Grok Bot saved skills can go stale: a pre-run checklist
Citing xAI docs, the author warns that Grok Bot's saved skills can outlive pricing and policy, and offers a five-step pre-run check.
Grok Bot can execute yesterday’s rule perfectly and still damage today’s business xAI’s docs warn that memory cannot replace authoritative sources. > A saved skill can outlive the pricing, approval policy or data format it was built around. > A routine can then repeat that mismatch on schedule. Before the next run: 1. Name the source of truth and its owner. 2. Compare its date, version and fields with the saved workflow. 3. Flag contradictions before changing records or sending messages. 4. Correct the stale assumption, test one safe example and verify the result. 5. Keep approval rules in force when updating instructions. Example: a sales Bot remembers an old discount. Ask it to show the pricing policy, quote the conflicting rule and draft the correction for review. Knight Capital shows the stakes: a 2012 deployment error activated old code, sent over 4 million orders in 45 minutes and ultimately caused losses above $460 million. Ask for a conflict report before an action report.
This post quotes another post: “Grok Bot handoff tax: why one owner beats many parallel agents”