Amazon just told its devs to stop tokenmaxxing. I learned that lesson with my own bots. Here’s what keeps my token bill sane.
Review before you run anything: read the docs yourself, back up your config, and try it somewhere you can afford to break first.
1. One bot, one job
A general helper drags every tool and every bit of context into every task. A bot with one job carries only what it needs.
2. Keep relays short
Bot-to-bot messages say what to do and where the files are, not the whole backstory. Results come back as IDs and one-liners.
3. Write a batch log and read it instead of re-researching
Each run saves a small summary file (what it made, what it skipped, what’s next). The next run reads that instead of rescanning everything.
4. Filter before you read
Pull RSS feeds and filter by timestamp first, then fetch only the articles you’ll actually use, and only their text.
5. Cap the output
Set a hard number of posts or tasks per run. If fresh news is thin, write fewer. Never pad.
6. Skip pointless checks
Don’t re-verify things a script already verified. Check once, record it, move on.
God Bless, and May the Source Be with You!
