Zum Hauptinhalt springen
tsecurity.de LIVE
Echtzeit-Radar & Feeds
Alle RSS Feeds
👥 Community & Social
Web Security TippsNew manual calculation setting in Google Sheets(21.09.2026 um 20:54 Uhr)
Videos & KonferenzenTechquickie: The Steam Frame Shouldn't Work - Here's Why It Does(21.09.2026 um 21:17 Uhr)
Sichere ProgrammierungHow to Build a Production-Ready iOS App With AI-Generated Code(21.09.2026 um 21:00 Uhr)
Sichere ProgrammierungAfriex Integrations: Sandbox, Idempotency, and Webhook Simulation(21.09.2026 um 21:50 Uhr)
Sichere ProgrammierungBridging Local and Cloud Databases for Centralized Data Management(21.09.2026 um 21:51 Uhr)
Web Security TippsNew manual calculation setting in Google Sheets(21.09.2026 um 20:54 Uhr)
Videos & KonferenzenTechquickie: The Steam Frame Shouldn't Work - Here's Why It Does(21.09.2026 um 21:17 Uhr)
Sichere ProgrammierungHow to Build a Production-Ready iOS App With AI-Generated Code(21.09.2026 um 21:00 Uhr)
Sichere ProgrammierungAfriex Integrations: Sandbox, Idempotency, and Webhook Simulation(21.09.2026 um 21:50 Uhr)
Sichere ProgrammierungBridging Local and Cloud Databases for Centralized Data Management(21.09.2026 um 21:51 Uhr)
Intelligence View
⚡ tsecurity.de Intelligence

How to Run 5 Claude Code Agents in Parallel Without Them Overwriting Each Other

I was about an hour into what felt like a genuinely productive session. Three Claude Code instances running, each working on a different part of the codebase. It felt like having a small team. Then I checked the git diff. Two of the…

0
↗ Quelle (dev.to)
Reagiere als Erste:r — dein Feedback zählt!

I was about an hour into what felt like a genuinely productive session. Three Claude Code instances running, each working on a different part of the codebase. It felt like having a small team.



Then I checked the git diff.



Two of the agents had been editing auth.js at the same time. One of them finished first. The other one saved over it. Two hours of work from agent one — gone. No warning, no conflict marker, nothing. Just silently overwritten.



That's the thing nobody tells you about running parallel agents. Claude Code is incredible. Running multiple instances is even better. But they have absolutely no idea what each other is doing. There's no shared brain, no "hey, I'm working on that file" signal. Each one just does its thing and assumes it has the whole codebase to itself.



So I built Switchman to fix it. This is how to use it.









Before you start



You'll need:




  • Node.js 22.5+

  • Git 2.5+

  • Claude Code



Install Switchman:




npm install -g switchman-dev












The idea in plain English



Switchman does three things that Claude Code doesn't do on its own:



It gives your agents a shared task list. You add the work upfront, and each agent picks up one task at a time. No two agents ever get the same job.



It adds file locking. Before an agent touches a file, it checks in. If another agent already has that file, it gets told immediately and picks something else instead. The collision never happens.



It connects to Claude Code automatically via MCP. You don't write any coordination logic. You just open Claude Code in each workspace and it handles the rest.









Step 1 — Create your agent workspaces



The first thing you need is a separate workspace for each agent — its own copy of the repo, on its own branch. If two agents are working in the same folder, they'll step on each other no matter what.



Run this once from your project:




cd my-project
switchman setup --agents 3






That's it. Switchman creates three isolated workspaces and sets everything up:




✓ Switchman ready — 3 agent workspaces created

✓ /Users/you/my-project-agent1
branch: switchman/agent1

✓ /Users/you/my-project-agent2
branch: switchman/agent2

✓ /Users/you/my-project-agent3
branch: switchman/agent3






Each workspace is on its own branch. Whatever one agent does stays completely separate until you decide to merge it.









Step 2 — Tell Claude Code about Switchman



Open ~/.claude/claude_desktop_config.json and add this:




{
"mcpServers": {
"switchman": {
"command": "switchman-mcp",
"args": []
}
}
}






Restart Claude Code. Your agents can now use Switchman's coordination tools automatically — they don't need to be told how, that comes in the next step.









Step 3 — Add CLAUDE.md to your repo root



This is the file that tells your agents what to do. Without it, they won't know to coordinate.




curl -O https://raw.githubusercontent.com/switchman-dev/switchman/main/CLAUDE.md






It instructs each agent to check in with Switchman at the start of every session, claim files before editing them, and release everything when they're done. You write this once and never think about it again.









Step 4 — Add your tasks



Think about what you want to get done and break it into separate chunks. The more independent each task is, the better parallel agents work.




switchman task add "Implement OAuth login flow" --priority 9
switchman task add "Add rate limiting to API routes" --priority 7
switchman task add "Write tests for auth middleware" --priority 6
switchman task add "Update API documentation" --priority 3






Higher priority tasks get picked up first. Each agent gets one task at a time — when it finishes, it picks up the next one.









Step 5 — Open Claude Code in each workspace



Open a separate Claude Code window in each of the three folders that switchman setup created. Each agent will pick up a task, lock the files it needs, do the work, and release everything when it's done.



You just watch it happen.



Here's what it looks like when two agents try to grab the same file:




# Agent 1 locks auth.js
✓ Claimed src/middleware/auth.js

# Agent 2 tries to claim the same file
⚠ Conflict — auth.js is locked by agent1

# Agent 2 picks different files and carries on
✓ Claimed src/middleware/validate.js
✓ Claimed src/routes/tasks.js






Agent 2 doesn't stop or ask you what to do. It just adapts and keeps going. Both agents make progress and nothing gets overwritten.









Checking in



At any point you can see exactly what's happening:




switchman status









Tasks:
Pending 2
In Progress 3
Done 1

Active Tasks:
agent1 → "Implement OAuth login flow"
agent2 → "Add rate limiting to API routes"
agent3 → "Write tests for auth middleware"

Active File Claims:
agent1: src/auth/login.js, src/auth/token.js
agent2: src/middleware/rate-limit.js, src/server.js
agent3: tests/auth.test.js, tests/middleware.test.js






Three agents, all working, no overlap anywhere.









Before you merge



When everything's done, run this before merging any branches:




switchman scan






It checks for any file overlaps or branch conflicts and tells you exactly what to fix before you touch anything. Much better than finding out at merge time.









Does it actually work?



The first time I ran three agents through a real feature with Switchman, I genuinely just went and made a coffee. Came back twenty minutes later, all three were done, no conflicts, nothing to untangle. That hadn't happened before.



It won't solve everything. If your tasks aren't well scoped, agents will still get stuck. And if you're merging a lot of branches, you'll still need to think about that. But the silent overwrite problem — the one that cost me two hours that afternoon — that's gone.






Switchman is free and open source.



Install: npm install -g switchman-dev


GitHub: github.com/switchman-dev/switchman


Site: switchman.dev

Ähnliche Beiträge
🔍 Verwandte News

Auch interessante Nachrichten How to Run 5 Claude Code Agents in Parallel Without Them Overwriting Each Other

Thematisch verwandte Begriffe: Claude, Code, Agents, Parallel · 6 Treffer

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

Laden...

Beiträge werden geladen ...

Laden...

Videos werden geladen ...

Zum Aktualisieren ziehen
ZERO-DAY CVE-2026-94497 | jshERP through 3.6 fails to validate object ownership in by-id info, upd…
Advisory →
TTS Reader • tsecurity.de Voice
tsecurity.de Icon
tsecurity.de App
Offline-Lesen, Eilmeldungen & 0ms Ladezeit

Installiere tsecurity.de direkt auf deinen Home-Bildschirm für das ultimative Vollbild-Magazinerlebnis ohne Browser-Leisten.

Nächster Beitrag
Themen-Radar & Intelligence Matrix
Echtzeit-Taxonomie nach Angriffsvektoren & Plattformen

tsecurity.de Live Threat Radar

🔴 LIVE RADAR
MONITORING
AKTIV
CVE-DATENBANK
LIVE
🔍
Community Radar & Live Chat
Sentinel Bot online • Live-Stream
Dein Cluster: Security Explorer
Match:
lädt…
Verbindung zum Community-Stream wird aufgebaut...
Bearbeitungsmodus — Senden überschreibt deine Nachricht
Community-Puls — was gerade passiert
lädt…
Aktivitäten deiner Analysten
lädt…
Neues Thema oder Eilmeldung einreichen

Reiche interessante Links, Zero-Days oder Debatten ein. Die Community entscheidet per Upvote über die Veröffentlichung.

Heiß diskutierte Einreichungen
🔖 Gespeicherte Artikel
📂 Keine gespeicherten Artikel vorhanden.
Zurück Ziehen Vor
Links: vorheriger Artikel Rechts: nächster Artikel unten: schließen
News NIS-2 Frühwarnung Tier-1 Intel ⏱️ 3 Min vor 10 Min
Artikeldaten werden geladen...

Zurück: vorheriger Vor: nächster
↗ Original-Quelle
Social Reaktionen Deine Reaktion zählt
Einstufung & Relevanz-Poll 0 Stimmen
In sozialen Netzwerken teilen 1-Klick