garrett 1.2.0

• A fresh install settles in for a day (LEARN_HOURS). What is plainly an
  attack is stopped from the first second; a program that only its score
  speaks against is reported and left running until its owner judges it.
• On a machine with no service, garrett learns from its owner's verdicts:
  signals they keep calling theirs count for less, ones they keep
  confirming count for more, and two signals that confirmed attacks share
  become a sign of attack there. No model is involved.
• New commands: "garrett confirm " (that was an attack) and
  "garrett learned" (what this machine has learned; --reset forgets it).

garrett 1.1.0

• First release from the sentinels repository.
• The agent keeps itself current: it installs each new version by
  itself, after checking the project's signature on it. UPDATE=off in
  /etc/garrett/garrett.conf turns that off. "garrett update" does it now.
• Linux: a burst of failed SSH sign-ins from one address is reported.
• Windows: startup entries that appear while the agent watches are judged,
  and switched off when they run what an intruder would.
• A service that calls back the address that attacked it has that one
  address refused for an hour.
• Windows: the agent no longer takes its own Security-log lookup for an
  attack.