Preflight Checklist
Problem Statement
I run long Claude Code sessions (4-8 hours) where the agent accumulates significant context — codebase knowledge, background job status, in-progress work, conversation history. Currently, sessions are tied to the terminal/machine they started on.
Use case: I often need to step away from my desk but want to check status of background jobs, give quick commands, or continue working from my phone. Right now I have to start a completely fresh session from claude.ai/code on mobile, which loses all accumulated context and requires re-reading project files to get back up to speed.
What I'd like:
Ability to "mirror" or resume an active Claude Code session from another device (phone browser, different computer, etc.)
Read-only mode would even be useful — just being able to see what the agent is doing and its recent output from my phone
Bonus: ability to send commands to the running session from another device (like a shared terminal)
Current workaround: Start a new session on the other device and rely on project memory files (CLAUDE.md, etc.) to rebuild context. Works but loses conversation history and any in-memory state.
This would be especially valuable for sessions that involve long-running background tasks (data pipelines, scraping jobs, bulk operations) where you want to check in periodically without being physically at the original machine.
Proposed Solution
Claude Code sessions could be backed by a cloud-synced session state (similar to how claude.ai syncs conversation threads across devices). When a user opens claude.ai/code or the Claude Code desktop/mobile app, they'd see a list of active sessions and could "attach" to any of them.
Implementation could be tiered:
Phase 1 — Read-only mirror: Stream the session transcript to other devices in real-time. User can see tool calls, outputs, and agent responses but can't send commands. Low complexity — essentially a log viewer with auth.
Phase 2 — Send commands: Allow typing into the session from any attached device. Only one device can have input focus at a time (like screen sharing). The session process still runs on the original machine; remote devices just send input and receive output.
Phase 3 — Portable sessions: Session state fully cloud-hosted so it doesn't depend on the original machine staying online. This is the hardest but most valuable — a session started on a desktop Mac could be resumed from a phone even if the Mac sleeps.
The key architectural insight: the conversation context and tool permissions are the valuable part, not the terminal process. If those could be serialized and resumed elsewhere, most of the value is unlocked.
Alternative Solutions
No response
Priority
High - Significant impact on productivity
Feature Category
Interactive mode (TUI)
Use Case Example
I'm building a property data platform. A typical work session looks like this:
9:00 AM (desk) — I start a Claude Code session. The agent reads ~20 project context files, builds up knowledge of the codebase, database schema, and current priorities. This takes 5-10 minutes.
9:15 AM — I kick off a data scraping job that will run for 2 hours. The agent launches it with nohup and monitors progress. We continue working on frontend features while it runs.
10:30 AM — I need to leave for a meeting. The scraper is 60% done. I want to check if it finished from my phone during the meeting, and if it did, tell the agent to start the next pipeline step.
Today's reality: I can't. I either wait at my desk, or start a brand new session on my phone that has zero context about what's running, what we were working on, or what comes next. Re-reading all context files burns 5-10 minutes and the agent still doesn't know about the in-flight background job.
With this feature: I open Claude Code on my phone, see my active session, tap to attach. I see the agent's last message ("Scraper at 82%, 340 properties remaining"). I type "when it finishes, run the SOS enrichment pipeline for the new county." Done — 15 seconds from my phone during a meeting break.
This pattern repeats daily for anyone running long builds, data pipelines, test suites, or deployments through Claude Code.
Additional Context
Similar features in other tools:
VS Code Remote / Live Share — attach to a running VS Code session from any device, shared terminal included
tmux / screen — terminal multiplexers let you detach from a session on one machine and reattach from another over SSH. This is essentially "tmux for Claude Code" but cloud-native
Google Docs — multiple people can view/edit the same document simultaneously from any device. The session transcript is analogous to a shared document
GitHub Codespaces — cloud-hosted dev environments that persist across devices and browser sessions
Technical considerations:
Session state is mostly text. The conversation transcript, tool call history, and file context are serializable. The heavy part (the LLM context window) is already server-side at Anthropic — the local terminal is mainly an I/O layer.
Background processes are the tricky part. A nohup job runs on the original machine's OS. Phase 1-2 (mirror/send commands) would still require the original machine to be online. Phase 3 (portable sessions) would need cloud-hosted compute, which is a bigger lift.
Permission model matters. If someone attaches from a new device, the existing tool permissions (file access, bash execution) should carry over without re-prompting. But there should be auth to prevent unauthorized session access.
Conflict resolution is simple. Unlike collaborative editing, Claude Code sessions are single-threaded — one human, one agent. At most you'd need "only one device has input focus" (like a mutex), not full CRDT-style conflict resolution.
Mobile UX consideration. Even a read-only transcript viewer on mobile would cover 80% of the use case. Most phone interactions would be quick status checks and short commands, not writing code. A simplified mobile UI that shows recent agent output + a text input would be sufficient.
Preflight Checklist
Problem Statement
I run long Claude Code sessions (4-8 hours) where the agent accumulates significant context — codebase knowledge, background job status, in-progress work, conversation history. Currently, sessions are tied to the terminal/machine they started on.
Use case: I often need to step away from my desk but want to check status of background jobs, give quick commands, or continue working from my phone. Right now I have to start a completely fresh session from claude.ai/code on mobile, which loses all accumulated context and requires re-reading project files to get back up to speed.
What I'd like:
Ability to "mirror" or resume an active Claude Code session from another device (phone browser, different computer, etc.)
Read-only mode would even be useful — just being able to see what the agent is doing and its recent output from my phone
Bonus: ability to send commands to the running session from another device (like a shared terminal)
Current workaround: Start a new session on the other device and rely on project memory files (CLAUDE.md, etc.) to rebuild context. Works but loses conversation history and any in-memory state.
This would be especially valuable for sessions that involve long-running background tasks (data pipelines, scraping jobs, bulk operations) where you want to check in periodically without being physically at the original machine.
Proposed Solution
Claude Code sessions could be backed by a cloud-synced session state (similar to how claude.ai syncs conversation threads across devices). When a user opens claude.ai/code or the Claude Code desktop/mobile app, they'd see a list of active sessions and could "attach" to any of them.
Implementation could be tiered:
Phase 1 — Read-only mirror: Stream the session transcript to other devices in real-time. User can see tool calls, outputs, and agent responses but can't send commands. Low complexity — essentially a log viewer with auth.
Phase 2 — Send commands: Allow typing into the session from any attached device. Only one device can have input focus at a time (like screen sharing). The session process still runs on the original machine; remote devices just send input and receive output.
Phase 3 — Portable sessions: Session state fully cloud-hosted so it doesn't depend on the original machine staying online. This is the hardest but most valuable — a session started on a desktop Mac could be resumed from a phone even if the Mac sleeps.
The key architectural insight: the conversation context and tool permissions are the valuable part, not the terminal process. If those could be serialized and resumed elsewhere, most of the value is unlocked.
Alternative Solutions
No response
Priority
High - Significant impact on productivity
Feature Category
Interactive mode (TUI)
Use Case Example
I'm building a property data platform. A typical work session looks like this:
9:00 AM (desk) — I start a Claude Code session. The agent reads ~20 project context files, builds up knowledge of the codebase, database schema, and current priorities. This takes 5-10 minutes.
9:15 AM — I kick off a data scraping job that will run for 2 hours. The agent launches it with nohup and monitors progress. We continue working on frontend features while it runs.
10:30 AM — I need to leave for a meeting. The scraper is 60% done. I want to check if it finished from my phone during the meeting, and if it did, tell the agent to start the next pipeline step.
Today's reality: I can't. I either wait at my desk, or start a brand new session on my phone that has zero context about what's running, what we were working on, or what comes next. Re-reading all context files burns 5-10 minutes and the agent still doesn't know about the in-flight background job.
With this feature: I open Claude Code on my phone, see my active session, tap to attach. I see the agent's last message ("Scraper at 82%, 340 properties remaining"). I type "when it finishes, run the SOS enrichment pipeline for the new county." Done — 15 seconds from my phone during a meeting break.
This pattern repeats daily for anyone running long builds, data pipelines, test suites, or deployments through Claude Code.
Additional Context
Similar features in other tools:
VS Code Remote / Live Share — attach to a running VS Code session from any device, shared terminal included
tmux / screen — terminal multiplexers let you detach from a session on one machine and reattach from another over SSH. This is essentially "tmux for Claude Code" but cloud-native
Google Docs — multiple people can view/edit the same document simultaneously from any device. The session transcript is analogous to a shared document
GitHub Codespaces — cloud-hosted dev environments that persist across devices and browser sessions
Technical considerations:
Session state is mostly text. The conversation transcript, tool call history, and file context are serializable. The heavy part (the LLM context window) is already server-side at Anthropic — the local terminal is mainly an I/O layer.
Background processes are the tricky part. A nohup job runs on the original machine's OS. Phase 1-2 (mirror/send commands) would still require the original machine to be online. Phase 3 (portable sessions) would need cloud-hosted compute, which is a bigger lift.
Permission model matters. If someone attaches from a new device, the existing tool permissions (file access, bash execution) should carry over without re-prompting. But there should be auth to prevent unauthorized session access.
Conflict resolution is simple. Unlike collaborative editing, Claude Code sessions are single-threaded — one human, one agent. At most you'd need "only one device has input focus" (like a mutex), not full CRDT-style conflict resolution.
Mobile UX consideration. Even a read-only transcript viewer on mobile would cover 80% of the use case. Most phone interactions would be quick status checks and short commands, not writing code. A simplified mobile UI that shows recent agent output + a text input would be sufficient.