name	remote-development
description	Use this skill when the user wants to develop in a remote server directory through this project's SSH tools, including connecting over SSH, mounting a remote project into a local workspace path, editing mounted files locally, and running project commands remotely instead of on the local machine.

Remote Development

When to use

Use this skill when the user asks for any of the following:

edit a remote project through an SSH mount
develop inside a remote workspace while keeping runtime commands on the server
connect to a server, mount the project locally, then work on it without using the local runtime
develop in a remote server directory
mount a remote project locally before making edits
run tests, builds, dev servers, or diagnostics over SSH in the remote environment
run commands like uv run, npm dev, cargo test, git status, or git diff against a remote project

Core rules

Prefer mount-first development. The default path is: connect, mount, edit locally in the mounted tree, execute project commands remotely.
Local file edits inside the mounted directory are allowed.
Local shell commands against the mounted project are forbidden.
Treat any command as project-aware if its cwd, arguments, globs, or target paths touch the mounted tree. Those commands must run through SSH tools on the remote host.
NEVER run remote-project commands on the local machine for convenience. This includes rg, find, ls, git status, git diff, uv run, npm dev, cargo test, build scripts, migrations, formatters, watchers, and service startup when they target the remote project.
If a mount is present but looks stale or broken, stop using local operations on that tree until the mount is verified or replaced.

Preferred workflow

Read references/mount-workflow.md for the full mount-first procedure, including state inspection, connection reuse, deterministic mount-target selection, remote cwd alignment, and remote command execution after the mount is established.

Fallback workflow

Use the fallback path only when mount-first is unavailable or unsafe. Read references/ssh-fallback.md for the exact trigger conditions and the no-mount file-operation workflow.

Tool selection

Read references/operations.md for tool-selection guidance, mount failure handling, stale-mount recovery, and cleanup behavior.

Failure handling

Read references/operations.md when you need recovery steps for missing mount capability, policy-bound mount targets, stale mounts, duplicate state, or tunnel-based access to remote services.

Cleanup

Do not eagerly unmount or disconnect at the end of a normal development task.
Preserve healthy reusable connections and mounts unless the user asks for cleanup or the current state is broken.

Project memory

If the user specifies or confirms the remote-development target for a project, record the reusable non-secret connection details in that target project's AGENTS.md so future connection requests for that project can reuse them directly.
Store only reusable metadata such as host alias, host, user, port, auth kind, remote project path, preferred mount target naming, and expected tunnel ports.
Never store secrets in AGENTS.md, including passwords, tokens, private key contents, or passphrases.
If the current user request conflicts with existing remote-development details in the target project's AGENTS.md, follow the current request and update that target project's AGENTS.md if the new values should become the project default.

Remote Development

When to use

Use this skill when the user asks for any of the following:

edit a remote project through an SSH mount

develop inside a remote workspace while keeping runtime commands on the server

connect to a server, mount the project locally, then work on it without using the local runtime

develop in a remote server directory

mount a remote project locally before making edits

run tests, builds, dev servers, or diagnostics over SSH in the remote environment

run commands like uv run, npm dev, cargo test, git status, or git diff against a remote project

Core rules

Prefer mount-first development. The default path is: connect, mount, edit locally in the mounted tree, execute project commands remotely.

Local file edits inside the mounted directory are allowed.

Local shell commands against the mounted project are forbidden.

Treat any command as project-aware if its cwd, arguments, globs, or target paths touch the mounted tree. Those commands must run through SSH tools on the remote host.

NEVER run remote-project commands on the local machine for convenience. This includes rg, find, ls, git status, git diff, uv run, npm dev, cargo test, build scripts, migrations, formatters, watchers, and service startup when they target the remote project.

If a mount is present but looks stale or broken, stop using local operations on that tree until the mount is verified or replaced.

Preferred workflow

Fallback workflow

Use the fallback path only when mount-first is unavailable or unsafe. Read references/ssh-fallback.md for the exact trigger conditions and the no-mount file-operation workflow.

Tool selection

Read references/operations.md for tool-selection guidance, mount failure handling, stale-mount recovery, and cleanup behavior.

Failure handling

Read references/operations.md when you need recovery steps for missing mount capability, policy-bound mount targets, stale mounts, duplicate state, or tunnel-based access to remote services.

Cleanup

Do not eagerly unmount or disconnect at the end of a normal development task.

Preserve healthy reusable connections and mounts unless the user asks for cleanup or the current state is broken.

Project memory

If the user specifies or confirms the remote-development target for a project, record the reusable non-secret connection details in that target project's AGENTS.md so future connection requests for that project can reuse them directly.

Store only reusable metadata such as host alias, host, user, port, auth kind, remote project path, preferred mount target naming, and expected tunnel ports.

Never store secrets in AGENTS.md, including passwords, tokens, private key contents, or passphrases.

If the current user request conflicts with existing remote-development details in the target project's AGENTS.md, follow the current request and update that target project's AGENTS.md if the new values should become the project default.