When gnupg-agent is enabled and seahorse is used for managing ssh keys, it crashes as soon as a key is added because the seahorse-agent cache is not initialised.

Steps to reproduce:
1. Make sure gnupg-agent is installed and running (check that GPG_AGENT_INFO is in your environment)
2. Run ssh-add to add an ssh key to your agent (seahorse).

seahorse-agent dies almost exactly when I have just run ssh-add.

This is Debian bug #407800: http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=407800

I will attach a patch. Here is an analysis of the bug.

Program received signal SIGABRT, Aborted.

+ Trace 146265

Here's what happens.

main [0]
  seahorse_agent_prefork [1]
  seahorse_agent_init [2]
    seahorse_agent_cache_init [3]
  gtk_main
    { ... }
      g_timeout_add (update_status) [4]
    update_status
      seahorse_agent_status_update
        seahorse_agent_cache_count
          g_assert [5]


[0] Program starts.

[1] seahorse_agent_prefork() detects another GPG agent already running 
and returns early, leaving seahorse_agent_enabled (global) set to FALSE.

[2] seahorse_agent_init() returns immediately because 
seahorse_agent_enabled was FALSE.

[3] seahorse_agent_cache_init() is /not/ called, leaving the global 
g_cache variable as NULL.

[4] A key is added by the user with ssh-add, it gets picked up by 
seahorse and the update_status timer is installed.

[5] The update_status timer is called and asserts because g_cache is NULL.