Agent Skills: Tailscale Operator

Tailscale Operator

Author: Daniel Li
Copyright © Daniel Li. All rights reserved.

Use this skill for fast path tailnet ops: find the device, get connected, recover SSH, then solve permissions.

The goal is not to produce a long diagnosis. The goal is to restore a working remote path as fast as possible.

Default data sources

Check these first:

• Shared tailnet inventory: /home/aa/.openclaw/shared/tailnet/README.md
• Device records: /home/aa/.openclaw/shared/tailnet/devices/*.md
• Local notes: /home/aa/.openclaw/workspace-ops/TOOLS.md

Treat the shared tailnet inventory as the source of truth for:

• device aliases / nicknames
• Tailscale IPs
• SSH usernames
• any known-good SSH command
• platform notes (macOS/Linux, sudo caveats, network quirks)

If the user gives a nickname like 小m, studio, aa, or a human name tied to a device, first map it to a documented device record. If multiple matches are plausible, confirm before acting.

If the target device is not documented, say so clearly and ask for one of:

• Tailscale IP
• device name
• SSH username
• exact SSH command

Quickstart

When speed matters, default to this sequence:

# 1) find the documented device record first
# 2) test tailnet reachability

tailscale ping 100.x.y.z

# 3) try the known-good SSH path
ssh -o IdentitiesOnly=yes -i ~/.ssh/id_ed25519 user@100.x.y.z

# 4) verify identity once in
ssh -o IdentitiesOnly=yes -i ~/.ssh/id_ed25519 user@100.x.y.z 'hostname; whoami; uname -a'

If step 2 fails, fix Tailscale first. If step 2 works but step 3 fails, fix SSH auth / sshd / Tailscale SSH.

Core workflow

Decision tree

1. Resolve the target device first

   • Read shared/tailnet/devices/*.md
   • Prefer the documented SSH command exactly as stored
   • Do not improvise usernames or addresses until the inventory has been checked

2. Test Tailscale reachability before deep SSH debugging

   • Fast path:
     • tailscale ping <ip-or-name> when available
     • ping -c 1 100.x.y.z
     • nc -vz 100.x.y.z 22
   • This separates “tailnet is broken/offline” from “tailnet works but SSH is broken”

3. Try the simplest known-good SSH command

   • Example:
     • ssh -o IdentitiesOnly=yes -i ~/.ssh/id_ed25519 user@100.x.y.z

4. If SSH fails, classify fast by exact error text

   • Permission denied (publickey) → key/auth problem
   • Operation timed out / No route to host → device offline, Tailscale broken, ACL, firewall, wrong IP
   • Host key verification failed → stale known_hosts entry or rebuilt machine
   • Connection refused → sshd not listening, Tailscale SSH not enabled, firewall, wrong port
   • asks for password but none known → key missing, wrong user, or password auth fallback
   • sudo: a password is required → account has sudo but needs password
   • not in the sudoers file → privilege problem; need another admin path

5. Apply the fastest safe fix

   • Fix host key mismatch
   • Try documented alternate username if one exists
   • Install public key on remote account
   • Use Tailscale SSH if ordinary SSH is the blocker and the tailnet policy allows it
   • For permissions, prefer an existing admin account over changing system policy

6. Verify and leave a stable path

   • Re-test SSH
   • Confirm the remote host identity: hostname; whoami; uname -a
   • If a new working path was discovered, record it in the device note

Fast fixes by failure mode

Error → fastest next command

| Failure mode | Likely cause | Fastest next command | |---|---|---| | Host key verification failed | stale host key | ssh-keygen -R 100.x.y.z | | Permission denied (publickey) | wrong user/key or missing public key | ssh -o IdentitiesOnly=yes -i ~/.ssh/id_ed25519 user@100.x.y.z | | timeout / No route to host | node offline / Tailscale broken / wrong IP | tailscale ping 100.x.y.z | | Connection refused | SSH service closed / disabled | nc -vz 100.x.y.z 22 | | asks for password unexpectedly | key auth not working | cat ~/.ssh/id_ed25519.pub | | sudo: a password is required | sudo works but needs password | stop and ask for the admin password | | not in the sudoers file | wrong account / no admin rights | switch to an existing admin account |

Then follow the detailed branch below.

1) Host key verification failed

Use:

ssh-keygen -R 100.x.y.z
ssh-keygen -R hostname

Then reconnect.

2) Permission denied (publickey)

Try, in order:

1. Confirm username from device notes
2. Force the intended key:

ssh -o IdentitiesOnly=yes -i ~/.ssh/id_ed25519 user@100.x.y.z

3. If you already have another path onto the machine, install your key:

mkdir -p ~/.ssh && chmod 700 ~/.ssh
echo 'YOUR_PUBLIC_KEY' >> ~/.ssh/authorized_keys
chmod 600 ~/.ssh/authorized_keys

4. On macOS/Linux, verify ownership:

chown -R "$USER":$(id -gn) ~/.ssh

3) Connection refused

Check whether the host is up but SSH is closed.

Useful checks on the remote side (if you have console/another path):

sudo systemctl status ssh
sudo systemctl status sshd
sudo ss -ltnp | grep ':22'

On macOS:

sudo systemsetup -getremotelogin
sudo systemsetup -setremotelogin on
sudo launchctl list | grep ssh

4) Timeout / no route

Suspect Tailscale before SSH.

From the local side, prefer these checks first:

tailscale ping 100.x.y.z
ping -c 1 100.x.y.z
nc -vz 100.x.y.z 22

If you have another path onto the remote side, check:

tailscale status
tailscale ip -4
tailscale whois 100.x.y.z

If deeper Tailscale diagnosis is needed on the node itself, use:

tailscale netcheck
tailscale status --json

If the node is missing/offline, ask the user to wake it, log in, or open Tailscale.

5) Password prompt but goal is no-password login

Set up public-key auth or Tailscale SSH.

Preferred: install your SSH public key

Display local public key if needed:

cat ~/.ssh/id_ed25519.pub

Install it on the target account's ~/.ssh/authorized_keys.

Alternate: Tailscale SSH

Use when normal sshd management is painful and Tailscale policy allows it.

Prefer this branch when:

• the node is clearly online in Tailscale
• ordinary SSH is blocked by key setup, user mismatch, or disabled sshd
• the user wants the fastest passwordless path

Useful checks:

tailscale status --json
tailscale ping 100.x.y.z
tailscale up

Decision rule:

• if tailnet policy already allows Tailscale SSH, prefer enabling/using it over spending too long fixing legacy sshd
• if tailnet policy does not allow it or cannot be changed quickly, fall back to standard sshd + authorized_keys

Permission / sudo playbook

Fast rule

• If the user only needs ordinary commands, do not touch sudo first.
• If the requested task needs system changes, say early that admin rights are required.
• If sudo needs a password and none is available, stop escalating and say exactly that.

Case A: user has sudo, but password is required

• This is normal on many macOS/Linux hosts.
• Fastest route: ask for the account password or switch to an already-admin session.
• Do not weaken sudo policy just to avoid the password.

Case B: user is not in sudoers

• Find an existing admin account first.
• On macOS, an admin user can add another admin user:

sudo dseditgroup -o edit -a username -t user admin

• On Linux:

sudo usermod -aG sudo username
# or on some distros
sudo usermod -aG wheel username

• Only do this with clear user approval.

Case C: file permission blocks key usage

Check:

ls -ld ~/.ssh
ls -l ~/.ssh

Typical safe permissions:

• ~/.ssh = 700
• authorized_keys = 600
• private key = 600

Preferred command patterns

Standard SSH over tailnet

ssh -o IdentitiesOnly=yes -i ~/.ssh/id_ed25519 user@100.x.y.z

Quick remote sanity check

ssh -o IdentitiesOnly=yes -i ~/.ssh/id_ed25519 user@100.x.y.z 'hostname; whoami; uname -a'

Copy local public key when password login is temporarily available

ssh-copy-id -i ~/.ssh/id_ed25519.pub user@100.x.y.z

If ssh-copy-id is unavailable, append manually.

macOS specifics

For Macs, check these early:

• Remote Login enabled
• correct account short name
• FileVault/login state if the machine was just rebooted
• Tailscale app running and signed in
• whether an admin password or physical unlock is still required after reboot

Common commands:

scutil --get ComputerName
whoami
systemsetup -getremotelogin

Tailnet troubleshooting heuristics

Use the fastest likely fix, not a long generic checklist.

• If the IP changed recently, re-read tailscale ip -4 on the remote node or ask the user for the new IP.
• If the device exists in inventory but SSH fails, prefer fixing auth over replacing the whole connection method.
• If multiple devices match the same human label, confirm the exact host before making changes.
• If tailscale ping fails, stop deep SSH debugging and fix tailnet reachability first.
• If tailscale ping works but TCP 22 fails, focus on SSH/Tailscale SSH rather than general network theory.
• For recurring tasks, update the relevant shared/tailnet/devices/*.md file with the working command/path.

Record-back rules

When new durable facts are discovered, write them back to the relevant device note when appropriate:

• confirmed Tailscale IP
• confirmed SSH username
• confirmed working SSH command
• whether passwordless public-key auth works
• whether Tailscale SSH is enabled/usable
• whether sudo requires a password
• platform-specific caveats (for example: network is slow, needs physical unlock after reboot)

Do not write uncertain guesses back into the inventory.

Do not do these by default

• Do not modify sudo policy just to avoid a password prompt.
• Do not promote a user to admin/sudoers without explicit approval.
• Do not wipe or replace the remote ~/.ssh directory to "fix" login.
• Do not change firewall/network settings before confirming the actual failure mode.
• Do not operate on an ambiguously matched device name.

Output style

When helping the user:

• State the likely root cause in one line.
• Give the fastest next command first.
• Only expand into deeper branching steps if that command fails.
• If access is impossible without user action (password, physical unlock, admin approval), say exactly what is needed.

What success looks like

A good outcome is not just “I diagnosed it”. A good outcome is one of:

• SSH works now
• Tailscale SSH works now
• public key installed and future logins are passwordless
• sudo/admin path is known and documented
• blocker is explicit and minimal (e.g. “need admin password once on the Mac console”)