Is anything known about the code I'm about to change?
When a task starts, your agent checks the backlog for pins on the files it's about to touch. Known bugs and gotchas change the plan before any code does.
Pinning things is only half of it. The other half is that they come back when they matter, and the moment they matter most is right before someone changes that code.
One check, at the start
When an agent starts a task, it calls pins_near once, on the files and folders the task involves. Not before every edit, just at the start, while what comes back can still change the plan.
What comes back is ranked by how close each pin is to what you're touching: the same file first, then the same folder, then further out. A pin that points at a specific function or a few lines beats one that just lists the file among a dozen others.
What the agent does with it
Known bugs and gotchas change the plan. If there's a pin saying the retry logic swallows 409s, an agent about to touch webhooks should know that before it builds on top of it. At the very least it tells you.
A claimed pin means someone's already in there. If another person's agent has claimed a pin in the same file, that's work in flight. Better to know now than at merge time.
A pin this change fixes gets linked. If the task happens to fix a known pin, the agent links its PR to it as fixed_in when the PR opens. Fixing a nearby pin that isn't part of the task is your call, not the agent's.
The same check, before a PR
The Claude Code plugin has a nearby skill that runs the same check over everything the branch changed, just before you open the PR. It's a good last look: anything known about the code you touched, anything you might have just fixed without meaning to.
You can also just ask, any time: "is anything known about the code I'm about to change?"