Claude Usage

Calculate Claude Max subscription usage from OpenClaw session data. Shows credits consumed, weekly budget percentage, 5-hour rate limit window, and per-session breakdown.

Security Scan

The skill's code and instructions are consistent with its stated purpose: it reads local OpenClaw session JSONL files, computes Claude Max credits usage, and saves a small local config file — it does not request credentials or perform network exfiltration.

Purpose & Capability

Name/description (Claude usage from OpenClaw session data) match the included files: a Python scanner that reads OpenClaw session JSONL files, computes credits using the documented formula, and reports per-session and windowed usage. No extraneous cloud credentials or unrelated dependencies are requested.

Instruction Scope

SKILL.md instructs the user to provide the weekly reset time and plan and then run the bundled Python script. The runtime instructions and script operate on session files under ~/.openclaw/agents/... and save a small config; they do not instruct reading unrelated system files, scanning other home directories, or sending data to external endpoints.

Install Mechanism

There is no install spec in the skill bundle (instruction-only install), which is low risk. The package includes package.json and a small Node wrapper that execs the Python script (normal convenience). README suggests npm installation or ClawHub, but the skill itself does not perform any remote downloads during runtime.

Credentials

The skill declares no required environment variables or credentials. The script reads OpenClaw session files and writes a local config (~/.openclaw/claude-usage.json), which is proportionate to its purpose. It does not request unrelated secrets.

Persistence & Privilege

The script will create and update a small config file in the user's home (~/.openclaw/claude-usage.json) to store reset_time/plan/tz — this is expected behavior for a local CLI tool. The skill is not marked always:true and does not modify other skills or system-wide agent settings.

Assessment

This tool appears to do what it says: it reads local OpenClaw session JSONL files (auto-detected under ~/.openclaw/agents/main/sessions), computes Claude credits, prints reports, and saves a config at ~/.openclaw/claude-usage.json. Before installing/running: (1) Verify you have Python 3.9+ and are comfortable the script will read those session files; (2) inspect the included scripts if you want to confirm there's no network activity (the provided code shows no HTTP/remote calls); (3) be aware session transcripts may contain sensitive content — the tool processes them locally and does not request external credentials. If you want additional assurance, run the Python script in a sandbox or review the full script contents (it is included in the bundle).

Claude Max Usage Calculator

Track your Claude Max subscription usage based on the reverse-engineered credits system from she-llac.com/claude-limits.

When to use

When the user asks about Claude usage, credits, subscription consumption, rate limits, or wants to know how much budget is left.

Setup (first time)

Ask the user for:

1. Weekly reset time — visible on claude.ai Settings > Usage (e.g. "Resets Mon 2:00 PM")
2. Plan — pro ($20), 5x ($100), or 20x ($200). Default: 5x

Save it so they never have to repeat:

python3 {SKILL_DIR}/scripts/claude-usage.py "2026-02-09 14:00" --plan 5x --save

Commands

# Weekly overview (uses saved config after first --save)
python3 {SKILL_DIR}/scripts/claude-usage.py

# Override plan or timezone
python3 {SKILL_DIR}/scripts/claude-usage.py --plan 20x --tz "America/New_York"

# Top N sessions only
python3 {SKILL_DIR}/scripts/claude-usage.py --top 5

# Single session detail (substring match on key or id)
python3 {SKILL_DIR}/scripts/claude-usage.py --session "main"
python3 {SKILL_DIR}/scripts/claude-usage.py --session "9aadee"

# JSON output
python3 {SKILL_DIR}/scripts/claude-usage.py --json

What it shows

Weekly overview

• Total credits used vs weekly budget (with progress bar)
• 5-hour sliding window — warns if approaching the per-5h rate limit
• All sessions ranked by credits consumed
• Model breakdown (Opus/Sonnet/Haiku/non-Claude)

Single session detail (--session)

• Credits consumed and % of weekly budget
• % of total usage (how much of your spending this session accounts for)
• Token breakdown (input/output/cache)
• Per-model detail

Credits formula

credits = (input_tokens + cache_write_tokens) × input_rate + output_tokens × output_rate

Model	Input rate	Output rate
Haiku	2/15	10/15
Sonnet	6/15	30/15
Opus	10/15	50/15

Cache reads are free. Non-Claude models (Gemini, Codex, etc.) don't consume Claude credits.

Plan budgets

Plan	$/month	Credits/5h	Credits/week
Pro	$20	550,000	5,000,000
5×	$100	3,300,000	41,666,700
20×	$200	11,000,000	83,333,300

Requirements

• Python 3.9+
• OpenClaw with session JSONL files (auto-detected at ~/.openclaw/agents/main/sessions/)