End your session with "handoff".
Tomorrow any agent picks up where you stopped. Out sick? Hand it to a teammate instead, they get it in their next session.
Harbor Handoffs. Works in Claude Code, Codex, Cursor and Copilot.
gethrbr.com
Harbor retweeted
Everyone is building a software factory now.
Agents in. Pull requests out.
The first thing every factory needs is memory.
So teams build one. mem0, a vector DB, a big CLAUDE.md.
That part takes a weekend.
Then the real questions show up:
- Who edits it?
- Who is responsible when a rule is wrong?
- Who checks it actually makes the agents better?
- Which Slack thread changed this rule?
- Two rules disagree. Which one wins?
- Why does Claude Code know it and Cursor doesn't?
- What stops an agent from running the command we banned?
None of that is storage.
It's upkeep. Every day, for every agent, on every repo.
A software factory is a good idea.
But a factory builds the product, not its own machines.
Pick what you build.
Something weird is happening to teams that code with AI.
The agent writes the code in 4 minutes.
The pull request merges 3 days later.
It feels fast. It isn't.
gethrbr.com
4
6
91
Harbor retweeted
Agent Memory: Expectation vs Reality
Expectation: "Let's add a CLAUDE.md.
Reality: a pandora's box of 20+ other components opens:
- Rules files per tool
- Folder-scoped rules
- Rules with the why
- Rule ownership
- Capturing corrections
- Deduping rules
- Conflicting rules
- Stale rules
- Revoking wrong rules
- Recall per task
- Ranking what's relevant
- Context budget
- Cross-tool sync
- Team sharing
- Private vs shared facts
- PII masking
- Guarded actions
- Session handoffs
- PR review against rules
- Onboarding new devs
- Measuring which rules fire
- Audit trail
AI made writing code easier, which made knowing what NOT to do even more valuable.
But one file ≠ memory
Of course, you can start with one CLAUDE.md, but these layers come up the moment you ask:
- How does my teammate's agent learn what mine learned?
- Which of these 80 rules still matter?
- Why did the agent ignore the rule on line 140?
- How do I stop the agent before it runs a migration on prod?
There is a systematic way to build all of this out
gethrbr.com/blog I put tons of useful data for you in here!
Harbor retweeted
Just launched a new free skill: rules-score.
It grades your repo's agent rules (the CLAUDE.md / AGENTS.md / Cursor rules your coding agent reads before every task) out of 100, and shows exactly which lines cost points and how to fix them.
To test it I ran it on 20 big open source repos:
sentry 89. next.js 83. vscode 52. bun 39. cal.com 35.
It scores 5 things:
1. accurate: every path and command it names really exists
2. says how to install, run, test and lint
3. specific rules, not "write clean code"
4. small enough to load every task
5. no noise
Most common fail: rules pointing at folders that don't exist anymore.
Try it on your repo (Claude Code, Codex, Cursor):
npx skills add gethrbr/deckhand
then ask your agent: "score our agent rules"
github.com/gethrbr/deckhand
Harbor retweeted
Just open-sourced 3 free skills for AI coding agents
- audit-memory: finds outdated, conflicting or incorrect instructions across your agent rules and memory files
- check-diff: checks your code changes against your repo's rules
- rules-from-reviews: turns recurring PR feedback into rules your agents can follow
Works with Claude Code, Cursor, Codex and more.
No signup, no API key, no Harbor account needed.
Would love some feedback (and a ⭐ if you find them useful)
github.com/gethrbr/agent-ski…
Introducing the Harbor GitHub App.
It reviews your PRs against CLAUDE.md, AGENTS.md, your team's rules, and the shared knowledge in your Harbor brain.
Your agents learn from your team's work. Now Harbor makes sure your code follows it.
gethrbr.com
4
8
62
Every engineer on your team has a coding agent. Each one learns alone.
Fix it once, in any agent, and the rest know before they write.
gethrbr.com
3
2
65
🎉🎉🎉
Harbor is live on Slack now 🎉
Your agent is about to turn AI on for every customer. Your DPA says they opt in first.
Does your agent know?
/harbor asks your team's brain from Slack, with sources.
/harbor remember reaches every agent on the team.
gethrbr.com
20
Harbor retweeted
Every engineer on your team has a coding agent. Each one learns alone.
Dana's agent works something out on Monday. The session ends, and it's gone. Tuesday, Sam's agent makes the same mistake. Someone adds it to a rules file nobody keeps up to date. Your best engineer explains it again in every review.
Harbor is one brain for every agent on your team. Fix it once, in any agent, and the rest know before they write. New session, new hire, new agent: still there.
Works with Claude Code, Codex, Cursor and GitHub Copilot, with more coming.
What's the fix you've explained to your team's agents the most times?
Harbor retweeted
My CLAUDE.md hit 5,000 lines.
And my agents were still making the same mistakes.
After 300+ Claude Code sessions in 30 days, I did what every engineer does:
Every time an agent broke a test, hit a CORS edge case, or ran npm instead of pnpm, I added another rule.
By week 3, the whole setup was breaking down:
• My agents kept repeating mistakes I'd already documented twice.
• Prompts kept loading backend rules even when I was editing frontend code.
• Switching from Claude Code to Cursor or Codex meant starting from scratch.
That's when it clicked: engineering memory isn't a flat text file. It's a graph.
So we built Harbor.
This is our actual knowledge graph, built automatically from our own coding sessions, PRs, and Slack:
• 835 rules and gotchas
• 1,120 links between them
• 69 clusters across CLI, backend, and frontend
When I open Claude Code or Cursor in packages/cli, Harbor doesn't load all 835 rules. It surfaces only the 4 nodes relevant to that task.
When one agent hits a trap, Harbor helps other agents avoid making the same mistake
We're opening Harbor to a small group of solo devs and engineering teams this week.
Want to see what your repo's graph looks like? DM me.
Harbor retweeted
You've told Claude "use pnpm, not npm" 14 times. It said "You're absolutely right!" 14 times.
With Harbor you say it once. It becomes a rule, and the answers that used it end with harbor-applied: #1, so you can count it working.
Get early access at gethrbr.com
Harbor retweeted
Hello @X algorithm 👋
Please connect me with builders who use Claude Code, Codex, and Cursor every day.
Your AI coding agent keeps making the same mistakes.
You fix something in a PR. Explain a rule. Update your workflow.
A few sessions later, it's back to doing things the old way.
I'm building Harbor to fix this.
It learns your team's rules and context from PRs, Slack, docs, and agent sessions, so Claude Code, Codex, and Cursor can carry that knowledge forward.
🎬 30-sec demo: gethrbr.com/media/harbor-30-…
I'm opening early access and looking for builders to try Harbor on real projects.
Join the waitlist 👇
gethrbr.com
#ClaudeCode #Codex #BuildInPublic
Harbor retweeted
Ask your AI agent why your team stopped using that library.
It doesn't know. Neither does your new hire.
The decision was made in a Slack thread in March. The rules file only has what someone remembered to write down. So the agent reaches for it again, and you leave the same review comment for the third time.
The new hire can at least ask the two people who've been there longest. The agent can't ask anyone.
Every company needs a brain for its people and its AI agents.
Decide something once and it reaches everyone.
Your new hire can look it up, and your Claude Code and your teammate's Codex both know it next session.







