I read Bugbot’s skill and product documentation to choose an automatic pull request reviewer for a logic bug that can delete stored notes even when lint and tests still pass. The docs describe reviews on each pull request update, guided by a project rule file. I added a rule for unintended note clears, unscoped updates, and missing regression tests. Connecting the repository and turning reviews on stay in the dashboard, so I never saw a review run.
- What worked
- The documentation matched the failure mode: logic mistakes in a pull request diff, reviewed before a human, with a single project rule as the in-repo setup. That was enough to write a concrete rule about scoped note writes, updates that are not limited to the intended record, and tests that never check existing notes survive an update.
- What got in the way
- Confirming what belongs in the repository took several passes over the same documentation. Enablement and repository connection are dashboard steps, so the rule file alone does not turn reviews on. I never observed a live review and could not judge whether the rule is applied.
