Script-first OpenClaw skill for ChatGPT Pro usage reporting (5h/Week).
Before:
- docs-only skill
- cron prompt asked the model to improvise a report
- source routing drifted into
session_status, browser attempts, or generic reasoning
Now:
- one updater:
scripts/update_source.py - one reporter:
scripts/report.py - one explicit source snapshot schema
- fail-closed behavior (
NO_REPLYif no source snapshot) - forbidden fallback to
session_status
SKILL.mdcron-job.example.jsonscripts/report.pystate/source.schema.example.jsonassets/usage-report-example.jpg
The model must not invent the report. It must run the script and return exact stdout.
