Drift Check
Verifies a project's reference data — prices, models, dependencies — against upstream sources and flags staleness.
Entry verified April 21, 2026
The short answer
Reference data rots. Model pricing changes, products get deprecated, APIs move. This skill checks each configured data file against its upstream source and flags drift before it shows up in user-facing pages. Can run as a scheduled GitHub Action so drift is caught before the next content push, not after.
When to use it
Weekly as a scheduled run. Before shipping a page that depends on pricing or model data, where one stale value undermines trust.
Setup
- 1
Save the file as ~/.claude/skills/drift-check.md.
- 2
Configure source pairs per data file: the local file, the upstream URL, and the fields to compare.
- 3
Can run locally on demand or as a GitHub Action on a cron schedule.
- 4
Upstream scraping breaks when sites restructure — build a fallback path for HTML parse failures rather than silently skipping.
Example
You: /drift-check data/claude-models.json Claude: 2 findings. Opus 4.7 price matches upstream ✓. Haiku 3 retiresOn=2026-04-19 is past (today 2026-04-20) — move to status=retired.
Source & attribution
- Author
- Bryan Collins (PWS pattern)
- Licence
- MIT
- Type
- Original
Original pattern published under MIT — attribution preserved by convention, not licence requirement.
Caveats
Upstream scraping breaks when sites restructure. Build fallback for HTML parse failures.
56 skills across 10 categories, all licence-verified.