#29Basics & Safety
Token-lean workflow: BRIEF.md, HANDOFF.md, and reuse before research
Keep the working chat short, store confirmed facts and the next action in files, search saved results before re-researching, and stop a failing operation after two attempts. Fresh research only starts when no saved result answers the question, a source changed, or a fact expired.
Why this scoreSensible cost technique, but the headline savings and the claimed official source are unverified.
"Create BRIEF.md, HANDOFF.md, sources/, results/, and a run log. Keep the working context short. Search saved results first... Give each task one owner. Save progress and stop the same failing operation after 2 attempts. Check every result against the brief."
Reported results
Post claims 100,000 to 12,500 tokens per task; not verified.
Verification
The post says xAI documented this; no matching official page was found this run. Shared /workspace files across bots are documented. Token figures are unverified; treat the method as a reasonable practice, not a measured result.