Releases on aitasks

v0.32.0: Board columns you can actually manage, Grouped cards, and Reject a concern and it stays rejected

Thu, 13 Aug 2026 00:00:00 +0000

v0.32.0 is a big one for anyone who lives in the board and the monitors. Columns became something you actually manage, cards learned to group themselves, and the shadow reviewer finally stops repeating objections you’ve already dismissed.

Board columns you can actually manage

Press e on the board and you get a real Columns dialog — add, rename, recolour, reorder with Shift+Up/Shift+Down, and merge one column into another in a single move. Column creation also came to minimonitor: when a task prompt pops up, you can send it to an existing column or spin up a brand-new one right there, without ever opening the board.

v0.31.0: Mark cards and move them in bulk, The board got noticeably faster, and Every shortcut is visible again

Wed, 05 Aug 2026 00:00:00 +0000

If you spend your day on the board, this one is for you. v0.31.0 makes the kanban board feel like a real tool for handling many tasks at once — and quietly fixes a whole class of “it said it worked but it didn’t” failures underneath.

Mark cards and move them in bulk

Hit Space on any card to mark it. The checkbox glyph is always visible, so you can see at a glance what’s selected, and the selection sticks as you navigate around. Then press m, pick a destination column, and every marked task moves at once — with a review dialog first so you can back out.

v0.30.0: The shadow companion, in the full monitor, Manage every linked repo from the syncer, and Triage that keeps up with you

Fri, 31 Jul 2026 00:00:00 +0000

v0.30.0 brings the shadow companion to the full monitor, turns the syncer into a real cross-repo control panel, and adds a pile of triage shortcuts you’ll wonder how you lived without.

The shadow companion, in the full monitor

Everything the minimonitor could do with a shadow agent, the full monitor can now do too — and then some. Your shadow sits in its own column right beside the agent it’s watching, with a live preview you can type into. When it raises concerns, you get a badge on the agent’s card and a toast the moment they appear; hit c to page through them. Spawning is e / E, same as before, with a guard so you can’t accidentally end up with two shadows on one agent.

v0.29.0: Implementation trails, Work reports, and A real setup wizard for chat intake

Sun, 26 Jul 2026 00:00:00 +0000

v0.29.0 is a big one — two brand-new planning and reporting workflows, a proper setup wizard for chat intake, and Opus 5 / Sonnet 5 wired in as defaults.

Implementation trails

Big features rarely fit in one task, and figuring out what to do first across a dozen related tasks is a chore you end up redoing every week. /aitask-trail does that analysis once and stores the answer: a wave-by-wave implementation plan, backed by evidence from the actual task files, plans, and gate state. It lives as a versioned artifact, so it survives the session — and when the underlying tasks change, the trail tells you it’s stale and why. The board picks it up with a new By-Trail view (z): waves become columns, each card shows its classification and confidence, and enter opens the full narrative.

v0.28.0: Report a bug from Discord, get back a ready-to-work task, Store and share artifacts with your team, Slack, and not just Discord

Fri, 10 Jul 2026 00:00:00 +0000

v0.28.0 is a big one — a whole new way to file bugs from chat, team-wide artifact sharing, Slack support, and zero-config remote access for the mobile companion.

Report a bug from Discord, get back a ready-to-work task

Drop a bug report in Discord and a sandboxed agent takes it from there: it explores your codebase, asks you clarifying questions right in the thread, and files a structured task when it has what it needs. The agent runs in an isolated Docker container, and you watch the whole thing from the new ait chatlink status screen.

v0.27.1: Better cross-repo task creation, A tidier board, and The shadow reviews your finished work

Sun, 05 Jul 2026 00:00:00 +0000

A small follow-up release that smooths out cross-repo work and sharpens the shadow companion.

Better cross-repo task creation

If you work across linked projects, you can now create a child task directly in another repo by pointing ait create at both a project and a parent. Project-routed creation also hands back an absolute path to the new task, so downstream tooling always knows exactly where it landed.

A tidier board

The task board now hides footer actions that don’t apply to whatever you’ve got selected. No more staring at a Commit or Brainstorm button that would just no-op — you only see the moves you can actually make right now.

v0.27.0: Attach files to your tasks, Quality gates that actually run, and Teach it a new skill

Fri, 03 Jul 2026 00:00:00 +0000

v0.27.0 is a big one — task attachments, a real quality-gate framework, and a way to teach the tool new skills all landed in the same release.

Attach files to your tasks

You can now attach files to any task with the new ait attach command. Attachments are stored content-addressed and reference-counted, so the same file shared across tasks is only stored once, and blobs are garbage-collected when nothing points at them anymore — including the tricky cases where you archive or fold a task.

v0.26.1: Spin Up an Agent, No Task Required, and Upgrades That Survive GitHub Rate Limits

Thu, 25 Jun 2026 00:00:00 +0000

A small but handy maintenance release that smooths out two rough edges.

Spin Up an Agent, No Task Required

Sometimes you just want a code agent in front of you without ceremony. Hit e in the TUI switcher and you get a fresh agent window with nothing attached — pick the agent and model right there and start hacking. Perfect for quick experiments and one-off questions.

Upgrades That Survive GitHub Rate Limits

If GitHub’s API throttled you, ait upgrade used to give up with a confusing “No releases found.” Now it quietly falls back to git tags and tells you the truth about where you stand, so checking for updates keeps working even when the API doesn’t.

v0.26.0: Gate orchestration, end to end, Your terminals, on your phone, and Project groups

Wed, 24 Jun 2026 00:00:00 +0000

v0.26.0 is a big one — it lands a full gate-driven verification system, brings the mobile companion’s live-terminal data plane online, and adds project groups plus a bunch of brainstorm TUI polish.

Gate orchestration, end to end

Tasks now run their declared verification gates through a real orchestrator that handles retries, parallelism, and even detects when a gate is stuck. You can resume an interrupted task right where it left off with aitask-resume, aitask-pick surfaces in-flight tasks as resume candidates, and both the board (a new In-Flight view) and the monitor (a compact pass/pending/failed summary) show you gate progress at a glance.

v0.25.0: Task gates that actually drive the workflow, Your phone can now talk to your workspace, and Meet the shadow companion

Mon, 15 Jun 2026 00:00:00 +0000

v0.25.0 is a big one — it lands a whole task-gating system, opens the door to the mobile companion app, and introduces an advisory “shadow” agent that watches over your coding agents.

Task gates that actually drive the workflow

Tasks can now carry named approval checkpoints — gates like plan approved, build verified, and merge approved — recorded in a durable ledger. Once you turn on record_gates, those gates aren’t just notes: they decide when dependent tasks unblock, hold a task back from archival until everything’s green, and let you pick up an in-flight task right where you left off instead of starting over. It’s the connective tissue that makes the workflow resumable and dependency-aware.

v0.24.0: See your proposal while you work, Your tmux sessions stop dying, and Smarter module decomposition

Fri, 12 Jun 2026 00:00:00 +0000

v0.24.0 is a big one — 49 tasks landed, headlined by a ground-up tmux gateway, a much nicer brainstorm experience, and some quality-of-life wins for keyboard-driven workflows.

See your proposal while you work

The brainstorm TUI now shows the relevant proposal side-by-side as you configure your next step — in the explore wizard, in module-decompose, with a section minimap and adjustable split ratios. You can hit Ctrl+Shift+L to flip the preview into a syntax-highlighted, line-numbered source view. No more bouncing between screens to remember what you were building on.

v0.23.1: Installs that just re-run cleanly, and Less untracked clutter from generated skills

Thu, 04 Jun 2026 00:00:00 +0000

v0.23.1 is a small housekeeping release that smooths out installs and setup.

Installs that just re-run cleanly

Running the installer again won’t trip over itself anymore. If the global ait shim is already in place, setup quietly moves on instead of bailing out — and any stray packaging/ directory the installer used to leave behind now gets cleaned up on its own.

Less untracked clutter from generated skills

ait setup now lays down the right gitignore rules so the skill variants your tools render locally stay out of your way, while the committed headless prerenders are still tracked. Your git status stays clean.

v0.23.0: Brainstorm module decomposition, Risk evaluation in planning, Cross-repo planning, and straight from explore

Wed, 03 Jun 2026 00:00:00 +0000

v0.23.0 is a big one — two major new capabilities land, plus a pile of macOS/portability fixes and UI polish.

Brainstorm module decomposition

You can now split a brainstorm design into independent module subgraphs and work each one on its own track. Decompose a design into modules, merge or sync them as first-class brainstorm operations, and watch each module’s status update live. When a module is ready to build, “Fast-track this module” extracts it into a linked aitask in a single pass.

v0.22.1: Track risk on your tasks, Risk evaluation, and when you want it

Mon, 01 Jun 2026 00:00:00 +0000

A focused follow-up to v0.22.0’s cross-repo release, this one is all about risk-awareness in your task workflow.

Track risk on your tasks

Tasks can now carry a risk level (high/medium/low) plus a list of risk_mitigation_tasks. Set them straight from the CLI with ait update --risk high, see risk at a glance in ait ls, and edit it right in the board TUI. It’s a lightweight way to flag the work that needs extra care before it ships.

v0.22.0: Tasks that span multiple repos, Make every shortcut your own, and Manual verification that runs itself

Sun, 31 May 2026 00:00:00 +0000

v0.22.0 is a big one — cross-repo dependencies, fully customizable keyboard shortcuts, autonomous manual verification, and Opus 4.8 as the new default.

Tasks that span multiple repos

You can now wire up dependencies between aitasks projects, not just within one. A task in your frontend repo can declare it’s blocked by a task in your backend repo using the new xdeps fields, and the board, planner, and blocking logic all understand it. Projects are referenced by a logical name you register once — no fragile ../ paths — so cross-repo planning, task creation, and context lookups just work.

v0.21.0: Profile-aware skill templating, Brainstorm: auto-apply, DAG navigation, and operation detail, and Cross-repo project registry

Wed, 27 May 2026 00:00:00 +0000

v0.21.0 is a big one — a foundational refactor of how skills are authored and dispatched, the brainstorm TUI graduating from “experiment” to a real DAG-driven planning workflow, first-class cross-repo project plumbing, and a fresh mobile-companion bridge.

Profile-aware skill templating

Every aitask skill is now authored once as a .md.j2 Jinja template and rendered per execution profile, per agent. You stop maintaining four near-duplicate copies of each skill across Claude, Codex, Gemini, and OpenCode; the renderer fans out per-profile variants behind a thin stub. A new ait skillrun command lets you launch any agent with any skill at any profile (with --profile-override for ad-hoc tweaks and --dry-run previews), and the Per-Run profile editor lets you tweak the active profile right from the launch dialog. Saving a profile invalidates the right rendered variants automatically — running agents pick up the new values on next invocation.

v0.21.1: Choose-sibling picker in monitor, Brainstorm retry-apply actually retries, and Release-post YAML hardening

Wed, 27 May 2026 00:00:00 +0000

v0.21.1 is a small patch — a couple of brainstorm/monitor papercuts and a fix for the v0.21.0 release-post pipeline so the next blog post doesn’t break.

Choose-sibling picker in monitor

The monitor TUI’s next-task dialog now lets you pick any ready sibling of the current task from a list, with blocked-by-sibling annotations on each row. Mid-family pivots no longer require backing out to the board.

Brainstorm retry-apply actually retries

The ctrl+shift+x/y/d retry-apply bindings in brainstorm were silently doing nothing once their internal tracking set drained — typically right after the original apply ran. They now rescan the worktree for completed agents on every invocation, and surface a clear notify when there’s nothing to retry.

v0.20.3: Linux install pages unified, Curl-first install, with native packages as the alternative, and A page on updating model lists

Wed, 13 May 2026 00:00:00 +0000

v0.20.3 is mostly a documentation release — a big sweep across the website to make it easier to find what you need, install the framework the way you prefer, and discover features that have been there for a while but weren’t obvious.

Linux install pages unified

The three separate install pages for Arch, Debian, and Fedora are gone. There’s now a single Linux page with per-distro sections, and the Installation index has been reorganized into clear “Operating systems” and “Setup topics” groups. If you’ve ever bounced between three near-identical pages trying to find your distro, you’ll like this.

v0.20.1 Release

Sun, 10 May 2026 00:00:00 +0000

A quick maintenance release. The nfpm GitHub Action we relied on for building Debian and RPM packages was deleted upstream, which broke our release pipeline. v0.20.1 swaps it for a direct Docker invocation of the same tool, so .deb and .rpm packages are once again attached to every release.

Full changelog: v0.20.1 on GitHub

v0.20.2: A redesigned home page, Native installs for every platform, and Code browser History screen no longer crashes

Sun, 10 May 2026 00:00:00 +0000

A polish-focused release. The website got a visual makeover, install instructions now match how you actually want to install software, and a couple of nasty bugs got squashed.

A redesigned home page

The website home page now leads with a split hero — clear pitch on the left, a screenshot of the Board on the right — and gets straight to the point. Below that, a new “Take the tour” mosaic shows off the suite of TUIs (Board, Code Browser, Monitor, Settings, Stats) at a glance, with each tile linking through to the relevant docs. The top feature cards are clickable too, so you can jump straight into the part of the framework that interests you.

v0.20.0: Install via your distro's package manager, A dedicated syncer TUI, and Opt-in PyPy fast-path

Wed, 06 May 2026 00:00:00 +0000

v0.20.0 is a big one — a brand new sync TUI, package-manager installs across four ecosystems, an opt-in PyPy fast-path that makes the long-running TUIs noticeably snappier, and a wave of polish across brainstorm.

Install via your distro’s package manager

Aitasks now ships as proper packages on Homebrew, the AUR, Debian/Ubuntu (.deb), and Fedora/RHEL (.rpm). Each tag triggers a CI workflow that builds, tests, and publishes the appropriate format, so brew install aitasks, yay -S aitasks, apt install ./aitasks_*.deb, and dnf install ./aitasks-*.rpm all just work. The ait shim itself was extracted to a single canonical file so every package consumes the same binary.

v0.19.2: Python wrappers that actually stay inside the venv, Warnings before you build on stale main, and Source-tree setup gets the starter tmux config too

Wed, 29 Apr 2026 00:00:00 +0000

v0.19.2 is a reliability-focused release for the everyday setup and release workflows. It fixes a few sharp edges around Python environments, tmux setup, macOS docs, and stale task data during changelog generation.

Python wrappers that actually stay inside the venv

The framework Python launchers now use wrapper scripts instead of symlink chains. That keeps ait board and the other Python tools inside the aitasks virtual environment, so packages like Textual, PyYAML, and linkify-it are found reliably after setup.

v0.19.0: A modern Python venv that installs itself, `ait setup` on a fresh clone, polished end-to-end, Brainstorm: fewer crashes, and more visibility

Tue, 28 Apr 2026 00:00:00 +0000

v0.19.0 is mostly an infrastructure release — the Python venv story, ait setup on a fresh clone, and brainstorm stability all got a lot of attention. There’s also a new audit-wrappers skill that closes a recurring drift class.

A modern Python venv that installs itself

ait setup now requires Python ≥3.11 and will auto-install a modern interpreter for you when the system Python is too old — Homebrew on macOS, uv-managed builds on Linux. Once installed, all aitasks scripts resolve through a single helper (lib/python_resolve.sh) and a scoped ~/.aitask/bin symlink, so the framework picks up the right interpreter without ever touching your shell rc files. If you’ve been getting “Python too old” failures on Debian or older macOS, this just works now.

v0.19.1: Releases and changelogs no longer miss remote tasks, and macOS test suite is back at parity with Linux

Tue, 28 Apr 2026 00:00:00 +0000

v0.19.1 is a quick follow-up to v0.19.0 with two reliability fixes you’ll appreciate if you contribute to aitasks itself or develop on macOS.

Releases and changelogs no longer miss remote tasks

If you cut a release while local main was a few commits behind origin/main, the changelog could silently skip those tasks and the release tag could land on stale code. The release script and the /aitask-changelog skill now fetch and offer to pull before doing anything destructive, so you can run them confidently from any clone.

v0.18.3: Brainstorm agents run interactive by default, Brainstorm sessions that don't get stuck, and MissedHeartbeat is gone (and that means a small migration)

Mon, 27 Apr 2026 00:00:00 +0000

v0.18.3 is mostly a brainstorm release — the TUI got a lot more transparent and a lot more resilient when things flake out. The agent-crew state machine also took a small but breaking turn.

Brainstorm agents run interactive by default

Every brainstorm agent type — initializer, detailer, explorer, comparator, synthesizer, patcher — now launches in a tmux pane by default instead of headless. You can watch each agent work and step in if you need to, without having to override launch_mode in codeagent_config.json. There’s also a new dim-cycling activity indicator next to the initializer banner and the Status tab that flashes on each poll, so you can tell the agent is alive even when nothing visible has changed.

v0.18.1: Triage manual-verification checklists in batches, Filter the board by issue type, and `ait upgrade` actually upgrades branch-mode setups now

Sun, 26 Apr 2026 00:00:00 +0000

v0.18.1 is a polish-and-fix release. The headline changes are a faster manual-verification flow, a new issue-type filter in the board, and a fix for branch-mode upgrades that were silently leaving framework files uncommitted.

Triage manual-verification checklists in batches

Manual-verification tasks used to walk you through items one at a time. Now the skill re-renders the whole numbered checklist with state markers on every turn, and you can answer in batch through the Other field — 1 pass, 3 defer, 5 skip not applicable lands four state changes in a single response. The single-item Pass/Fail/Skip/Defer prompt is still there for the items that actually need careful thought.

v0.18.2: Multi-session stats in the TUI, Monitor and minimonitor handle cross-session agents correctly, and Agent crews that recover instead of erroring out

Sun, 26 Apr 2026 00:00:00 +0000

v0.18.2 is mostly a polish release for multi-session — stats picks up every session you’ve got, and the monitor finally stops mixing up which project a foreign-session agent belongs to. Agent crews and brainstorm sessions also got a lot more forgiving when something flakes.

Multi-session stats in the TUI

The ait stats TUI now picks up every aitasks session on your machine. A new Session panel on the left lets you cycle between them with ← / → or click to jump to one. There’s also a new sessions preset with a grouped bar chart comparing today / 7-day / 30-day activity across all of them — useful for getting a single read on where your time has been going.

v0.18.0: Your monitor and minimonitor now see every project at once, The TUI switcher teleports across sessions, and `ait install` is now `ait upgrade`

Fri, 24 Apr 2026 00:00:00 +0000

v0.18.0 is the multi-session release. If you keep two or three aitasks projects open in separate tmux sessions, every TUI now sees them all at once instead of pretending only the current session exists.

Your monitor and minimonitor now see every project at once

Both ait monitor and ait minimonitor now aggregate code-agent panes across every tmux session rooted in an aitasks project. No more switching sessions just to check whether the other project’s agent finished — it’s all one list, grouped by session under divider rows. Tap M in either TUI to flip back to single-session view if you want it.

v0.17.1: Import proposals straight into brainstorm, Lazygit with a built-in dashboard, and `ait setup` that tells you what's going on

Thu, 23 Apr 2026 00:00:00 +0000

v0.17.1 is a small but focused release: you can now bring your own proposal into the brainstorm engine, lazygit gets a dashboard companion, and ait setup is noticeably more trustworthy.

Import proposals straight into brainstorm

If you already have a markdown spec for a feature, you no longer have to paste it into a blank brainstorm session by hand. ait brainstorm init --proposal-file my_proposal.md now hands the file to a new initializer agent that reformats it into the brainstorm node format — structured sections, dimension metadata, the whole shape — then auto-starts the crew runner so you can jump straight into the interactive flow.

v0.17.2: Import a proposal when you start a brainstorm, and Fresh installs actually work now

Thu, 23 Apr 2026 00:00:00 +0000

v0.17.2 is mostly a stabilization release — a new way to kick off a brainstorm session, plus a handful of install and test-scaffolding fixes that have been biting people trying out the framework for the first time.

Import a proposal when you start a brainstorm

When you run ait brainstorm <N> on a fresh task, the init modal now gives you three choices: Blank, Import Proposal…, or Cancel. Picking Import opens a markdown file picker (filtered to .md / .markdown), runs an initializer agent over the file you chose, and applies its output to seed the first brainstorm node — so you can bring an existing design doc or proposal straight into the tree instead of starting from scratch.

v0.17.3: Task IDs now start at 1, Tmux targeting no longer confuses sibling projects, and `ait setup` adds `pycache/` to `.gitignore`

Thu, 23 Apr 2026 00:00:00 +0000

v0.17.3 is a small bug-fix release — three quality-of-life fixes aimed at people running the framework across multiple projects or starting fresh ones.

Task IDs now start at 1

New projects used to begin at t10 because of a buffer that made room for future renumbering. The buffer is gone: --peek returns 1, the first ait claim returns 1, and fresh projects finally look the way most people expect them to.

v0.17.0: Manual verification workflow, Stats TUI, and Shared section viewer across TUIs

Tue, 21 Apr 2026 00:00:00 +0000

v0.17.0 is a workflow-and-UX release: a brand-new manual-verification loop, a full interactive stats TUI, section-aware navigation across three TUIs, and a ground-up rewrite of the website and landing page.

Manual verification workflow

Some things you just can’t test with a bash assertion — you have to load the TUI, click through it, and see whether the row reorders. v0.17.0 gives those checks a proper home: mark a task with issue_type: manual_verification, list the items to verify, and /aitask-pick walks you through a Pass / Fail / Skip / Defer loop for each one. Failures become linked bug-fix follow-ups automatically, deferred items carry over into a fresh task on archival, and the archival gate makes sure you can’t forget half-finished verification runs.

v0.16.1: Claude Opus 4.7 is now the default, Fuzzy file search in the codebrowser, and Structured brainstorming

Sat, 18 Apr 2026 00:00:00 +0000

v0.16.1 ships with Claude Opus 4.7 as the new default and a pile of codebrowser TUI upgrades. Structured brainstorming lands too — you can now zoom brainstorm operations in on individual sections of a plan.

Claude Opus 4.7 is now the default

Opus 4.7 is registered in two variants — standard and 1M context — and the 1M variant is the new default for pick, explore, and all brainstorm ops. If you want the standard variant or need to swap models later, the new aitask-add-model skill registers models and promotes them to defaults with a single command, including dry-run diffs so you can see exactly what it will change.

v0.16.0: Interactive agent launch mode, File references on tasks, and Plan verification tracking

Wed, 15 Apr 2026 00:00:00 +0000

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.

v0.15.1: Scroll back through your agent's output, `ait ide` from a fresh shell just works, TUI switcher, and now documented

Mon, 13 Apr 2026 00:00:00 +0000

A quick follow-up to v0.15.0 with one notable feature, a shim fix, and the final piece of the TUI switcher docs.

Scroll back through your agent’s output

The monitor preview has been pretty tight until now — you saw the last few lines and that was it. v0.15.1 gives it real scrollback: mouse-wheel through the last 200 lines, toggle a scrollbar with b, or cycle to an XL preset that fills the whole terminal. It still follows the tail automatically, so you only lose the auto-scroll when you actually scroll up to read something.

v0.15.0: One-step startup with `ait ide`, Live monitor TUI for agent panes, and Minimonitor side panel

Sun, 12 Apr 2026 00:00:00 +0000

v0.15.0 is a big release centered on live tmux monitoring. Running several code agents in parallel is now a first-class experience, with two new monitor TUIs, a one-keystroke TUI switcher, and a single ait ide command to get everything going.

One-step startup with `ait ide`

Spinning up your workspace used to take four steps: open a terminal, cd into the project, start tmux, then start the monitor. Now it’s just ait ide. The new command creates (or attaches to) your project’s tmux session and opens a monitor window for you — whether you’re running it fresh outside tmux, inside an existing session, or on a second terminal to get another view of the same workspace.

v0.14.0: Browse Your Completed Tasks, Process Monitoring and Hard Kill, and Unified Launch Dialog with tmux Support

Sun, 29 Mar 2026 00:00:00 +0000

v0.14.0 is a big one — headlined by a full history browser, process monitoring, and a unified tmux-aware launch dialog across all TUIs.

Browse Your Completed Tasks

The codebrowser now has a history screen (press h) that lets you browse every archived task. Search and filter by labels, read the full task details and implementation plans, navigate to sibling tasks, and jump straight to the source files that were changed. It’s the fastest way to understand why code looks the way it does.

v0.13.0: Diff Viewer TUI, Brainstorm Engine & TUI, and Standalone QA Skill

Mon, 23 Mar 2026 00:00:00 +0000

v0.13.0 is a big one — the diff viewer is fully operational, brainstorming has its own TUI, and there’s a dedicated QA skill so you stop forgetting to write tests.

Diff Viewer TUI

You can now visually compare implementation plans side-by-side (or interleaved) with ait diffviewer. It supports classical line-by-line diffs and structural section-aware diffs, word-level highlighting so you can spot exactly what changed within a line, markdown syntax coloring, and a unified mode for comparing multiple plans at once. There’s even a merge screen where you can cherry-pick individual hunks from one plan into another.

v0.12.1: View Implementation Plans Right in the Board, and More Reliable Satisfaction Feedback

Wed, 18 Mar 2026 00:00:00 +0000

A smaller release this time with two quality-of-life improvements — one for the board UI and one under the hood for agent reliability.

View Implementation Plans Right in the Board

You can now toggle between viewing a task and its implementation plan directly in the TUI board detail screen. Hit v to switch views — the border turns orange so you always know which file you’re looking at. Editing is context-aware too, so pressing edit while viewing a plan opens the plan file, not the task.

v0.12.2: macOS Compatibility Fix

Wed, 18 Mar 2026 00:00:00 +0000

A quick patch release focused on macOS compatibility.

macOS Compatibility Fix

If you’re running aitasks on macOS, the codebrowser Python files now include future annotations so they work correctly with the system Python version. A small fix, but one less thing to worry about when setting up on a Mac.

Full changelog: v0.12.2 on GitHub

v0.12.0: AgentCrew: Run Multiple Agents in Parallel, Revert Any Completed Task, and Smarter Contribution Management

Tue, 17 Mar 2026 00:00:00 +0000

v0.12.0 brings multi-agent orchestration and the ability to undo any task you’ve ever completed. Two big additions that change how you work with aitasks.

AgentCrew: Run Multiple Agents in Parallel

You can now decompose a large task into subtasks and have multiple AI agents work on them simultaneously — each in its own git worktree. ait crew init sets up the session, ait crew addwork assigns subtasks with dependencies, and ait crew runner handles the rest: launching agents in the right order, monitoring heartbeats, and managing concurrency. There’s even a full TUI dashboard (ait crew dashboard) so you can watch everything happen in real time.

v0.11.0: Contribution Pipeline, Satisfaction Feedback & Verified Scores, and Explain Context

Sun, 15 Mar 2026 00:00:00 +0000

v0.11.0 is a big one — it introduces a complete contribution management pipeline, a satisfaction feedback system that tracks how well each AI model performs, and a bunch of board and settings TUI improvements.

Contribution Pipeline

You can now receive external contributions as GitHub/GitLab/Bitbucket issues and have them automatically checked for overlap with your existing tasks. CI/CD templates handle the automation, and a new contribution review skill walks you through analyzing, merging, and importing contributions. You can even merge multiple related issues into a single task or update an existing task with new contribution content.

v0.10.0: Contribute Back Without Forking, Code Agent Commit Attribution, and Code Agent and Model Statistics

Tue, 10 Mar 2026 00:00:00 +0000

v0.10.0 brings a major new contribution workflow, smarter commit attribution, and a bunch of quality-of-life improvements across the board.

Contribute Back Without Forking

The new /aitask-contribute command lets you open structured issues against upstream repositories directly from your local changes — no fork required. It works with GitHub, GitLab, and Bitbucket, and even parses contributor metadata when issues are imported back. If your project defines code_areas.yaml, you get hierarchical area drill-down to scope your contributions precisely.

v0.9.0: Gemini CLI and OpenCode Are First-Class Citizens, Model Discovery and Status Tracking, and Directory Rename: aiscripts to .aitask-scripts

Sun, 08 Mar 2026 00:00:00 +0000

v0.9.0 is a big one — full Gemini CLI and OpenCode support, a cleaner directory layout, and several workflow fixes that make multi-agent development smoother.

Gemini CLI and OpenCode Are First-Class Citizens

Both Gemini CLI and OpenCode now have complete skill and command wrapper sets, matching what Claude Code and Codex CLI already had. Run ait setup in any project and the framework automatically detects which agents you have installed, configuring each one with the right skills, permissions, and instructions. Gemini CLI commands also moved to TOML format with automatic permission policy merging, so setup is truly hands-off.

v0.8.2: Codex CLI Support, Unified Install Pipeline, and macOS Compatibility

Thu, 05 Mar 2026 00:00:00 +0000

v0.8.2 brings Codex CLI into the aitasks family — if you use OpenAI’s Codex CLI, your aitask skills now work there too.

Codex CLI Support

All 17 aitask skills now have Codex CLI wrappers. Run them with $skill-name syntax just like you would in Claude Code. A shared tool mapping file handles the translation between Claude Code and Codex CLI conventions, so skills behave consistently across both agents.

Unified Install Pipeline

Running ait setup now automatically detects which AI code agents you have installed and configures each one. Codex CLI gets its skills, config, and instructions assembled from a layered seed system. A new marker-based system (>>>aitasks/<<<aitasks) makes instruction injection idempotent — your existing config files stay clean, and aitasks content is neatly delimited and replaceable.

v0.8.3: Python-powered Stats with Charts, Codex CLI Gets Proper Guardrails, and Safer Task Ownership

Thu, 05 Mar 2026 00:00:00 +0000

v0.8.3 is a stability and polish release focused on making Codex CLI integration rock-solid and improving the stats experience.

Python-powered Stats with Charts

The ait stats command has been rewritten in Python, making it noticeably faster. Even better, you can now get visual charts right in your terminal with --plot — just enable the optional plotext dependency during ait setup.

Codex CLI Gets Proper Guardrails

If you’re using Codex CLI with aitasks, interactive skills now properly require plan mode before running. No more cryptic failures when Codex tries to prompt you mid-execution. We also fixed broken YAML in skill definitions and added agent attribution tracking across all remote/async workflows.

v0.8.0: Pull Request Import Pipeline, Settings TUI, and Code Agent Wrapper

Tue, 03 Mar 2026 00:00:00 +0000

v0.8.0 is a big one — three major features that change how you work with aitasks day-to-day, plus a ton of polish across the board.

Pull Request Import Pipeline

You can now import pull requests directly as aitasks. Run ait primport and point it at a PR from GitHub, GitLab, or Bitbucket — it creates a structured task with the PR metadata, contributor info, and a ready-to-go implementation plan. When you’re done and archive the task, the original PR gets closed automatically. Contributor attribution flows through to your commits too, so the original author gets credit.

v0.8.1: Works Without a Remote, and Smarter Update Checks

Tue, 03 Mar 2026 00:00:00 +0000

A small but important patch release fixing usability issues when working without a git remote and cleaning up the auto-update experience.

Works Without a Remote

You can now use ait create and task locking in repositories that don’t have a remote configured yet. The task ID counter runs locally and seamlessly upgrades to the remote-based atomic counter the moment you add a remote — no manual steps needed.

Smarter Update Checks

The auto-update notification no longer suggests “upgrading” to an older version. Version comparisons now use proper semver ordering instead of string comparison, so you’ll only see update prompts when there’s actually a newer release available.

v0.7.1: Code Browser TUI, Task Annotations at a Glance, and Smarter Explain Runs

Thu, 26 Feb 2026 00:00:00 +0000

v0.7.1 introduces the code browser — a brand new TUI for exploring your codebase with full task traceability — along with a batch of board improvements and developer experience fixes.

Code Browser TUI

The headline feature of this release is a full code browser you can launch with ait codebrowser. It gives you a file tree on the left and a syntax-highlighted code viewer on the right, complete with task annotation gutters that show exactly which tasks modified each line. Navigate with keyboard or mouse, select code ranges, and jump straight into the explain skill for deeper analysis.

v0.7.0: Run Tasks from Anywhere, Claude Code Web Support, and Full macOS Compatibility

Wed, 25 Feb 2026 00:00:00 +0000

v0.7.0 is a big one — this release makes aitasks work everywhere: on macOS, on remote servers, and even in Claude Code Web.

Run Tasks from Anywhere

The new /aitask-pickrem skill lets you run task implementation on remote servers, CI pipelines, or SSH sessions — completely hands-free. No interactive prompts, no fzf, just autonomous execution. Pair it with the new ait sync command to keep your task files in sync across machines, and the auto-merge engine handles any YAML frontmatter conflicts automatically.

v0.6.0: Code Explanations, Wrap Skill, and Auto-Refresh Board

Sun, 22 Feb 2026 00:00:00 +0000

aitasks v0.6.0 is out, and it’s a feature-packed release. Here are the highlights.

Code Explanation Skill

Ever wanted to document how a piece of code evolved over time? The new /aitask-explain skill generates structured code explanations with evolution tracking. Point it at a file or module, and it produces a narrative that captures not just what the code does, but how it got there — complete with data extraction pipelines and run management for iterative analysis.

v0.5.0: Code Review, Multi-Platform Support, and Documentation Site

Fri, 20 Feb 2026 00:00:00 +0000

v0.5.0 is the biggest release yet. Code review capabilities, support for all three major git platforms, and a proper documentation website.

AI-Powered Code Reviews

The /aitask-review skill brings structured code reviews to your workflow. Point it at a file, a directory, or your recent changes, and it runs a review using configurable review guides — sets of rules and patterns that define what to look for. It comes with 9 seed templates out of the box, plus Google style guides for 7 languages. Findings become tasks automatically, so nothing falls through the cracks.

v0.4.0: Auto-Bootstrap, Explore Skill, and Task Folding

Tue, 17 Feb 2026 00:00:00 +0000

v0.4.0 is a big one. It makes getting started easier, adds new ways to investigate your codebase, and gives you more control over how you organize tasks.

Auto-Bootstrap for New Projects

Setting up aitasks used to require downloading the installer manually. Now just run ait setup in any directory and it bootstraps everything automatically — the framework files, the task directory structure, all of it. One command, done.

Interactive Codebase Exploration

The new /aitask-explore skill is for when you have a vague idea and need to figure out the right approach. Point it at a problem area, and it guides you through an interactive investigation of your code — asking follow-up questions, exploring related files, and eventually creating a well-scoped task from what you discover. It even checks for existing tasks that might overlap with your idea and offers to fold them together.

v0.3.0: Atomic Task IDs, Task Locking, and Framework Updater

Fri, 13 Feb 2026 00:00:00 +0000

aitasks v0.3.0 is all about making multi-device and multi-developer workflows rock-solid.

Atomic Task IDs

Task IDs used to be assigned locally, which meant two people creating tasks at the same time could end up with the same ID. Not anymore. IDs now come from a shared atomic counter on a separate git branch, so every task gets a unique number no matter how many PCs are creating tasks against the same repo. Tasks start as local drafts and get their final ID when you commit.

v0.2.0: Documentation, Execution Profiles, and Changelog Generation

Thu, 12 Feb 2026 00:00:00 +0000

aitasks v0.2.0 lays the groundwork for a polished developer experience. Here’s what’s new.

Comprehensive Documentation

The project now ships with full documentation covering installation, command reference, Claude Code skills, platform support, and the task file format. Whether you’re setting up for the first time or looking up a specific command, everything is in one place.

Execution Profiles

Tired of answering the same workflow prompts every time you pick a task? Execution profiles let you pre-configure your answers. The built-in “fast” profile skips confirmations, uses your stored email, and jumps straight to implementation. Create your own profiles by dropping a YAML file in aitasks/metadata/profiles/.

Releases on aitasks

v0.32.0: Board columns you can actually manage, Grouped cards, and Reject a concern and it stays rejected

Board columns you can actually manage

v0.31.0: Mark cards and move them in bulk, The board got noticeably faster, and Every shortcut is visible again

Mark cards and move them in bulk

v0.30.0: The shadow companion, in the full monitor, Manage every linked repo from the syncer, and Triage that keeps up with you

The shadow companion, in the full monitor

v0.29.0: Implementation trails, Work reports, and A real setup wizard for chat intake

Implementation trails

v0.28.0: Report a bug from Discord, get back a ready-to-work task, Store and share artifacts with your team, Slack, and not just Discord

Report a bug from Discord, get back a ready-to-work task

v0.27.1: Better cross-repo task creation, A tidier board, and The shadow reviews your finished work

Better cross-repo task creation

A tidier board

v0.27.0: Attach files to your tasks, Quality gates that actually run, and Teach it a new skill

Attach files to your tasks

v0.26.1: Spin Up an Agent, No Task Required, and Upgrades That Survive GitHub Rate Limits

Spin Up an Agent, No Task Required

Upgrades That Survive GitHub Rate Limits

v0.26.0: Gate orchestration, end to end, Your terminals, on your phone, and Project groups

Gate orchestration, end to end

v0.25.0: Task gates that actually drive the workflow, Your phone can now talk to your workspace, and Meet the shadow companion

Task gates that actually drive the workflow

v0.24.0: See your proposal while you work, Your tmux sessions stop dying, and Smarter module decomposition

See your proposal while you work

v0.23.1: Installs that just re-run cleanly, and Less untracked clutter from generated skills

Installs that just re-run cleanly

Less untracked clutter from generated skills

v0.23.0: Brainstorm module decomposition, Risk evaluation in planning, Cross-repo planning, and straight from explore

Brainstorm module decomposition

v0.22.1: Track risk on your tasks, Risk evaluation, and when you want it

Track risk on your tasks

v0.22.0: Tasks that span multiple repos, Make every shortcut your own, and Manual verification that runs itself

Tasks that span multiple repos

v0.21.0: Profile-aware skill templating, Brainstorm: auto-apply, DAG navigation, and operation detail, and Cross-repo project registry

Profile-aware skill templating

v0.21.1: Choose-sibling picker in monitor, Brainstorm retry-apply actually retries, and Release-post YAML hardening

Choose-sibling picker in monitor

Brainstorm retry-apply actually retries

v0.20.3: Linux install pages unified, Curl-first install, with native packages as the alternative, and A page on updating model lists

Linux install pages unified

v0.20.1 Release

v0.20.2: A redesigned home page, Native installs for every platform, and Code browser History screen no longer crashes

A redesigned home page

v0.20.0: Install via your distro's package manager, A dedicated syncer TUI, and Opt-in PyPy fast-path

Install via your distro’s package manager

v0.19.2: Python wrappers that actually stay inside the venv, Warnings before you build on stale main, and Source-tree setup gets the starter tmux config too

Python wrappers that actually stay inside the venv

v0.19.0: A modern Python venv that installs itself, `ait setup` on a fresh clone, polished end-to-end, Brainstorm: fewer crashes, and more visibility

A modern Python venv that installs itself

v0.19.1: Releases and changelogs no longer miss remote tasks, and macOS test suite is back at parity with Linux

Releases and changelogs no longer miss remote tasks

v0.18.3: Brainstorm agents run interactive by default, Brainstorm sessions that don't get stuck, and MissedHeartbeat is gone (and that means a small migration)

Brainstorm agents run interactive by default

v0.18.1: Triage manual-verification checklists in batches, Filter the board by issue type, and `ait upgrade` actually upgrades branch-mode setups now

Triage manual-verification checklists in batches

v0.18.2: Multi-session stats in the TUI, Monitor and minimonitor handle cross-session agents correctly, and Agent crews that recover instead of erroring out

Multi-session stats in the TUI

v0.18.0: Your monitor and minimonitor now see every project at once, The TUI switcher teleports across sessions, and `ait install` is now `ait upgrade`

Your monitor and minimonitor now see every project at once

v0.17.1: Import proposals straight into brainstorm, Lazygit with a built-in dashboard, and `ait setup` that tells you what's going on

Import proposals straight into brainstorm

v0.17.2: Import a proposal when you start a brainstorm, and Fresh installs actually work now

Import a proposal when you start a brainstorm

v0.17.3: Task IDs now start at 1, Tmux targeting no longer confuses sibling projects, and `ait setup` adds `__pycache__/` to `.gitignore`

Task IDs now start at 1

v0.17.0: Manual verification workflow, Stats TUI, and Shared section viewer across TUIs

Manual verification workflow

v0.16.1: Claude Opus 4.7 is now the default, Fuzzy file search in the codebrowser, and Structured brainstorming

Claude Opus 4.7 is now the default

v0.16.0: Interactive agent launch mode, File references on tasks, and Plan verification tracking

Interactive agent launch mode

v0.15.1: Scroll back through your agent's output, `ait ide` from a fresh shell just works, TUI switcher, and now documented

Scroll back through your agent’s output

v0.15.0: One-step startup with `ait ide`, Live monitor TUI for agent panes, and Minimonitor side panel

One-step startup with ait ide

v0.14.0: Browse Your Completed Tasks, Process Monitoring and Hard Kill, and Unified Launch Dialog with tmux Support

Browse Your Completed Tasks

v0.13.0: Diff Viewer TUI, Brainstorm Engine & TUI, and Standalone QA Skill

Diff Viewer TUI

v0.17.3: Task IDs now start at 1, Tmux targeting no longer confuses sibling projects, and `ait setup` adds `pycache/` to `.gitignore`

One-step startup with `ait ide`