v0.37.0

Features

• Test-map engine launcher (t1852_4): Added the ait testmap command. It finds the test-map engine and checks that the engine's version matches your framework version before handing over.
• Test-map engine installed by setup and upgrade (t1852_5): ait setup and ait upgrade now download the engine for your exact framework version and verify its checksum. New flags (--local-engine, --engine-from-source, --force-engine, --no-testmap) and developer verbs (ait engine build|test|cross|prune) control how it is installed.
• Framework home report and migration (t1852_6): ait engine home shows where the framework's home directory lives. ait engine home --migrate moves the legacy ~/.aitask root into $AITASKS_HOME and leaves a symlink behind; if any step fails, it rolls the move back.
• Tabbed multi-trail sessions (t1895): ait trails can keep several trails open as tabs. Switch with [ / ] or 1–9 and close with ctrl+w. Background tabs refresh on their own and show a marker when they change.
• Pick a task from a trail (t1946): Press p in ait trails to open the same /aitask-pick dialog as the board for the focused trail member. If an agent window for that task already exists, it is focused instead.
• Newer models offered for superseded defaults (t1910): When a default model in your config has a newer replacement, ait upgrade offers to switch. ait codeagent check-superseded runs the same check on demand.
• Code agent for data sync conflicts (t1911): When a task-data sync conflicts, the conflict dialog in the syncer and the board can launch a code agent to resolve it. The agent works in a separate worktree, and other sessions' work is preserved.
• Claude Haiku 5.5 (t1916): Registered Claude Haiku 5.5 and its 1M-context variant, recorded as the successor to Haiku 4.5.
• Superseded defaults in the Settings TUI (t1921): The Settings TUI marks code-agent defaults that a newer model replaces. Press o or Enter on a marked row to switch it.

Bug Fixes

• Plan-path extraction keeps multi-part extensions (t1889): A path like SKILL.md.j2 is no longer cut down to SKILL.md. Trail and parallel-work conflict checks therefore report fewer false overlaps.
• Codex bullet concern markers parsed (t1892): Codex renders concern markers as • bullets, and monitor and minimonitor now pick those up.
• Two shadow findings no longer merge into one (t1897): A marker split across lines now stops rejoining at the next item, whatever its bullet glyph.
• Codex shadows launch inline (t1900): Codex shadows now run without the alternate screen, so their output stays readable. A leftover update dialog in the scrollback no longer makes the pane look like it is waiting for input.
• Codex 0.160 update dialog detected (t1904): The monitor recognizes the update dialog's new wording in codex-cli 0.160.
• OpenCode shadow concerns reach the monitor (t1908): OpenCode's renderer strips square brackets, so OpenCode shadows now wrap their concern markers in code spans. Rows that still arrive without brackets are reported instead of silently dropped.
• Hard-wrapped concern text rejoined correctly (t1914): A line break in the middle of a word (history.csv, follow-up) no longer corrupts the concern text or loses its disposition and effort fields.
• Syncer "Run in terminal" works (t1915): This button used to do nothing. It now runs the command you edited, in the right repository. The sync dialogs also no longer close a different dialog by mistake.
• Co-author trailer for 1M and major-only Claude IDs (t1918): Commits now say "Claude Opus 5.5" instead of the raw model id for [1m] and major-only Claude models.
• One stray pane no longer remaps a tmux session (t1922): A single pane that changes directory into another repository no longer moves its whole session to that project in the monitors. ait ide also refuses to take over a session that belongs to another project.
• Upgrades remove retired framework scripts (t1926): ait upgrade and ait setup now delete script files that the framework no longer ships. The TUIs refuse to start if a stale module would load in place of the framework's own.
• No lost updates in the model registry (t1928): The OpenCode model refresh and the usage and verified-score counters no longer overwrite each other's changes when they run at the same time.
• Codebrowser "Run in terminal" runs your edited command (t1929): The command you edited in the dialog is the one that runs. Projects whose path contains spaces also work now.
• Settings reload with no profiles (t1930): Pressing r in Settings no longer crashes when there are no execution profiles.
• Retired-skill pruning is path-aware (t1932): The pruner no longer deletes your own file just because its contents match a file the framework once shipped.
• Session field-list test updated (t1933): A test that pins the fields of a tmux session object was updated to include a recently added field.
• Verified-score test isolated (t1937): The verified-score flags test no longer commits and pushes junk scores to the real model registry.
• Settings reload survives a malformed config (t1938): Reloading with a broken config file now shows an error that names the file. The TUI keeps running.
• Codebrowser open-in-editor hardened (t1936): A missing $EDITOR program no longer crashes codebrowser, an $EDITOR with arguments (such as code -w) now works, and opening the editor no longer cancels annotation loading.
• Board editor launch hardened (t1943): The board's editor launch gets the same fixes and shares the same $EDITOR parsing rule.
• Codex working timer ticks again (t1944): Animations are back on in Codex panes; only the starfield effect is disabled.

Improvements

• Shadow >pd plan-decisions shortcode (t1894): Ask a shadow agent to list a plan's main design decisions, with the alternatives and reasons behind each. The shadow docs now include a shortcode reference table.
• >pd reports only real trade-offs (t1901): >pd now lists only choices that have a real downside, instead of filling a fixed number of entries.
• All Codex launches run inline (t1902): Every Codex pane the framework starts now runs without the alternate screen, so scrollback is kept.
• Profiles can turn off satisfaction feedback (t1907): A profile can turn off the end-of-task rating question, and the fast profile now does. Usage stats are still recorded.
• Explicit workflow end banners (t1913): Every task-workflow run now ends with a clear Complete or Stopped banner that states the task's outcome.
• Model lineage at registration (t1927): The add-model skill asks which model a new one replaces. Lineages were also filled in for older Claude and Codex models.
• OpenCode model lineages (t1940): Offers to switch superseded default models now cover OpenCode, with 54 recorded lineages.

Documentation

• Simpler concept pages (t1891): The shadow-agent, implementation-trails, task-notes and cross-repo concept pages were rewritten in a plain what / why / how structure. Their detailed internals moved to the workflow pages.
• Screen-recording docs (t1893_1): Added a reference page for the screen-recording skill, plus workflow guides for bug reports, animation timing and recording tips.

Maintenance

• OpenCode concern rendering measured (t1899): Recorded how OpenCode renders shadow concern blocks. This measurement led to the t1908 fix.
• OpenCode model registry refresh (t1919): Added 25 OpenCode models, including the Claude 5.5 family. Refreshing the registry no longer resets usage counters.
• OpenCode add-model wrapper (t1920): The OpenCode add-model wrapper now lists the --no-supersession flag.