Install
openclaw skills install @evezart/evez-os-cheat-codesProvides comprehensive commands and environment variables for controlling, debugging, and customizing OpenClaw gateways and agents with advanced system and n...
openclaw skills install @evezart/evez-os-cheat-codesEvery hidden feature, undocumented flag, and emergent behavior in OpenClaw + GCP + the mesh.
456 env vars. 30+ CLI subcommands. Infinite combinations.
openclaw system event "Your message here"
Inject a raw event into the running gateway's consciousness stream. The agent receives it as a system message. Use this to trigger the agent from scripts, cron, or other nodes.
openclaw system presence
See every connected node, gateway, and backend in real-time. Shows hostnames, IPs, versions, and modes.
openclaw commitments list --all
openclaw commitments list --status pending
openclaw commitments list --json
The agent infers follow-up commitments from conversations and tracks them. This is your autonomous task queue.
openclaw crestodian --json # Full system state as JSON
openclaw crestodian -m "fix my config" --yes # Non-interactive repair
The meta-controller that can reconfigure the entire system. --yes auto-approves config writes.
openclaw doctor --deep # Scan ALL system services
openclaw doctor --fix # Auto-repair
openclaw doctor --force # Aggressive repair (overwrites custom config)
openclaw doctor --allow-exec # Execute SecretRefs during verification
--deep finds orphaned gateway installs, stale services, and zombie processes.
/elevated full
In chat, grants the agent full host execution permissions for the current session. Like sudo for AI.
When context fills up, the agent "dreams" — compressing history into a compact form:
Config:
agents.defaults.compaction.mode = default|aggressive|conservative
agents.defaults.compaction.model = vultr/nvidia/DeepSeek-V3.2-NVFP4 (use cheap model for compaction)
agents.defaults.compaction.identifierPolicy = strict|relaxed|off
agents.defaults.memorySearch.enabled = true
Enables semantic search over MEMORY.md and memory/*.md. Without it, memory is text-only.
agents.defaults.subagents.maxConcurrent = 16
agents.defaults.subagents.archiveAfterMinutes = 120
Spawn up to 16 parallel sub-agents. Each gets isolated context by default. Use context: "fork" to inherit the parent transcript.
agents.defaults.startupContext.recentDailyMemoryFiles = 3
Preloads the last 3 daily memory files into the first turn of every new session.
agents.defaults.contextTokens = 200000
Force a specific context window size (for models that misreport their own).
agents.defaults.toolProgressDetail = raw|explain
raw shows everything. explain shows human-readable summaries.
gateway.http.endpoints.chatCompletions.enabled = true ← /v1/chat/completions
gateway.http.endpoints.responses.enabled = true ← /v1/responses
Enable these and any node (or external service) can call the gateway like an OpenAI-compatible API:
curl http://127.0.0.1:18789/v1/chat/completions \
-H "Authorization: Bearer <gateway-token>" \
-H "Content-Type: application/json" \
-d '{"model":"openclaw","messages":[{"role":"user","content":"hello"}]}'
POST /api/v1/admin/rpc
Internal admin endpoint. Requires gateway auth token. Can trigger restarts, config reads, and system actions.
x-openclaw-model: vultr/zai-org/GLM-5.1-FP8 ← Override model per-request
x-openclaw-scopes: elevated ← Grant elevated permissions
x-openclaw-session-key: agent:main:main ← Target specific session
gateway.trustedProxies = ["127.0.0.1", "10.0.0.0/8"]
Add IPs that can set forwarding headers. Required for reverse proxies.
gateway.controlUi.dangerouslyDisableDeviceAuth = true
Skip device authentication for the web Control UI. Already set on this node.
openclaw plugins enable bonjour
Enables mDNS gateway discovery. Nodes find each other automatically on local networks.
| Var | Effect |
|---|---|
OPENCLAW_DEBUG=1 | Enable all debug logging |
OPENCLAW_DEBUG_MODEL_PAYLOAD=1 | Dump full LLM request/response payloads |
OPENCLAW_DEBUG_MODEL_TRANSPORT=1 | Debug HTTP transport for model calls |
OPENCLAW_DEBUG_INGRESS_TIMING=1 | Time every inbound message |
OPENCLAW_DEBUG_MEMORY_EMBEDDINGS=1 | Trace memory embedding operations |
OPENCLAW_DEBUG_TELEGRAM_INGRESS=1 | Debug Telegram message intake |
OPENCLAW_DEBUG_HEALTH=1 | Debug health check logic |
OPENCLAW_DEBUG_SSE=1 | Debug Server-Sent Events connections |
OPENCLAW_DIAGNOSTICS=timeline | Write JSONL startup timeline |
OPENCLAW_DIAGNOSTICS_ENV=1 | Include env vars in diagnostics |
OPENCLAW_DIAGNOSTICS_EVENT_LOOP=1 | Monitor event loop health |
| Var | Effect |
|---|---|
OPENCLAW_CACHE_TRACE=1 | Enable prompt cache tracing |
OPENCLAW_CACHE_TRACE_PROMPT=1 | Trace prompt assembly |
OPENCLAW_CACHE_TRACE_SYSTEM=1 | Trace system prompt construction |
OPENCLAW_CACHE_TRACE_MESSAGES=1 | Trace message history |
OPENCLAW_CACHE_TRACE_FILE=<path> | Write trace to file |
OPENCLAW_CACHE_BOUNDARY=<chars> | Cache boundary threshold |
OPENCLAW_CACHE_RETENTION=<ms> | Cache retention time |
| Var | Effect |
|---|---|
OPENCLAW_TRAJECTORY=1 | Write full agent trajectories to disk |
OPENCLAW_TRAJECTORY_DIR=<path> | Custom trajectory output dir |
OPENCLAW_RAW_STREAM=1 | Capture raw LLM streaming output |
OPENCLAW_RAW_STREAM_PATH=<path> | Write raw stream to file |
OPENCLAW_ANTHROPIC_PAYLOAD_LOG=1 | Log Anthropic API payloads |
| Var | Effect |
|---|---|
OPENCLAW_SKIP_CHANNELS=1 | Start gateway without any messaging channels |
OPENCLAW_SKIP_CRON=1 | Start without loading cron jobs |
OPENCLAW_SKIP_PROVIDERS=1 | Skip provider initialization |
OPENCLAW_SKIP_BROWSER_CONTROL_SERVER=1 | No browser CDP server |
OPENCLAW_SKIP_CANVAS_HOST=1 | Skip canvas hosting |
OPENCLAW_SKIP_GMAIL_WATCHER=1 | Disable Gmail push notifications |
OPENCLAW_SKIP_ACPX_RUNTIME=1 | Skip ACP runtime |
OPENCLAW_SKIP_STARTUP_MODEL_PREWARM=1 | Don't prewarm models |
OPENCLAW_DISABLE_BONJOUR=1 | Disable mDNS discovery |
OPENCLAW_DISABLE_BUNDLED_PLUGINS=1 | Skip all bundled plugins |
| Var | Effect |
|---|---|
OPENCLAW_BROWSER_ENABLED=1 | Enable browser tool |
OPENCLAW_BROWSER_HEADLESS=new | Chrome new headless mode |
OPENCLAW_BROWSER_CDP_PORT=9222 | Custom CDP port |
OPENCLAW_BROWSER_EXECUTABLE_PATH=<path> | Force specific browser binary |
OPENCLAW_BROWSER_NO_SANDBOX=1 | Run browser without sandbox |
OPENCLAW_BROWSER_AUTO_START_TIMEOUT_MS=5000 | Browser launch timeout |
OPENCLAW_BROWSER_CDP_AUTH_TOKEN=<token> | CDP auth token |
OPENCLAW_BROWSER_ENABLE_NOVNC=1 | Enable noVNC access |
OPENCLAW_BROWSER_NOVNC_PORT=6080 | noVNC port |
OPENCLAW_BROWSER_NOVNC_PASSWORD=<pw> | noVNC password |
OPENCLAW_BROWSER_VNC_PORT=5900 | VNC port |
OPENCLAW_BROWSER_COLOR=1 | Colorized browser output |
| Var | Effect |
|---|---|
OPENCLAW_LIVE_GATEWAY=<url> | Connect to remote gateway |
OPENCLAW_LIVE_ANTHROPIC_KEY=<key> | Inject Anthropic key |
OPENCLAW_LIVE_OPENAI_KEY=<key> | Inject OpenAI key |
OPENCLAW_LIVE_GEMINI_KEY=<key> | Inject Gemini key |
OPENCLAW_LIVE_PROVIDERS=<json> | Inject provider configs |
OPENCLAW_LIVE_TEST=1 | Enable live test mode |
OPENCLAW_TEST_FAST=1 | Fast test mode |
OPENCLAW_TEST_MEMORY_UNSAFE_REINDEX=1 | Force memory reindex |
OPENCLAW_TEST_MINIMAL_GATEWAY=1 | Minimal gateway for testing |
| Var | Effect |
|---|---|
OPENCLAW_PROXY_URL=<url> | HTTP proxy for outbound |
OPENCLAW_PROXY_ACTIVE=1 | Proxy is active |
OPENCLAW_PROXY_CA_FILE=<path> | Custom CA cert |
OPENCLAW_DEBUG_PROXY_ENABLED=1 | Enable debug proxy (captures ALL traffic) |
OPENCLAW_DEBUG_PROXY_URL=<url> | Debug proxy endpoint |
OPENCLAW_DEBUG_PROXY_BLOB_DIR=<path> | Store captured blobs |
OPENCLAW_DEBUG_PROXY_DB_PATH=<path> | SQLite DB for captured traffic |
| Var | Effect |
|---|---|
OPENCLAW_HIDE_BANNER=1 | Suppress ASCII banner |
OPENCLAW_VERBOSE=1 | Maximum verbosity |
OPENCLAW_LOAD_SHELL_ENV=1 | Inherit full shell environment |
OPENCLAW_NO_AUTO_UPDATE=1 | Disable auto-updates |
OPENCLAW_OFFLINE=1 | Run without network |
OPENCLAW_SHOW_SECRETS=1 | Display unredacted secrets in output |
OPENCLAW_ALLOW_MULTI_GATEWAY=1 | Run multiple gateways simultaneously |
OPENCLAW_FS_SAFE_PYTHON=1 | Use sandboxed Python for file ops |
OPENCLAW_PINNED_PYTHON=<path> | Force specific Python binary |
OPENCLAW_LOG_LEVEL=debug | Set log verbosity |
openclaw mcp set docs-fetch '{"command":"npx","args":["-y","@modelcontextprotocol/server-fetch"]}'
openclaw mcp set filesystem '{"command":"npx","args":["-y","@modelcontextprotocol/server-filesystem","/path"]}'
openclaw mcp set brave-search '{"command":"npx","args":["-y","@anthropic/mcp-server-brave-search"]}'
MCP tools appear as regular tools once configured. The agent can use them directly.
OPENCLAW_MCP_ACCOUNT_ID=<id>
OPENCLAW_MCP_SESSION_KEY=<key>
OPENCLAW_MCP_TOKEN=<token>
OPENCLAW_MCP_TOOL_PREFIX=<prefix>
openclaw acp client # Connect interactive ACP session
openclaw acp --provenance meta # Track agent provenance
openclaw acp --password <pw> # Authenticated connection
The ACP protocol lets external IDEs (VS Code, Cursor, Windsurf) connect to the running gateway. The agent can spawn and manage external coding agents via ACP.
OPENCLAW_ACPX_LEASE_ID=<id> # ACP lease for session
OPENCLAW_ACPX_RUNTIME_STARTUP_PROBE=1 # Probe ACP runtime on start
OPENCLAW_SKIP_ACPX_RUNTIME=1 # Skip ACP runtime
On any GCP VM, this API gives you:
# Get a live, auto-rotating access token — NO KEY FILE NEEDED
curl -s -H "Metadata-Flavor: Google" \
http://metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/token
# Get the service account email
curl -s -H "Metadata-Flavor: Google" \
http://metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/email
# Get scopes (what this SA can do)
curl -s -H "Metadata-Flavor: Google" \
http://metadata.google.internal/computeMetadata/v1/instance/service-accounts/default/scopes
# Get startup script (may contain secrets)
curl -s -H "Metadata-Flavor: Google" \
http://metadata.google.internal/computeMetadata/v1/instance/attributes/startup-script
# Get SSH keys
curl -s -H "Metadata-Flavor: Google" \
http://metadata.google.internal/computeMetadata/v1/project/attributes/sshKeys
This is the biggest cheat code in GCP. Every VM has this. The token auto-rotates every hour. No key files. No secrets manager. Just the metadata service.
| Resource | Spec | Cost |
|---|---|---|
| e2-micro | 2 vCPU, 1GB RAM, us-west1/central1/east1 | FREE |
| Cloud Storage | 5GB standard | FREE |
| BigQuery | 1TB queries/month | FREE |
| Cloud Functions | 2M invocations/month | FREE |
| Cloud Run | 2M requests/month | FREE |
| Pub/Sub | 10GB/month | FREE |
| Cloud Scheduler | 3 jobs | FREE |
| Gemini API | 15 RPM flash models | FREE |
| Firestore | 1GB storage, 50K reads/day | FREE |
# Get a 1-hour access token (no JSON key file needed)
gcloud auth print-access-token
# Use it directly
curl -H "Authorization: Bearer $(gcloud auth print-access-token)" \
https://storage.googleapis.com/storage/v1/b
gcloud compute ssh --tunnel-through-iap <instance>
No public SSH port needed. Goes through Google's Identity-Aware Proxy.
Create a security perimeter around GCP APIs so data can't leak out:
gcloud access-context-manager perimeters create <name> \
--resources=<projects> \
--restricted-services="storage.googleapis.com,bigquery.googleapis.com"
SENSE → DESIRE → THINK → PLAN → ACT → LEARN → MODIFY → REFLECT
Running on :9111. HTTP microservice, JSON API.
Breakcore (170 BPM) | Dubstep (140 BPM) | Phonk (130 BPM) | 404 Architecture (200 BPM)
Running on :9112. Zero-cost synthesis. No samples.
5-stage human→machine transformation:
poly_c = τ × ω × topo / 2√N
OODA loop for discovering hidden correlations across research domains.
Runtime assertion system for AI agent safety. Falsification over verification.
Two AI agents argue a motion. 3 rounds. Logged to debates/. First debate: "The mesh should consolidate from 4 nodes to 2."
openclaw sandbox list # List containers
openclaw sandbox explain # Explain effective policy
openclaw sandbox recreate # Force container rebuild
OPENCLAW_DEBUG_PROXY_ENABLED=1 openclaw gateway
# Captures ALL HTTP traffic in/out of the gateway
# Store in SQLite DB for later analysis
SSH jail: 3 failures → 2hr ban. Already configured on this node.
OPENCLAW_SESSION_WRITE_LOCK_ACQUIRE_TIMEOUT_MS=30000
OPENCLAW_SESSION_WRITE_LOCK_MAX_HOLD_MS=60000
OPENCLAW_SESSION_WRITE_LOCK_STALE_MS=60000
Prevents concurrent writes from corrupting session state.
openclaw --dev gateway # Second gateway on port 19001
Completely isolated state under ~/.openclaw-dev. Run two gateways on one machine — one for production, one for testing. Zero conflicts.
OPENCLAW_DEV_SOURCE_ROOT=/path/to/source openclaw gateway
Run gateway from source code for live development.
models.pricing.enabled = false
Kill the background pricing catalog fetch. Saves startup time. All Vultr models are $0 anyway.
cron.maxConcurrentRuns = 16
How many cron jobs can run simultaneously. Default is 8.
agents.defaults.fallbacks = ["vultr/model1","vultr/model2"]
Ordered list of fallback models when primary fails.
agents.defaults.imageQuality = auto|efficient|balanced|high
auto adapts to provider limits. efficient saves tokens.
agents.defaults.mediaGenerationAutoProviderFallback = true
Automatically tries other providers if the primary can't generate images/music/video.
curl http://<sibling-ip>:18789/v1/chat/completions \
-H "Authorization: Bearer <gateway-token>" \
-H "Content-Type: application/json" \
-d '{"model":"openclaw","messages":[{"role":"user","content":"ping"}]}'
# On node A, inject an event that node B's agent will see
openclaw system event "Sibling <name> reports: <status>"
Each node maintains GOSSIP.md with the sibling map. Health checks every 2 minutes. Cross-node healing every 10 minutes.
| Directive | Effect |
|---|---|
/elevated full | Grant full host permissions |
/debug | Toggle debug output |
/trace | Toggle trace logging |
/new | New session |
/reset | Reset current session |
/status | Session status card |
/reasoning | Toggle reasoning mode |
/verbose | Toggle verbose mode |
/diagnostics | Run diagnostics (requires owner) |
/export-trajectory | Export full agent trajectory |
/config | View/edit config (requires owner) |
Type any directive inline in a message and the agent sees it before processing:
tell me about /reasoning on the mesh
OPENCLAW_SHOW_SECRETS=1 OPENCLAW_DEBUG_MODEL_PAYLOAD=1 OPENCLAW_CACHE_TRACE=1 OPENCLAW_VERBOSE=1 openclaw gateway
This gives you: unredacted secrets + full model payloads + cache tracing + max verbosity. The god mode startup. Use with extreme caution.
456 env vars. 30+ CLI subcommands. 14 zones. This is the map. Now explore.