merge-and-throughput.md

Merge and Throughput Philosophy

Agent throughput can exceed human attention; this changes which costs matter most. The goal is fast iteration without letting drift compound.

Default PR shape

• Small, focused diffs
• Clear reproduction/validation steps
• Links to the source-of-truth doc/spec that motivated the change

Minimal safety bar (don’t skip)

• npm run lint
• npm run build
• If gateway protocol/client changes: npm run probe

When follow-ups are better than blocking

Prefer merging + follow-up PRs when:

• The change is localized and the failure mode is low blast radius
• The harness can detect drift quickly (lint/build/probe/manual smoke check)
• The alternative is waiting on uncertain investigation that can be done after merge

Do not treat this as permission to merge broken builds or skip boundary validation.

Config change throughput (OpenClaw)

Config-driven systems have a different throughput dynamic than code:

• Prefer small, focused config patches over large multi-key edits. A single agent's tool policy change is safer than rewriting the entire agents block.
• Run openclaw doctor after every config change — it's fast and catches most regressions.
• OpenClaw creates dated .bak files automatically. Verify the backup exists before making the next change.
• Cron job edits are higher risk: the runtime writes state back to cron/jobs.json periodically. After editing, verify changes persist after the next runtime state save.