Onur Polat

Building 7+ products with AI + automation
Sharing what works (and what doesn't)
Claude Code power user | n8n addict

[Onur Polat @onurpolat05]

Stop using Claude Code just for coding.

That's like buying a smartphone only to make calls.

I turned it into my personal operating system:

→ CLAUDE(.)md as the agent's constitution
→ Skills over MCP (saved 80k+ tokens)
→ Hooks for deterministic control
→ Subagents for delegated tasks

The result? An executive assistant that manages my Trello, plans my projects, and actually remembers context.

Full breakdown of my opAgent system:
https://t.co/1yQOOc959h

[Onur Polat @onurpolat05]

@egebese Tebrikler

[Daron Yondem @daronyondem]

Hala "100% AI yakaladim" kafasinda olanlar var 😂

Benim yazdigim yazilarin, hele uzun form yazilarin 99%'u 100% AI 😂 Artik oturup da yazi yazacak hamaliyeye giremem. Audio kayit alip, uzerine AI'a yazdiriyorum. En son, belki 3-5 kelime degistiriyorumdur hosuma gitmeyen yerlerde. Herkese de tavsiye ederim 😊

Kisa zamanda da bitmeyecektir bu "Aha 100% AI" diye ortalikta gezenler. At arabasi ile takilsinlar, ne diyim 😂

Bu arada, benim yazilari da AI'a ozetlettirin, sesli okutun.. isinize nasil geliyorsa ozgursunuz 🤣

[Burak Can Polat @devburakcan]

https://t.co/dB1OECHmta

[Onur Polat @onurpolat05]

This is gold. Two things I'd add from running skills in production daily:

1. Progressive disclosure for memory — don't load everything into CLAUDE.md. Lazy-load from markdown files. Cuts token waste significantly.
2. Self-improvement hooks — PostToolUse that auto-logs errors into a gotchas file. Next session, Claude reads its own mistakes. Compounds fast.

Packaged these patterns into an open-source template: https://t.co/RG6KFfXhsf (9 skills, 6 hooks, interactive setup)

[Onur Polat @onurpolat05]

what it does:

/start → loads your priorities from yesterday
/end → saves progress, records learnings

next day /start → picks up exactly where you left off

zero dependencies. pure markdown. git-tracked.

[Onur Polat @onurpolat05]

every Claude Code session starts from zero.

no memory. no context. same mistakes.

I got tired of explaining myself every morning.

so I built a system that gives it persistent memory.

11 months of daily use. now it's open source.

[Onur Polat @onurpolat05]

it's called ALBA

https://t.co/QePXXOJdcQ

template repo — click "Use this template" and answer 7 questions. agent ready in 10 min.

9 skills, 6 hooks, 3-tier memory. MIT license.

[Onur Polat @onurpolat05]

@marcelfolaron YAML frontmatter keeping diffs clean is smart. solid setup.
same core pattern, different contexts - you're coordinating devs, i'm coordinating AI agents. file-based memory works for both.

[Onur Polat @onurpolat05]

"How do you manage 7 projects at once?"

I don't. I built a system that does.

Claude Code = my dev team

n8n = my operations team

AI agents = my project managers

Total employees: 0

Total projects shipping: 7

Solopreneurship in 2026 looks nothing like 2023.

[Onur Polat @onurpolat05]

@marcelfolaron that's a solid pattern. never thought about branch-level versioning tbh.
i keep mine project-scoped instead of branch-scoped - one https://t.co/UYAV2Hw8SB per project, one dashboard for all.
curious: do you ever hit merge conflicts on context files when branches converge?

[Onur Polat @onurpolat05]

@marcelfolaron the key insight: don't try to keep context in your head OR in chat history.
make it file-based. a simple markdown file per project beats any fancy tool.
my /start command reads a central dashboard + project context = instant "where was i?"

[Onur Polat @onurpolat05]

the customizability is what makes cc feel like a different thing entirely. i turned it into a personal OS - custom agents, persistent memory, hooks, skills for content/email/research, MCP integrations... manages my 10+ projects now, not even for coding. open sourcing
it, haven't pushed latest but core idea is here: https://t.co/XTQ0VeLj6p

curious if you use cc for non-coding stuff yourself?

[Onur Polat @onurpolat05]

Congrats Felix! 👋 I've been tackling the exact same 'knowledge work' continuity challenge using Claude Code.

I built a system that architects a persistent file-based memory (state/knowledge/daily logs). It allows the agent to maintain context across sessions without hitting token limits effectively.

Would love to hear how Cowork handles long-term memory! https://t.co/XTQ0VeLj6p

[Onur Polat @onurpolat05]

Love seeing the expansion to non-coding tasks! I actually used Claude Code to bridge that exact gap.

I built a starter kit that adapts the agent to ANY role (PMs, Designers, Devs). It asks 5-7 questions about your workflow and generates a custom memory system & toolset automatically.

The self-building capability is wild. 👇 https://t.co/XTQ0VeLj6p

[Onur Polat @onurpolat05]

@acerionsjournal Claude code ile kullanım için ben de şöyle bir şey yaptım müsaitliğinizde incelersiniz. https://t.co/xGV9t0DLTs

[Matt Shumer @mattshumer_]

Claude Code Addiction is real

[Onur Polat @onurpolat05]

Built opAgent — my executive assistant setup. Non-coding use cases:

→ Research synthesis: "Summarize everything I know about [client]" — pulls from my notes + memory
→ Meeting prep: Generates briefing docs by connecting Trello cards + past conversations
→ Content repurposing: Takes one idea, outputs LinkedIn post + tweet thread + newsletter draft

The to-do list skepticism is valid IF you just ask it to "manage tasks." The unlock is making it the orchestration layer, not the task list itself.

Wrote about the full architecture: https://t.co/1yQOOc959h

[Onur Polat @onurpolat05]

Yes! Building exactly this. Claude Code + Obsidian as my PKM backbone. Custom skills, hooks, and agents tied together. It now manages my automation consulting workflows — connects notes, tasks, and client projects automatically.
Documented the full setup here: https://t.co/1yQOOc959h