XSource
The rule that makes my agent fleet trustworthy
by Randy Northrup (@starsoldier420) · 2026-10-07
✓ Opened via the X API on 2026-10-15
Read the original ↗Part 1 · What the source did
A fleet rule: whichever model wrote a piece of work never reviews it; a different vendor's agents review every lane, and repeated failure forces a redesign.
- Muse and Codex agents build; Grok agents review every lane.
- A third failed review triggers a redesign instead of another patch.
“the model that writes a lane never reviews it”— Randy Northrup (@starsoldier420), X
Summary written by Agent Cookbook. See the original for the full context.
Part 2 · Our copy-paste version
Adapted by us from what the source described. Fill in the [BRACKETS].
The prompt
REVIEWER (different model than the author):
You did not write this. Review lane [NAME] against its spec [FILE].
Return PASS or FAIL with numbered reasons. This is review #[N].
If N = 3 and still FAIL, don't suggest patches: write a one-paragraph redesign proposal instead.
🛡
Approval ruleNothing merges without a PASS from a non-author model and your sign-off.
Set it up
- Label each work item with its author model.
- Route reviews to a different vendor.
- Track the review count per item.
Common failures
⚠
What usually goes wrong
- Same-model reviewers rubber-stamp shared blind spots.
- Endless patch loops without the 3-strike rule.
Variations
- Use Grok Bot as reviewer for Claude Code work.
- Apply to docs and specs, not just code.