All in One SSH Studio

English | 简体中文

An all-in-one SSH client inside VS Code: connection management, SFTP browsing, remote editing, system/network/process monitoring, port forwarding, plus native integration with GitHub Copilot Chat.

No binaries are pushed to the remote host, no telemetry is reported, and nothing is written to ~/.ssh. All destructive operations always require a second confirmation.

Get Started in 30 Seconds

1. Click the All in One SSH Studio icon in the Activity Bar on the left → open the Connections panel → click ＋ in the title bar to create a new connection.
2. Fill in host / port / username / authentication (password · private key · SSH Agent), then click Test Connection to validate it (this does not write any configuration).
3. After validation succeeds, save and connect. Once connected, the Dashboard and Toolbox on the right become available automatically.

For jump-host scenarios: first configure the already-whitelisted host as a normal connection, then select it from the target connection's Advanced → Jump Host (ProxyJump) dropdown. Multi-hop nesting is supported (A→B→C).

Views and Entry Points

After opening All in One SSH Studio in the Activity Bar, you will see 5 panels:

Toolbox · Resource Management

Toolbox · System Management

Toolbox · Plugin Management

Migrated: Plugin management has been moved out of the Toolbox and is now an independent top-level panel in the sidebar. See Plugin Management below.

Plugin Management (Independent Sidebar Panel)

SSH Terminal

Integrated xterm; when Terminal Directory Sync is enabled, running cd in the terminal syncs the SFTP explorer directory via OSC 7 (disabled by default and available from the SFTP panel or settings).

Automatic Account Switching (Per Connection)

Enable sudo / su switching in the connection editor: after the interactive terminal starts, it injects sudo -i -u or su - user; the exec channel uses printf base64 | base64 -d | sudo -n -H -u 'user' bash -s wrapping, with no shell injection risk; all Copilot tool calls use the same privilege-escalation path.

Remote Editing

Double-click a remote file to download it into a local cache, then save to write it back automatically. The cache path can be configured with rss.editor.tempDir; if left empty, cache/ under extension storage is used.

Copilot Chat Tools

This extension integrates SSH/SFTP/system capabilities into GitHub Copilot Chat through VS Code native Language Model Tools. No service needs to be started, and no .vscode/mcp.json configuration is required. VS Code handles tool discovery, authorization, and lifecycle management.

Available Tools

To-do management remains in the extension's Toolbox, but is no longer registered as a Copilot Chat tool, to avoid occupying tool-list and model-selection context.

How to Use

1. Connect to the target host first in the Connections panel.
2. Open Copilot Chat, type #, choose sshExec, sftpList, and so on from the popup tool list; or describe your intent directly and let Copilot choose the tools.
3. Destructive operations are intercepted by VS Code's native second-confirmation UI.

Deployment Example

Help me deploy seller-api using the prod-1 connection:
1. Run ./service.sh backup in /data/api/seller-api;
2. Upload the local /Users/me/work/seller-api/target/seller-api.jar to the remote path with the same name;
3. Recursively upload the local target/lib/ to overwrite the remote lib/;
4. Run ./service.sh restart;
5. Read the last 200 lines of log/seller-api.log and check whether there are startup errors.

Copilot will break it down as: #sshExec backup → #sftpUpload jar → #sftpUploadDir lib → #sshExec restart → #tailLog log. Each dangerous operation triggers VS Code's native confirmation dialog.

Data Security and Privacy

Where Your Data Is Stored

Outbound Network Traffic

• The extension itself does not report any telemetry and does not send any data to the author's servers.
• SSH / SFTP traffic goes only to the hosts you configure yourself (or through the jump hosts you configure yourself).
• GeoIP resolution is disabled by default. When enabled, only peer public IPs from the ESTABLISHED list are sent to the provider you choose; API keys are stored in SecretStorage.
• When you invoke a Copilot Chat tool, its result can include connection metadata, remote command output, file or log content, process details, and login records. VS Code passes that result to your selected language-model provider under that provider's privacy terms. The extension does not operate a separate relay or author-owned backend.

Import / Export of Configuration

Use ☁⏫ / ☁⏬ in the title bar of the Connections panel to export / import; you can also search Export Connection Configuration / Import Connection Configuration in the Command Palette.

When importing, all IDs are reassigned, and existing connections are never overwritten; ProxyJump references are rewritten based on the new mapping.

Unified Protection Layer for SFTP / Copilot Operations

Write / delete / rename / upload / create folder operations all go through a unified path-protection layer. No matter whether the source is the SFTP browser, the built-in VS Code file explorer (rss-sftp://), terminal OSC rss-edit, or Copilot Chat #sftp* tools, the rules are the same:

• Protected root paths (exact match, only blocks the directory itself when used as the target): / /bin /sbin /boot /dev /lib /lib64 /proc /sys /usr /var /root /home /opt /run /srv /mnt /media
• Protected subtree prefixes (if matched, any child file underneath is rejected): /bin/ /sbin/ /boot/ /dev/ /lib/ /lib64/ /proc/ /sys/ /usr/ /root/
  • By default, /etc/ is no longer included. Editing service configuration such as /etc/nginx/*.conf and /etc/systemd/** is a legitimate ops scenario. If you want stricter protection, add "/etc/" to rss.security.protectedSubtrees in your settings.
• Sensitive filenames (even if the parent directory does not match a protected subtree, matching the basename still causes rejection): .ssh/{authorized_keys,known_hosts,config,id_*}, sudoers(.d/*), shadow, gshadow, passwd, crontab

Path normalization handles special characters strictly: control characters are rejected (NUL / CR / LF / TAB / DEL); // is collapsed; . / .. are resolved and prevented from escaping above root; under POSIX semantics, \ is treated as a normal character rather than a separator. Invalid regex patterns are silently ignored and written to the output log, and will not crash the extension.

Manual bypass: when an interactive source (SFTP browser / file explorer / OSC rss-edit) hits a blocked path, a modal dialog appears with an I understand, continue button. Clicking it allows only this single operation and writes a warn-level audit log entry (including path, operation type, and connection name). If you do not want any bypass at all, set rss.security.allowBypass to false; then all blocks are hard stops. Copilot Chat tool invocations are always hard-blocked and are not affected by allowBypass to prevent AI from bypassing protections on its own.

Configurable items:

If #sshExec matches a dangerous command (rm -rf /, mkfs, dd of=/dev/, shutdown, firewall flushes, crontab -r, curl|sh, kernel parameter writes, and so on), it is rejected immediately. The error message instructs the user to run it manually in a real terminal with the appropriate identity (sudo). The command-level dangerous-command allowlist never provides any bypass switch.

Controlled Interactive Sessions

#interactiveStart first resolves the real executable path through command -v + realpath / readlink -f. By default, rss.security.interactiveCommandMode is strict, which requires the result to match the optional path allowlist. When switched to auto, ordinary programs that are owned by root, not writable by group/others, and not located in user or temporary directories may be allowed automatically. Both modes reject system-level dangerous commands, shell syntax, credential parameters, as well as shells / interpreters / command wrappers / escape-capable programs. #interactiveRead reads incrementally by cursor and treats remote output as untrusted data; #interactiveSend shows the full one-line input before every send and requires confirmation, while also rejecting shell metacharacters and reusing the dangerous-command blacklist; #interactiveClose terminates the PTY and releases resources.

• Up to 2 active sessions per connection and 10 globally; an active session is terminated automatically after 15 minutes of inactivity, and output is retained for 5 minutes after termination.
• Output buffer limit is 256 KiB, and a single read returns at most 48 KiB; ANSI / OSC control sequences are removed before data is handed to the model.
• Common password, token, and API key output values are masked first; once prompts for password, passphrase, OTP, PIN, token, verification code, and similar inputs are detected, the session is blocked and terminated immediately, and the model must not continue sending input; a real SSH terminal must be used instead.
• The input body sent by #interactiveSend is not written to the audit log. Only length, connection, session, and whether Enter was sent are recorded, to avoid persisting secrets after accidental submission.

Credential Best Practices

• Prefer SSH Agent or private keys with a passphrase over raw passwords.
• Put jump hosts in front of critical servers to reduce public exposure.
• Use Test Connection instead of connecting directly so you can validate authentication without writing known_hosts.

What Commands Are Executed on the Remote Host?

This extension does not install any agent or download any binary to the remote host. All functionality is built on standard OpenSSH capabilities plus common Linux tools. The commands below are triggered by explicit user actions only, with no silent background invocation.

The five read-only collection categories Service Management / Scheduled Tasks / Security Audit / Software & Environment / Firewall Management are also exposed to Copilot Chat through the tools #listServices #serviceLogs #listScheduledTasks #securityAudit #listPackages #firewallOverview, using exactly the same commands as shown above; when #listPackages / #processDetail return environment variables, suspected password/key/token fields are automatically masked as ***.

Remote Dependency Check

The VS Code side never pushes any binary to the remote host, so the remote host must already provide tools such as ss / netstat, ps, /proc, ip / ifconfig, and tar / gzip. Common minimal Alpine images, container images, and customized RHEL systems often miss some of them.

Entry points:

• ✓ icon in the title bar of the Connected Sessions panel
• ✓ inline button on the right side of each active session row
• Click Click here to check remote dependencies → when Port Monitor is empty

What the panel provides:

• OS fingerprint (PRETTY_NAME / distro family / kernel / architecture / whether root)
• Probe results for 9 capability categories (command + details)
• Recommended install command automatically selected by distribution (apt / yum / apk / pacman) + one-click copy
• Open Terminal button to paste and run immediately in the SSH terminal

Settings

All settings are available in VS Code Settings → search for All in One SSH Studio, or edit settings.json directly:

Set or clear a GeoIP API key with All in One SSH Studio: Securely Set GeoIP API Key… in the Command Palette. The value is stored in VS Code SecretStorage and is not written to settings.json.

settings.json Example

{
  "rss.dashboard.refreshInterval": 5,
  "rss.network.refreshInterval": 2,
  "rss.geoip.provider": "ip-api",
  "rss.terminal.syncCwd": false
}

Command Palette Quick Reference

License

MIT