Cost Guardian by Dexter Labs

Monitor and control OpenClaw API costs. Tracks token usage across all sessions, estimates spend by model, alerts on budget overruns, and recommends cheaper m...

Security Scan

The skill's code and instructions match its stated purpose (reading local OpenClaw session data and estimating costs), it requires only python3 and does not request secrets or make external network calls.

Purpose & Capability

Name/description (cost monitoring and optimization) aligns with the included script and SKILL.md. The script reads OpenClaw session data (~/.openclaw/agents/<agent>/sessions/sessions.json) and computes per-model/session cost estimates and recommendations — this is exactly what the skill claims to do. Requiring python3 is appropriate.

Instruction Scope

SKILL.md stays focused on cost reporting and optimization, and the script only reads a local sessions.json and prints or emits JSON reports. Two items to note: (1) SKILL.md asks the agent to create cron jobs to run the script and to "deliver the report via your preferred channel" — that grants the agent discretion to modify system crontab and to send the report over whatever delivery channel the agent uses (chat, email, webhook, etc.). These behaviors are consistent with 'automated reports' but are broader system actions than simply reading session data. (2) The script will exit with an error if the session file is missing; not a security problem but could cause noisy cron output.

Install Mechanism

No install spec (instruction-only plus a python script) — nothing is downloaded or extracted. Lowest risk for install mechanism; only requires a local python3 interpreter which is declared.

Credentials

The skill requests no environment variables or credentials. It reads the user's OpenClaw session store (a local file) which is necessary for cost calculation. There are no unrelated secrets requested.

Persistence & Privilege

Metadata does not request elevated or persistent privileges (always:false). However, SKILL.md explicitly instructs the agent to create cron jobs (system crontab changes) for automated reports. That is a legitimate feature for automation but is a system modification that the user should approve or perform manually.

Assessment

This skill appears to do what it says: it reads the OpenClaw sessions.json in your home directory, computes estimated costs, and prints or outputs JSON. Before installing/using: 1) Inspect ~/.openclaw/agents/<agent>/sessions/sessions.json to understand what data will be read (it may include conversation metadata and token counts or even message text). 2) Run scripts/cost-report.py locally once to review output and ensure the path and agent-id are correct. 3) If you don't want an agent to modify your system, set up the cron job yourself rather than giving the agent instructions to create it; cron entries will run with whatever permissions the user has. 4) Be aware that "deliver via your preferred channel" means the agent may transmit report contents over chat/email/webhooks — if your reports contain sensitive text snippets, route them carefully. 5) If you need different pricing, update the MODEL_PRICING dict in the script. Overall the skill is coherent and contained, but grant automation (crontab or delivery) only with explicit consent.

Cost Guardian — OpenClaw Cost Monitor & Optimizer

Know what your agent costs. Before it becomes a problem.

When to Use

✅ USE this skill when:

• "How much am I spending?"
• "What's my API cost?"
• "Which sessions cost the most?"
• "Am I over budget?"
• "How can I reduce costs?"
• Setting up daily/weekly cost reports
• Optimizing model usage for crons and automated tasks

❌ DON'T use this skill for:

• Billing issues with Anthropic/OpenAI (contact providers directly)
• Managing API keys (use OpenClaw's secrets system)

Quick Start

Get a Cost Report (All Time)

python3 scripts/cost-report.py --all

Last 24 Hours

python3 scripts/cost-report.py --hours 24

With Budget Alert ($5/day)

python3 scripts/cost-report.py --budget 5.00

JSON Output (for dashboards)

python3 scripts/cost-report.py --all --json

What It Reports

• Estimated cost by model and session
• Token breakdown — input, output, context (cached)
• Top spending sessions — find what's burning tokens
• Model efficiency — how much you'd save switching routine tasks to cheaper models
• Budget alerts — 🟡 at 80% and 🔴 when over budget
• Cron cost tracking — automated jobs often account for 30-50% of spend

Setting Up Automated Reports

Daily Cost Report via Cron

Tell your agent:

"Set up a daily cost report that runs at 8 AM and alerts me if I'm over $5/day"

The agent should create a cron job that:

1. Runs python3 <skill-dir>/scripts/cost-report.py --hours 24 --budget 5.00
2. Delivers the report via your preferred channel

Model Optimization

The report flags when expensive models (Opus) are being used for routine tasks that Sonnet handles fine:

• Email checking crons → Sonnet
• Heartbeat checks → Sonnet
• Site health monitoring → Sonnet or Haiku
• Complex reasoning, strategy, writing → Keep on Opus

Supported Models & Pricing

Model	Input $/1M	Output $/1M	Cache $/1M
Claude Opus 4	$15.00	$75.00	$1.875
Claude Sonnet 4	$3.00	$15.00	$0.30
Claude Haiku 3.5	$0.80	$4.00	$0.08
GPT-4o	$2.50	$10.00	$1.25
GPT-4.1	$2.00	$8.00	$0.50
GPT-4.1-mini	$0.40	$1.60	$0.10
GPT-4.1-nano	$0.10	$0.40	$0.025
OpenRouter/auto	~$3.00	~$15.00	~$0.30

Pricing is estimated and may vary. Update scripts/cost-report.py MODEL_PRICING dict for current rates.

Roadmap

• Auto model-switching recommendations per session type
• Historical trend tracking (daily/weekly/monthly)
• Cost anomaly detection (sudden spikes)
• Per-project cost allocation
• Web dashboard with charts