v0.16.0: Interactive agent launch mode, File references on tasks, and Plan verification tracking
v0.16.0 is a big one — 46 tasks landed, headlined by interactive agents you can actually watch, a file-references system that ties tasks to specific lines of code, and smarter plan verification that stops duplicated work across agents.
Interactive agent launch mode
You can now run agentcrew agents in interactive mode instead of headless, which means they spawn inside a tmux window you can attach to and watch live. Flip the mode per-agent from the brainstorm wizard, from the Status tab with e, or via the new ait crew setmode CLI — each agent type ships with a sensible default that you can override in the Settings TUI.
File references on tasks
Tasks can now carry a file_references list pointing at specific files and line ranges like foo.py:10-20^30-40. Open the codebrowser, select a block, press n, and a new task is created pre-seeded with that exact range. If the new task overlaps with an existing pending task’s file refs, you’ll get offered an auto-merge. The board’s task-detail modal shows these refs as a clickable row that jumps straight back into the codebrowser at the right line.
Plan verification tracking
Plans now record which agents have verified them against the current codebase. Combined with the new plan_verification_required and plan_verification_stale_after_hours profile keys, a pick can skip re-verification when another agent validated the plan recently — no more repeating the same work across agent runs.
ANSI log viewer and task restart
A new ait crew logview TUI tails agent log files with ANSI color rendering, live search, and a raw-mode toggle. Press L from the brainstorm Status tab or monitor to open it for the focused agent. And when an agent goes off the rails, R on an idle pane in ait monitor now kills the window and restarts the task cleanly.
Monitor preview that actually stays put
The monitor preview remembers where you scrolled on each pane, freezes with a PAUSED badge when you scroll up from the tail, and re-engages with t. Tmux refreshes run async now, so arrow keys don’t get eaten by refresh ticks and the whole TUI stays responsive even with a lot of agents.
Full changelog: v0.16.0 on GitHub