Docs rot silently: a runbook says auto-deploy is on when it is not, names a command that no longer exists, or labels a repo that was renamed — and the next person (or agent) acts on the stale claim. Each iteration takes one unverified operational claim from the audit ledger, tests it against reality (run the documented command, check the referenced system or setting), and on mismatch fixes the doc — or the system, if the doc described the intended state. Distinct from changelog-sync-after-ship (adds entries for new work): this verifies EXISTING claims against the live system.
Loop
- Goal
- every operational claim in the in-scope docs is verified against the live system — confirmed accurate, or fixed where it had drifted
- Exit when
- DRIFT.md lists every operational claim in scope and zero remain unchecked — each verified accurate or corrected
Check between iterations
test -f DRIFT.md && grep -c '^- \[ \]' DRIFT.md || echo NO_AUDIT
$
Kickoff prompt
Start the "Docs Drift Sweep" loop.
Goal: every operational claim in the in-scope docs is verified against the live system — confirmed accurate, or fixed where it had drifted
Max iterations: 8
Between iterations run: test -f DRIFT.md && grep -c '^- \[ \]' DRIFT.md || echo NO_AUDIT
Exit when: DRIFT.md lists every operational claim in scope and zero remain unchecked — each verified accurate or corrected
Step 1: Write DRIFT.md listing each operational claim from the in-scope docs as an unchecked box, then verify the first one by running the documented command or inspecting the referenced system.
›