Remote Engineering with Claude Code: Managing Agentic Workflows via Telegram

Remote Slop with Claude Code

Mladen Stepanić tested the new Claude Code Channels integration to manage a production refactoring session entirely through a Telegram bot. The experiment successfully coordinated 22 agents across 5 teams to merge 5 pull requests without direct terminal access.

Why This Matters

Remote agentic workflows face a technical ‘permission wall’ where security prompts in the local terminal stall the remote session, often requiring the use of high-risk flags like —dangerously-skip-permissions. This highlights the friction between local security guardrails and the desire for asynchronous development cycles where feedback loops shift from milliseconds to minutes due to deployment latency.

Key Insights

• Anthropic’s Claude Code Channels (2026) uses the Model Context Protocol (MCP) to bridge local filesystem and git access with external chat interfaces like Telegram and Discord.
• The ‘Permission Wall’ occurs because standard Claude Code permission prompts for destructive actions do not forward to the remote bot, causing silent session stalls.
• Context management in remote sessions is limited by the 1M token window of Claude Opus and the current inability to use the /clear command from Telegram.
• The ‘beads’ concept, popularized by Steve Yegge, allows for decomposing work into small, scoped tasks that sub-agents can implement without exhausting the main session’s context.
• Playwright can be utilized by remote agents to perform visual verification by providing screenshots of layout changes directly within the chat interface.

Working Examples

The flag required to bypass local permission prompts when controlling Claude Code via remote channels like Telegram.

claude --dangerously-skip-permissions

Practical Applications

• Coordinating high-level refactorings via OpenSpec where agents implement details independently. Pitfall: Disabling permissions can lead to unauthorized branch history rewrites if branch protection is not enabled.
• Using Playwright for automated visual verification of Vercel preview builds during remote sessions. Pitfall: High latency in the feedback loop makes this unsuitable for rapid UI iteration or pixel-pushing.

References:

• https://dev.to/crawleyprint_71/remote-slop-with-claude-code-329c