Stores durable facts in a categorized, plain-markdown vault on disk, alongside your agent's built-in memory.
by Iván559 installs18 stars
Documents
Token Optimizer for OpenClaw & Hermes
Try itReduce recurring OpenClaw token waste via lazy context, Sonnet/Opus routing, and cache-aware heartbeats.
What it does
Local scripts for OpenClaw 2026.6.x agents that recommend the minimum context for each prompt, route routine work to Sonnet and complex reasoning to Opus, schedule heartbeat checks with quiet hours and cache-TTL awareness, and track local daily token budgets. Writes to the workspace only when explicitly requested, and backs up HEARTBEAT.md before replacement. Includes a SHA256 manifest (.clawhubsafe) for verifying installed files.
When to use it
- Picking the minimum context bundle for a prompt
- Routing a routine reply to Sonnet instead of Opus
- Scheduling heartbeat checks around quiet hours and cache TTL
- Checking a local daily token budget before expensive work
The skill document
OpenClaw Token Optimizer v3.2.0
Built for current OpenClaw 2026.6.x agents by Asif2BD · GitHub · Security Notes
Security notice: local-only optimization toolkit. The Python scripts make no network requests and do not execute dynamic code. Commands that write files are explicit, documented, and backup-safe.
Install
openclaw skills install @asif2bd/openclaw-token-optimizer
What's New in v3.2
Modern ClawHub Card
The skill card now leads with a clearer product promise, current OpenClaw compatibility, security posture, and practical commands instead of the older long-form v3.0 copy.
Current OpenClaw Model Routing
Routine Anthropic work maps to anthropic/claude-sonnet-4-5; only explicitly complex reasoning routes to anthropic/claude-opus-4-5.
Audit-Safe Writes
context_optimizer.py generate-agentsprints to stdout by default.- Writing an optimized
AGENTS.mdrequires--outputor--workspace-output. ./scripts/optimize.sh heartbeatpreviews only by default../scripts/optimize.sh heartbeat installwritesHEARTBEAT.mdonly after backing up an existing file.
What It Optimizes
Context Loading
Stop loading every workspace file into every session. Recommend only the context needed for the current prompt.
python3 scripts/context_optimizer.py recommend "hi"
Typical result: SOUL.md + IDENTITY.md only for a greeting.
Model Routing
Keep standard work on Sonnet and reserve Opus for deep reasoning.
python3 scripts/model_router.py "thanks!"
python3 scripts/model_router.py "design a multi-agent architecture"
Heartbeat Scheduling
Run only checks that are due, respect quiet hours, and align heartbeat timing with prompt-cache TTL.
python3 scripts/heartbeat_optimizer.py plan
python3 scripts/heartbeat_optimizer.py cache-ttl
Token Budgets
Track local daily budget state and surface warnings before runaway spend.
python3 scripts/token_tracker.py check
Quick Start
1. Find the minimum context for a request
python3 scripts/context_optimizer.py recommend "write a deployment checklist"
2. Preview an optimized AGENTS.md
python3 scripts/context_optimizer.py generate-agents
Save a review copy only when you explicitly ask for it:
python3 scripts/context_optimizer.py generate-agents --output ~/.openclaw/workspace/AGENTS.md.optimized
3. Preview heartbeat installation
./scripts/optimize.sh heartbeat
Install with backup protection:
./scripts/optimize.sh heartbeat install
4. Route a task
python3 scripts/model_router.py "debug this error"
Savings Profile
| Area | Typical waste | Optimizer approach |
|---|---|---|
| Startup context | Loading docs, memory, tools, and old logs every turn | Lazy-load only what the prompt needs |
| Routine model use | Opus used for quick replies or simple checks | Sonnet for routine work, Opus for complex reasoning |
| Heartbeats | Every check runs on every heartbeat | Due-check planner + quiet hours |
| Cache TTL | Idle sessions trigger expensive cache rewrites | 55-minute heartbeat guidance for 1-hour cache TTL |
| Budget control | Spend discovered after the fact | Local budget checks before expensive work |
Expected savings: 50-80% in context-heavy OpenClaw workspaces, with context optimization usually delivering the biggest win.
Files
Scripts
scripts/context_optimizer.py— prompt classification and context bundle recommendationsscripts/model_router.py— Sonnet/Opus-aware task routingscripts/heartbeat_optimizer.py— due-check planning, quiet hours, cache-TTL interval guidancescripts/token_tracker.py— local budget trackingscripts/optimize.sh— convenience wrapper around bundled scripts
Assets
assets/HEARTBEAT.template.md— backup-safe heartbeat templateassets/cronjob-model-guide.md— scheduled-task model guidanceassets/config-patches.json— native and optional config examples
References
references/PROVIDERS.md— optional external-provider strategy notesSECURITY.md— complete security and data-handling breakdown
Security Posture
- No network calls in executable scripts
- No
eval, dynamic imports, shell command execution, or downloader behavior - Local JSON state only under the OpenClaw workspace memory directory
- Workspace writes require explicit commands
- Existing
HEARTBEAT.mdis backed up before replacement .clawhubsafecovers every published file with SHA256 hashes
Verify locally:
sha256sum -c .clawhubsafe
Native OpenClaw Diagnostics
Use OpenClaw's built-in context and usage commands alongside this skill:
/context list
/context detail
/usage tokens
/usage cost
Use the built-ins to see where tokens are going, then use Token Optimizer to reduce the recurring waste.
Best Fit
- Multi-agent OpenClaw workspaces with large instruction files
- Agents that run heartbeats or scheduled checks
- Teams trying to keep Opus for hard work instead of routine work
- xCloud, MissionDeck, or other hosted OpenClaw deployments where every saved prompt token compounds across users
More by Asif2BD
openclaw skills install @asif2bd/jarvis-mission-control
openclaw skills search asif2bd
Related skills
Find why your productivity system keeps failing, then apply the smallest fix — capacity math, bottleneck routing, durable local notes.
by Iván855 installs69 stars
Join a video meeting as an AI bot with voice, avatar, and screenshare across four operating modes.
by johnpatternai22 installs8 stars
Generate and edit Draw.io, Mermaid, and Excalidraw diagrams from natural language using a structured JSON spec.
by nssa.io1.0k installs47 stars
Fetch raw ad creative, app, ranking, and revenue data from AdMapix as structured JSON.
by fly0pants4.3k installs296 stars
Site audit, content writing, and competitor analysis for organic search rankings.
by Iván574 installs31 stars
More from asif2bd
Browse all skillsSelf-hosted dashboard for monitoring OpenCLAW AI sessions and managing tasks.
by asif2bd95 installs4 stars
Get a Docker project ready for xCloud with confirmation-gated file edits and port fixes.
by asif2bd38 installs
Publish content directly to WordPress sites via REST API with full Gutenberg block support. Create and publish posts/pages, auto-load and select categories from website, generate SEO-optimized tags, preview articles before publishing, and generate Gutenberg blocks for tables, images, lists, and rich formatting. Use when user wants to publish to WordPress, post to blog, create WordPress article, update WordPress post, or convert markdown to Gutenberg blocks.
by asif2bd189 installs10 stars
Universal travel planning skill for OpenClaw agents. Plan itineraries, check weather, discover attractions, and estimate budgets — all through natural conver...
by asif2bd33 installs
Install, authenticate, and use Claude Code CLI as a native coding tool for any OpenClaw agent system.
by asif2bd32 installs
Dramatically reduce per-session token usage by loading skills and context files only when needed — not at session start. Includes the SKILLS catalog pattern,...
by asif2bd36 installs