Feed
What gets in, and what stays.
Five kinds of knowledge, one setting that decides whether a capture waits for a person, and two queues that tell you what to take out. A rule a person approved is never removed by a timer.
The five kinds
Everything in the brain is one of five things. The kind is not filing: it decides how a capture is treated on the way in, and a memory is held for a person in two of the three settings below precisely because it is the kind least likely to be about code.
| Kind | What it is | For example |
|---|---|---|
| Convention | A standard your team holds to | "Database columns are snake_case, including in migrations." |
| Pattern | A shape that recurs, and where to copy it from | "Every endpoint validates in middleware, never in the handler." |
| Gotcha | A trap that cost somebody an afternoon | "The deploy script needs the VPN up or it fails at the last step." |
| Skill | A procedure, in order | "How to stand up a new service from the template." |
| Memory | A durable fact that is not about code | "We tried GraphQL in 2024 and dropped it, for these three reasons." |
They live in one filterable list on Knowledge, filtered by kind, by Space, or by whether anything has used them. There is no separate page per kind.
Always-on is not a sixth kind
A rule can be pinned, which means it is sent every session regardless of whether anything in that session looked relevant. Pinning is a property of a rule, not a kind of rule, and it is the one setting on this page that can quietly do nothing.
What decides whether a capture waits for you
When an agent records something, or a connector distils something out of a merged pull request, one workspace setting decides whether it goes into the brain or into a queue. It is the same setting for both, which is the point: a fact your agent noticed and the same fact written in Slack should not be governed differently.
| Setting | What happens to a capture | Who it suits |
|---|---|---|
| Manual | Everything waits for a person. | Teams with a compliance reason to have a human on every rule. |
| Hybrid | Routine captures go in provisionally. Team memories and contradictions wait. | The default, and right for almost everyone. |
| Auto | Everything from a named colleague goes in. Contradictions still wait. | Teams who would rather archive a bad rule than clear a queue. |
Hybrid is the default, and it is worth knowing that before you go looking for an empty queue and conclude nothing is working. A second setting sits on top of it, require review for, which can hold everything or hold just the risky kinds no matter what the first one says. Where the two disagree, the stricter one wins.
Two things are held in every setting, including Auto
- Anything that contradicts a live rule. Overwriting a rule a colleague wrote, without telling anyone, is the single thing that would make the brain not worth consulting. So a contradiction is never resolved automatically, in any setting. It arrives as a conflict for a person to settle.
- Anything Harbor cannot place inside your company. An outside contributor on a public repo, a guest in a Slack Connect channel, a customer filing a ticket. “Auto” is a statement about trusting your own team’s captures and was never a statement about trusting strangers. A private repo is treated as trusted even when the author is just a handle, because an admin decided who can push to it. A public one is not.
Provisional, and what earns permanence
A capture that goes in without a person does not go in on the same terms as one you approved. It enters provisionally, which means it is served to agents like any other rule, and it has to earn its place. Reinforcement is what earns it: the same thing being observed again, from somewhere else.
A provisional rule that goes a fortnight with no reinforcement is archived on the next weekly pass. Archived means dropped from what agents are served, not deleted: the body is kept, so a rule that decayed out is still readable and can be brought back.
The Review queue
Review is where held captures wait. Under the default setting that means team memories, contradictions, and anything whose author could not be placed inside the company, rather than everything. Each row says where it came from and, for a connector row, links back to the message or pull request it was distilled from, so you can judge the extraction against the source instead of on its own wording.
Rows arrive from more places than the two the old guide named: your agents’ own captures, every connector by name, pull request reviews, tasks, chat, and documents you uploaded. You can filter by any of them, which is how you approve six proposals from one PR review without reading past the Slack ones.
For each row you can approve it, edit it and then approve, change its Space first, or dismiss it. Editing before approving is worth the ten seconds: a distilled proposal is a machine’s reading of someone else’s prose, and it is usually right about the fact and clumsy about the sentence.
Seeing Review requires knowledge.publish, which owners, admins and curators hold. A member can propose but not approve, which is the whole distinction between the two roles. See who sees what, and who decides.
Putting documents in by hand
You do not have to wait for a connector. Upload on the Knowledge page and Harbor splits the document into candidate rules for you to judge, rather than adding the document itself: the useful unit is a rule an agent can be served, not a twelve-page file it has to read.
Markdown, plain text, CSV, PDF and Word documents, up to 10MB each. .mdc is accepted too, so a Cursor rules file can go straight in.
What leaves the brain
A brain that only grows is a brain that stops being read. Harbor keeps two lists of candidates for removal, and it keeps them separate on purpose, because they are two different problems and merging them hides the smaller one.
- Nobody has used this. Rules that were served and never cited, or never served at all, ranked by how long that has been true. These are the ones to archive.
- This is used, and nobody has re-checked it. Rules an agent cites regularly that no human has looked at since they landed. These are the dangerous ones, because a stale rule that gets cited is worse than one that gets ignored.
Each row carries the sentence Harbor is willing to defend about it, and there are three possible verdicts rather than two. A rule was served, or the instrument was watching for that rule’s whole life and never saw it serve, or it has not been measured. The third is not a fact about the rule. It means the rule predates the measurement, and Harbor prints the date it started watching alongside the verdict so you can see the edge of what it knows. A tool that showed you “0 uses” for all three would be telling you something it cannot know.
Acting on either list is a person clicking archive. Harbor is the instrument, not the editor. What it owes you is a defensible list and the evidence behind each row, which is the subject of served versus cited.
Next, who all of this is served to: who sees what, and who decides.