token-usage-tracker

Quick start

1. Configure defaults in skill-config.json (timezone, log_folder).
2. Install scripts (examples provided) into your workspace and wire the interceptor into your message pipeline.
3. Use scripts/context_summarizer.py before sending large contexts to reduce token usage.

What this skill provides

• Logging: token_tracker.py writes per-call token usage to a JSONL log. Includes timestamp normalization.
• Interceptor: token_interceptor.py example that normalizes timestamps and forwards sanitized messages to the tracker.
• Alerts: token_alerts.py example for threshold-based alerts (no external posting by default).
• Compression: context_summarizer.py produces short summaries to reduce token payloads.
• Utilities: migration and cleanup scripts (convert timestamps, dedupe log entries).

When to use

• Use this skill when you want transparent per-call token accounting, to keep token usage low, or to protect sensitive/verbose contexts by summarizing before sending to the model.

Files

• scripts/
  • token_interceptor.py — example interceptor (normalizes timestamps)
  • token_tracker.py — logging helper
  • token_alerts.py — alert examples
  • context_summarizer.py — compression helper
  • migrate_timestamps.py — migration utility
  • dedupe_log.py — dedupe utility
• references/
  • examples/systemd/ — example unit files (install manually)
• skill-config.json — configurable defaults
• README.md — usage and install notes

Configuration

See skill-config.json for defaults. The skill exposes:

• timezone: default UTC
• log_folder: default ./skills/logs (relative to OpenClaw workspace)
• compression settings: summary_target_tokens, max_context_tokens

Security and installation

• The scripts are examples and safe by default. They do not change system state or install services automatically.
• Example systemd/unit files are provided in references/systemd/ — apply them manually after review.

License: MIT (adapt as you prefer)