agent-fleet 0.2.1

Changed

• The PyPI distribution is named fleetq (pipx install fleetq): PyPI refuses agent-fleet
  because agentfleet already exists. The command is still fleet, and the repo, skill and plugin
  are still agent-fleet.
• Added .github/workflows/release.yml, a manual trusted-publishing (OIDC) workflow for PyPI.

Install: pipx install fleetq then fleet install-skill.

agent-fleet 0.2.0

First tagged release.

Added

• Standalone terminal windows — fleet run --windows opens one window per worker in
  gnome-terminal, kitty, alacritty, konsole, xfce4-terminal or xterm ($FLEET_TERMINAL picks one).
  Works on Wayland, needs no xdotool. fleet run now chooses visible terminals automatically
  (VS Code on X11, otherwise windows) and falls back to headless only with no display.
• fleet run --vscode to force VS Code terminals only.
• A person in the loop — fleet run --ask keeps workers interactive at a login, captcha, 2FA or
  question; fleet unblock ID --answer "..." appends your answer to the task's notes, and clearing a
  wall no longer counts as a failed attempt. fleet blocked shows how to answer each parked task.
• AGENT_DIR / FLEET_WORKSPACE — keep the queue under another name or below the repo root; the
  engine, launcher and ignore rule agree on where it is.
• fleet --version, a single VERSION source, and drift checks that keep CITATION.cff and the
  plugin manifests in step with it.
• Packaging — PyPI (pipx install agent-fleet, then fleet install-skill), a Claude Code plugin
  marketplace for the skill, and a release tarball with a SHA-256.
• Starter TASKS.md from fleet init; checklist / checklist_user_marker documented.
• Dispatch-flow diagram in the README.

Changed

• The parallel lane is opt-in and collision-aware (was opt-out).
• Workers start as claude --chrome; the browser probe measures the mode workers actually run in.
• Self-test suite: 107 checks, no desktop alerts raised while it runs.