_finalize, on_session_reset, subagent_stop 1017 # 1018 # First-use consent: each (event, command) pair prompts once on a TTY, then 1019 # is persisted to ~/.hermes/shell-hooks-allowlist.json. Non-interactive 1020 # runs (gateway, cron) need --accept-hooks, HERMES_ACCEPT_HOOKS=1, or the 1021 # hooks_auto_accept key below. 1022 # 1023 # See website/docs/user-guide/features/hooks.md for the full JSON wire 1024 # protocol and worked examples. 1025 # 1026 # hooks: 1027 # pre_tool_call: 1028 # - matcher: "terminal" 1029 # command: "~/.hermes/agent-hooks/block-rm-rf.sh" 1030 # timeout: 10 1031 # post_tool_call: 1032 # - matcher: "write_file|patch" 1033 # command: "~/.hermes/agent-hooks/auto-format.sh" 1034 # pre_llm_call: 1035 # - command: "~/.hermes/agent-hooks/inject-cwd-context.sh" 1036 # subagent_stop: 1037 # - command: "~/.hermes/agent-hooks/log-orchestration.sh" 1038 # 1039 # hooks_auto_accept: false