[zsh] Make ssh loading across platform more robust

bin/.local/bin/ssh-agent-doctor

@@ -0,0 +1,130 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+usage() {
+  cat <<'EOF'
+ssh-agent-doctor: diagnose and select the best SSH_AUTH_SOCK.
+
+Usage:
+  ssh-agent-doctor            # human-readable report
+  ssh-agent-doctor --export   # print: export SSH_AUTH_SOCK='...'
+  ssh-agent-doctor --json     # minimal JSON-ish output (no jq required)
+
+Exit codes:
+  0  OK (agent reachable)
+  1  Socket found but agent has no identities (reachable)
+  2  No usable socket found OR cannot connect to agent
+EOF
+}
+
+mode="report"
+case "${1:-}" in
+  --export) mode="export" ;;
+  --json) mode="json" ;;
+  -h|--help) usage; exit 0 ;;
+  "") ;;
+  *) echo "Unknown arg: $1" >&2; usage; exit 2 ;;
+esac
+
+os="$(uname -s 2>/dev/null || echo unknown)"
+uid="$(id -u)"
+user="$(id -un 2>/dev/null || echo unknown)"
+home="${HOME:-/}"
+
+linux_sock="/run/user/${uid}/ssh-agent.socket"
+darwin_sock="${home}/.ssh/agent.sock"
+env_sock="${SSH_AUTH_SOCK:-}"
+
+is_sock() { [[ -n "${1:-}" && -S "$1" ]]; }
+
+pick_sock=""
+pick_reason=""
+
+if is_sock "$linux_sock"; then
+  pick_sock="$linux_sock"
+  pick_reason="linux_systemd_default"
+elif is_sock "$darwin_sock"; then
+  pick_sock="$darwin_sock"
+  pick_reason="darwin_launchd_default"
+elif is_sock "$env_sock"; then
+  pick_sock="$env_sock"
+  pick_reason="existing_env"
+fi
+
+check_agent() {
+  local sock="$1"
+  SSH_AUTH_SOCK="$sock" ssh-add -l >/dev/null 2>&1
+  return $?
+}
+
+if [[ "$mode" == "export" ]]; then
+  if [[ -z "$pick_sock" ]]; then
+    echo "echo 'ERROR: No ssh-agent socket found.' >&2"
+    echo "return 2 2>/dev/null || exit 2"
+    exit 0
+  fi
+  printf "export SSH_AUTH_SOCK='%s'\n" "$pick_sock"
+  exit 0
+fi
+
+if [[ "$mode" == "json" ]]; then
+  if [[ -z "$pick_sock" ]]; then
+    cat <<EOF
+{"ok":false,"user":"$user","uid":$uid,"os":"$os","picked":"","reason":"","error":"no_socket","candidates":{"linux":"$linux_sock","darwin":"$darwin_sock","env":"$env_sock"}}
+EOF
+    exit 2
+  fi
+
+  rc=0
+  check_agent "$pick_sock" || rc=$?
+  # ssh-add -l exit codes: 0 has keys, 1 no keys, 2 cannot connect
+  ok=false
+  status="unknown"
+  if [[ $rc -eq 0 ]]; then ok=true; status="has_identities"; fi
+  if [[ $rc -eq 1 ]]; then ok=true; status="no_identities"; fi
+  if [[ $rc -eq 2 ]]; then ok=false; status="cannot_connect"; fi
+
+  cat <<EOF
+{"ok":$ok,"user":"$user","uid":$uid,"os":"$os","picked":"$pick_sock","reason":"$pick_reason","ssh_add_l_exit":$rc,"status":"$status","candidates":{"linux":"$linux_sock","darwin":"$darwin_sock","env":"$env_sock"}}
+EOF
+  exit $rc
+fi
+
+# Human report mode
+echo "ssh-agent-doctor"
+echo "  user: $user (uid $uid)"
+echo "  os:   $os"
+echo
+echo "Candidates:"
+printf "  linux systemd: %s %s\n" "$linux_sock"  "$(is_sock "$linux_sock" && echo '[socket]' || echo '[missing]')"
+printf "  macOS launchd: %s %s\n" "$darwin_sock" "$(is_sock "$darwin_sock" && echo '[socket]' || echo '[missing]')"
+printf "  env SSH_AUTH_SOCK: %s %s\n" "${env_sock:-<unset>}" "$(is_sock "$env_sock" && echo '[socket]' || echo '[not socket]')"
+echo
+
+if [[ -z "$pick_sock" ]]; then
+  echo "Result: ❌ No usable SSH agent socket found."
+  echo "Hint: start your agent (systemd user on Linux, launchd agent on macOS) and re-run."
+  exit 2
+fi
+
+echo "Picked:"
+echo "  SSH_AUTH_SOCK=$pick_sock"
+echo "  reason=$pick_reason"
+echo
+
+rc=0
+check_agent "$pick_sock" || rc=$?
+
+if [[ $rc -eq 0 ]]; then
+  echo "Agent check: ✅ reachable, has identities"
+  SSH_AUTH_SOCK="$pick_sock" ssh-add -l || true
+  exit 0
+elif [[ $rc -eq 1 ]]; then
+  echo "Agent check: ✅ reachable, but has no identities loaded"
+  echo "Next: run your pass-based loader (load_keys)."
+  exit 1
+else
+  echo "Agent check: ❌ cannot connect to agent via that socket"
+  echo "This usually means the socket is stale or the agent died."
+  exit 2
+fi

bin/.local/bin/ssh-agent-env

@@ -0,0 +1,32 @@
+#!/usr/bin/env bash
+set -euo pipefail
+
+uid="$(id -u)"
+
+linux_sock="/run/user/${uid}/ssh-agent.socket"
+darwin_sock="${HOME}/.ssh/agent.sock"
+
+pick_sock=""
+
+# Prefer Linux systemd socket if present
+if [[ -S "$linux_sock" ]]; then
+  pick_sock="$linux_sock"
+elif [[ -S "$darwin_sock" ]]; then
+  pick_sock="$darwin_sock"
+else
+  # Fall back to existing env if it's a socket
+  if [[ -n "${SSH_AUTH_SOCK:-}" && -S "${SSH_AUTH_SOCK}" ]]; then
+    pick_sock="$SSH_AUTH_SOCK"
+  fi
+fi
+
+if [[ -z "$pick_sock" ]]; then
+  echo "echo 'ERROR: No ssh-agent socket found.' >&2"
+  echo "echo 'Tried: $linux_sock and $darwin_sock' >&2"
+  echo "return 2 2>/dev/null || exit 2"
+  exit 0
+fi
+
+cat <<EOF
+export SSH_AUTH_SOCK='$pick_sock'
+EOF

emacs/.config/doom/config.el

@@ -8,11 +8,17 @@
 (after! envrc
   (envrc-global-mode))
 
-;; Force systemd user ssh-agent socket for all Emacs subprocesses.
+;; Force user ssh-agent socket for all Emacs subprocesses.
-(let* ((xdg (format "/run/user/%d" (user-uid)))
+(defun my/apply-ssh-agent-env ()
-       (sock (expand-file-name "ssh-agent.socket" xdg)))
+  "Set SSH_AUTH_SOCK in Emacs using ~/.local/bin/ssh-agent-env."
-  (setenv "XDG_RUNTIME_DIR" xdg)
+  (let* ((cmd (expand-file-name "~/.local/bin/ssh-agent-env"))
-  (setenv "SSH_AUTH_SOCK" sock))
+         (out (and (file-executable-p cmd)
+                   (string-trim (shell-command-to-string cmd)))))
+    ;; out looks like: export SSH_AUTH_SOCK='...'
+    (when (and out (string-match "SSH_AUTH_SOCK='\\([^']+\\)'" out))
+      (setenv "SSH_AUTH_SOCK" (match-string 1 out)))))
+
+(my/apply-ssh-agent-env)
 
 ;; load pinentry
 (when (require 'pinentry nil t)
@@ -448,33 +454,22 @@ Always open the result in `eww`."
 
 ;; Bind globally to C-c l
 (global-set-key (kbd "C-c l") #'life-scrobble-url)
-
 (after! magit
   (defvar my/ssh-key-injector-script (expand-file-name "~/.local/bin/load_keys"))
 
-  (defun my/systemd-ssh-auth-sock ()
-    (format "/run/user/%d/ssh-agent.socket" (user-uid)))
-
   (defun my/ssh-add-status ()
-    "ssh-add -l exit code: 0=has keys, 1=no keys, 2=no agent."
     (call-process "ssh-add" nil nil nil "-l"))
 
   (defun my/run-ssh-key-injector ()
+    (my/apply-ssh-agent-env)
     (let* ((buf (get-buffer-create "*ssh-key-injector*"))
-           (sock (my/systemd-ssh-auth-sock))
+           (process-connection-type t)
-           (process-connection-type t) ;; PTY for pinentry-curses
            (process-environment (copy-sequence process-environment)))
-      (setenv "SSH_AUTH_SOCK" sock)
-      (setenv "XDG_RUNTIME_DIR" (format "/run/user/%d" (user-uid)))
-
       (unless (file-executable-p my/ssh-key-injector-script)
         (user-error "SSH injector script not executable: %s" my/ssh-key-injector-script))
-
       (with-current-buffer buf
         (erase-buffer)
-        (insert (format "Emacs SSH_AUTH_SOCK=%s\n" (getenv "SSH_AUTH_SOCK")))
+        (insert (format "Emacs SSH_AUTH_SOCK=%s\n\n" (or (getenv "SSH_AUTH_SOCK") "<unset>"))))
-        (insert (format "Emacs XDG_RUNTIME_DIR=%s\n\n" (getenv "XDG_RUNTIME_DIR"))))
-
       (let ((proc (make-process
                    :name "ssh-key-injector"
                    :buffer buf
@@ -483,16 +478,15 @@ Always open the result in `eww`."
                    :noquery t)))
         (while (process-live-p proc)
           (accept-process-output proc 0.05))
-        (let ((exit (process-exit-status proc)))
+        (unless (eq (process-exit-status proc) 0)
-          (unless (eq exit 0)
+          (display-buffer buf)
-            (display-buffer buf)
+          (user-error "SSH key injection failed. See *ssh-key-injector*.")))))
-            (user-error "SSH key injection failed (exit %d). See *ssh-key-injector*." exit))))))
 
   (defun my/ensure-ssh-keys-loaded (&rest _ignore)
     (pcase (my/ssh-add-status)
-      (0 nil)                    ;; already has identities
+      (0 nil)
-      (1 (my/run-ssh-key-injector)) ;; agent reachable but empty
+      (1 (my/run-ssh-key-injector))
-      (2 (user-error "No reachable ssh-agent. SSH_AUTH_SOCK=%s"
+      (2 (user-error "No reachable ssh-agent (SSH_AUTH_SOCK=%s)"
                      (or (getenv "SSH_AUTH_SOCK") "<unset>")))
       (_ (my/run-ssh-key-injector))))
 

launchd/Library/LaunchAgents/com.powellc.ssh-agent.service

@@ -0,0 +1,25 @@
+<?xml version="1.0" encoding="UTF-8"?>
+<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN"
+ "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
+<plist version="1.0">
+<dict>
+
+  <key>Label</key>
+  <string>com.user.ssh-agent</string>
+
+  <key>ProgramArguments</key>
+  <array>
+    <string>/usr/bin/ssh-agent</string>
+    <string>-D</string>
+    <string>-a</string>
+    <string>$HOME/.ssh/agent.sock</string>
+  </array>
+
+  <key>RunAtLoad</key>
+  <true/>
+
+  <key>KeepAlive</key>
+  <true/>
+
+</dict>
+</plist>

zsh/.zshrc

@@ -7,6 +7,8 @@ export ZSH="$HOME/.oh-my-zsh"
 ZSH_THEME="robbyrussell"
 
 plugins=(git z fzf asdf direnv emacs yarn aws)
+
+eval "$($HOME/.local/bin/ssh-agent-env)"
 load_keys &>/dev/null
 
 source $ZSH/oh-my-zsh.sh