| name | openclaw-usage-trends |
| description | Collect local Codex quota snapshots and OpenClaw cron model health into a private SQLite history, then report consumption trends and rolling-window exhaustion forecasts. Use when the user asks for agent usage trends, quota history, token budget forecasting, OpenClaw cron model allocation, or scheduled low-cost usage monitoring. |
OpenClaw Usage Trends
Use the bundled script. It calls only the existing read-only Codex quota
endpoint and the local OpenClaw CLI. It never reads credentials or estimates
unavailable provider token usage.
Collect A Snapshot
python <skill-dir>/scripts/usage_trends.py collect
The default database is ~/.local/share/agent-quota-query/usage-trends.sqlite3.
Set AGENT_QUOTA_QUERY_DATA_DIR to keep the private history elsewhere.
Read Trends And Forecasts
python <skill-dir>/scripts/usage_trends.py report
python <skill-dir>/scripts/usage_trends.py report --days 30 --json
Explain that a forecast requires at least two samples in the same rolling reset
window. It projects the observed percentage slope, not an exact token cost.
Scheduling
Use an OpenClaw command cron for collection. Do not spend an LLM call on this
rule-based operation. Keep delivery disabled unless the user explicitly wants a
separate alert workflow.
Safety
- Report provider/model names, percentages, reset times, and cron counts only.
- Never display API keys, auth profiles, task prompts, recipient IDs, or raw
OpenClaw configuration.
- Distinguish actual Codex quota observations from Kimi/MiniMax health signals;
the latter do not prove token consumption without a provider usage API.
