Drift tooling: what exists, and what honestly doesn't
Only one of this repo's six doc mirrors has any drift-detection tooling at all: the Cloudflare mirror (docs/docs/cloudflare/), via scripts/cf-mirror-sync.py.
What it does
scripts/cf-mirror-sync.py --generate walks the mirror and computes a sha256:16 content hash per file, writing the result to docs/docs/cloudflare/HASHES.json (4,393 entries) -- the same local-integrity philosophy as workers/subagentcowork/scripts/gen-hashes.py, scaled from an ~80-entry TypeScript array to a JSON manifest sized for thousands of files. --diff recomputes and compares against that manifest to report which local files have changed since it was generated. This is pure local hashing -- zero network calls.
What it honestly doesn't do yet
This only answers "has my local copy changed since I last snapshotted it," not "is my local copy stale relative to the live site." The actual live-compare step -- a cheap HEAD/ETag "sha256 bloom crawl" that only re-fetches pages that have actually changed on developers.cloudflare.com -- is documented as a next step in the script's own docstring, specifically to avoid the token cost of re-crawling ~4,500 pages one at a time. It has not been built.
Every other mirror on this site (platform.claude.com, code.claude.com, claude.com, blog.cloudflare.com) has no drift tooling of any kind -- freshness for those is simply "whatever it was when the mirror was originally created," occasionally spot-checked by a live web_fetch during unrelated research (e.g. the 2026-07-01 Fable 5 model-overview verification), never systematically re-verified.
See also: mirror vs viewer, and subagentcowork.com's own sha256:16 drift-check table for the sibling pattern this script is modeled on.