Releases: mblua/AgentsCommander
Agents Commander v0.9.16
Agents Commander v0.9.16
19 merged PRs since v0.9.15 — config-folder seeding, in-app npm update awareness, runtime logging controls, and a batch of coordinator / Resource Monitor lifecycle fixes.
Features
- Live log-level selector + logging cleanup (#612) — choose runtime log verbosity in-app; noisy logs are downgraded rather than deleted.
- npm update notification (#609) — get notified in-app when a newer npm version is available.
- Config-folder seeding at spawn (#598) — seed a default config folder (e.g.
.claude) into a coding-agent replica when it spawns. - Profile content-hash drift detection + command concatenation (#592, #597) — detect when a profile's content drifts and compose coding-agent commands.
- MANUALLY-CLOSED coordinator badge + cascade close (#588) — new badge plus a Settings-gated option to cascade-close a coordinator's team.
- Resource Monitor: maximize + central-pane view (#587) — maximize the detached Resource Monitor window or view it as a central pane.
- Resource Monitor: activity filters + Agents rename + close-on-exit (#566) — filter by activity, "Groups" renamed to "Agents", and close-on-exit handling.
- Per-coding-agent profile name overrides (#548).
- Restart-failure toast (#574) — a reusable toast surfaces previously-silent restart failures.
- Unified coordinator idle clock (#580) — the idle badge and team auto-close now share one clock.
- AC_REPLICA_ROOT rename + new placeholders (#576) —
AC_REPLICA_ROOTrename, workspace/matrix tokens, and opencode config-dir auto-create. - Smarter replica assignment (#551) — "Assign to this replica" is disabled when the replica already uses that Coding Agent + Profile.
Fixes
- Profile env-var focus loss (#614) — env-var fields keep focus while typing.
- Wake-spawn idle gate (#611) — require sustained idle before wake-spawn PTY injection.
- Window-title prefix (#606) — shortened the taskbar window-title prefix to "AC".
- Orphaned auto-close member (#582) — reap an auto-close member orphaned when its coordinator is destroyed.
- Stale AUTO-CLOSED pill (#589) — clear a stale/wrong AUTO-CLOSED pill on a live coordinator.
- Restart-prompt error surfacing (#573) — surface a failed post-assign "Restart now" in the prompt modal.
Chore
- Sync
role-skill-boundary-auditskill seed (#602).
Full changelog: v0.9.15...v0.9.16
Agents Commander v0.9.15
Agents Commander v0.9.15
First release since v0.9.0 — bundles 18 merged PRs (9 features, 9 fixes), headlined by the new per-coordinator idle badge + configurable team auto-close, plus the new Resource Monitor and a wave of resource-monitor reliability/latency fixes.
Highlights
- #552 — Coordinator idle badge + configurable team auto-close: each coordinator now shows a color-coded idle badge (minutes since last activity), and teams can be configured to auto-close on inactivity.
- #516 — Resource Monitor: new live view of active agent process groups and their resource usage.
- #384 — Coding-agent profiles: configurable per-agent coding-agent profiles.
Features
- #552 — Coordinator idle badge (color-coded minutes) + configurable team auto-close on inactivity
- #565 — Configurable max-concurrent agent cap
- #558 — Root Agent write access + authority model
- #545 — Context-menu gray/red session-state coloring
- #516 — Resource Monitor (active process-group view)
- #515 — Sidebar regex filter
- #508 — Visible label for the selected workgroup
- #519 — Context template optimization
- #384 — Coding-agent profiles
Fixes
- #564 — Resource Monitor: reduce restart/stop/assign latency (kill_group wait)
- #559 — Resource Monitor: fix ACTIVE GROUPS eviction/counter leak
- #543 — Resource Monitor: handle PID reuse with unreadable root
- #563 — Fix hover-preselect layout shift
- #542 — Remove dead "new session" path and Ctrl+Shift+N
- #538 — Profiles: always show the command box
- #537 — Config write retry on transient failure
- #533 — Fix OpenCode prompt WG-agent diff badge
- #526 — Config selection UI fidelity
Assets
Prebuilt binaries for Windows, macOS (Intel + Apple Silicon), and Linux are attached below, with SHASUMS256.txt for verification.
Windows: Agents.Commander_0.9.15_x64-setup.exe (installer) or agentscommander-windows-x86_64.exe (standalone).
Full commit range: v0.9.0...v0.9.15
Agents Commander v0.9.0
See the assets below to download and install Agents Commander.
Agents Commander v0.8.51
See the assets below to download and install Agents Commander.
v0.8.50
AgentsCommander v0.8.50 focuses on steadier workgroup selection, a preview Spec Board for Mermaid/spec workflows, and smaller reliability and UI polish improvements.
Added
- Always show selected workgroup. New opt-in setting to keep your selected workgroup visible.
- Spec Board preview. New opt-in detached Spec Board window for Mermaid diagrams and spec work, with file open/save, snapshot history, live updates, and agent prompt integration. This is disabled by default and can be enabled with the
specBoardEnabledsetting.
Changed
- Telegram action polish. The sidebar Telegram action now uses an inline icon instead of a letter while keeping the existing hover, active color, tooltip, and click behavior.
Fixed
- More reliable workgroup selection. The selected workgroup now updates correctly across active agents and is preserved when there is no active match.
- Clearer workgroup deletion diagnostics. Delete-blocker reporting now distinguishes live sessions, ignored exited records, external blockers, and unidentified Windows locks, with raw delete and Restart Manager errors preserved in logs for troubleshooting.
- More reliable contributor installs. CLI npm installation is safer when replacing files that need executable permissions.
- Safer agent discovery. Discovery now guards against delete-sentinel artifacts affecting repository state.
Full Changelog: v0.8.47...v0.8.50
npm install
AgentsCommander is published on npm as @mblua/agentscommander@0.8.50.
npm install -g @mblua/agentscommander
agentscommander --helpThe npm package is scoped as @mblua/agentscommander; the installed CLI command remains agentscommander. Do not use npm install -g agentscommander, because npm rejected the unscoped package name as too similar to agents-commander.
v0.8.47
Added
- Role Experiments: Implemented CLI phases 1 through 3a, introduced dry-run artifacts, and formalized the report JSON contract.
- Telegram: Added configuration settings for network poll logging.
Fixed
- Root Agent: Resolved a startup failure that occurred when no coding agent was present, and ensured dormant root agents are preserved during launch configuration errors.
- Messaging: Fixed the
senddelivery confirmation mechanism. - Role Experiments: Fixed resume and metadata validation.
Removed
- CLI: Removed the sidecar binary entirely.
v0.8.45
This release resolves several identity verification issues that prevented agents from connecting in complex workgroup setups, removes legacy .ac-new workspace support, and improves the reliability of the wake command.
Added
- Test coverage for cross-workgroup peer discovery to prevent future regressions.
Changed
- Documentation: Clarified OS support status in the README and updated author social links.
Fixed
- Fixed an issue where the
wakecommand failed to resolve agent names correctly (#387). - Fixed identity trust issues that caused Codex agents to reject valid workgroup replicas and production releases (#380, #388).
- Fixed an issue that caused workgroup coordinators to be rejected due to stale identity references (#370).
- Fixed a bug where valid filesystem paths in team agent references were incorrectly rejected (#373).
Removed
- Removed legacy support for
.ac-newworkspaces. All workspaces must now use the.acfolder structure (#383).
Agents Commander v0.8.44
v0.8.44
Patch release after v0.8.43. This release adds CLI automation for workgroups and team membership, hardens task update events and session restore behavior, fixes Agent Matrix context handling, and expands public documentation around coordination, Telegram, screenshots, and the roadmap.
Added
- Added
workgroup list,workgroup add, andworkgroup removeCLI commands for registered projects, including auto-numbered workgroup creation,TASK.mdandmessaging/provisioning, replica creation, repo assignment flags, project refresh requests, and removal safeguards for live sessions and dirty repos (#241). - Added
team list,team add-member, andteam remove-memberCLI commands for managing team membership inside workgroups, including replica creation/removal, optional coordinator assignment for added members, repo assignment cleanup, and reachability coverage (#241). - Added default context metadata to newly created Agent Matrix config files, listing
$AGENTSCOMMANDER_CONTEXTandRole.md(#356).
Changed and Improved
create-agentnow creates a full Agent Matrix in a registered project and delegates to the same Matrix creation path ascreate-agent-matrix, including--description, optional--role-template, and the shared JSON result shape (#309).create-agent-matrix --projectnow resolves registered project folder names fromsettings.projectPathsand rejects direct path input, ambiguous matches, and unregistered project names before writing (#309).- Workgroup numbering is now allocated globally per project as the lowest free positive integer across teams, and deleted numbers can be reused (#241).
- Team configs are normalized to project-relative
_agent_*references, while the Edit Team UI continues to resolve older absolute references for compatibility. workgroup_task_updatednow uses a unifiedworkgroupRootfield and an explicitsource: "manual" | "poll"discriminator, with frontend routing based on the discriminator instead of payload shape (#306, #307, PR #337).task_ops::performnow returns post-edit task content and title from inside the write lock, so immediate task update emits no longer need an unlocked follow-up disk read (#308, PR #334).
Fixed
- Prevented removed-project sessions from being restored by purging persisted sessions whose working directories are outside current
projectPathsduring startup, settings updates, and session persistence (#310). - Fixed direct Agent Matrix launches that omitted the matrix agent's local
Role.mdfrom generated provider context (#356). - Fixed CLI team agent reference handling so absolute project paths are not persisted in team config, and repo assignment agent lists are normalized consistently.
- Improved project refresh path matching for Windows extended-length path prefixes.
Documentation
- Added an initial public README screenshot and removed stale/missing screenshot and GIF references (#343, PR #344).
- Added a demo and GIF capture checklist, including the role customization inheritance story (#345, PR #346).
- Updated README and home-page copy to mention Root Agent / Agent Commander coordination across workgroups (#347, PR #348).
- Updated public positioning for multiple teams and workgroup instances, and linked roadmap items for coordinator auto-handoff (#349) and Telegram coordinator routing (#350) (#351, #352, PR #353).
- Added cron-based scheduled executions to the roadmap and linked the implementation issue (#354, PR #355).
- Clarified the Telegram bridge security model, including
chat_idbinding, the Test and Save setup flow, private chat versus group behavior, and bot token handling (#359, PR #360). - Expanded CLI and teams/workgroups docs for the new workgroup/team commands, project-scoped numbering, repo assignment flags, task file editing, and deletion safeguards (#241).
Internal
- Added focused Rust and frontend tests for workgroup/team CLI flows, Matrix project resolution, task event races, direct Matrix context, session purge behavior, and project refresh path normalization.
- Removed stale legacy
BRIEF.mdassumptions from CLI tests and updated internal task panel comments from BRIEF naming to TASK naming. - Bumped package, Cargo, lockfile, and Tauri metadata from 0.8.43 to 0.8.44 (#361).
AgentsCommander v0.8.43
AgentsCommander v0.8.43
This is the first public release after the visible v0.4 line, and AgentsCommander has grown into a much fuller product.
AgentsCommander is a Tauri, Rust, and Solid terminal session manager for coordinating AI coding agents. It launches Claude Code, Codex, and Gemini as real OS processes in real PTYs, keeps coordination state in plain files, and gives you a visible desktop UI for supervising several agents working in real repositories.
Highlights Since v0.4
Agent Matrix
Agents now have a canonical Agent Matrix under .ac-new/_agent_<name>/. The matrix keeps the role definition, memory, plans, skills, inbox, and outbox in predictable directories, while workgroups run isolated __agent_<name> replicas for the active task. That makes long-lived agent identity separate from per-task scratch state.
Teams, Workgroups, and Tasks
You can group agents into Teams with a coordinator and one or more workers. Activating a Team creates a workgroup for a specific task, including a TASK.md brief, a shared messaging/ folder, and per-agent replicas.
The task workflow now has dedicated CLI support for coordinator-owned brief updates through task-set-title and task-append-body, with frontend and backend refresh paths for keeping the UI aligned.
File-Based Inter-Agent Messaging
Inter-agent coordination is now based on Markdown files, not hidden orchestration state. Agents write timestamped messages into a workgroup messaging/ directory, then call send --send <filename> --mode wake. The recipient gets a short terminal notification and reads the full payload from disk, so large messages are inspectable and not limited by PTY injection size.
The CLI also includes lean peer discovery, routing validation, coordinator-only operations, remote housekeeping commands, and Root Agent messaging to verified workgroup coordinators.
Coding-Agent Integrations
AgentsCommander has first-class profiles for Claude Code, Codex, and Gemini. The app handles agent-specific resume tokens, idle detection, transcript readers, and launch behavior while still letting you choose the coding agent per session.
Credentials are now delivered to managed agent sessions through environment variables instead of visible terminal text. Generated agent context also explains the current AgentsCommander CLI rules, messaging protocol, and write boundaries directly inside the session.
Role Templates and Skills
The New Agent flow can use role templates from the vendored Agents Agency catalog or from local templates in your config directory. Templates can enrich Role.md, provide metadata, and copy template skills into the new Agent Matrix.
Runtime skill discovery now resolves skills from the canonical Agent Matrix, so replica sessions can load the same role-owned workflows without treating replica scratch as the source of truth.
Portable Instances and Settings
AgentsCommander is now designed for portable, side-by-side instances. A renamed binary gets its own config directory, settings, sessions, ports, and instance label, so you can run isolated environments such as production, staging, or team-specific setups from separate copies of the app.
Settings and project registration also have expanded CLI coverage, including open-project, new-project, list-sessions, and stricter version-sync tooling for release maintenance.
Telegram, Voice, and Remote Workflow Support
The Telegram bridge can stream session output to a chat and send replies back into the session. It also has coding-agent-aware transcript readers for Claude Code, Codex, and Gemini, plus a telegram-send-image CLI for sending local images or files through a configured bot.
Voice-to-text support lets you dictate prompts into an active session through the Gemini API, with optional auto-execute. RTK integration is documented for Claude Code command-output compression.
Documentation and Public Project Cleanup
The public docs were rebuilt around the current product: README, quickstart, concepts, Teams and workgroups, inter-agent messaging, CLI reference, settings reference, troubleshooting, security, integrations, feature guides, roadmap, changelog, issue templates, and PR template. Public copy now uses the current naming, supported-agent list, unified-window model, and Teams terminology.
New in v0.8.43: create-agent-matrix
This release adds a new CLI command for creating full Agent Matrices from automation:
agentscommander create-agent-matrix \
--project "C:\path\to\project" \
--name "dev-rust" \
--description "Rust backend engineer" \
--role-template "agency:dev-rust" \
--launch codexcreate-agent-matrix creates the same full layout as the UI path, including Role.md, canonical Matrix directories, config.json, optional role-template resolution, best-effort template skills/ copy, settings-file updates, and optional launch request support.
It also hardens automation behavior:
- Invalid role templates fail before creating the target Matrix directory.
- The target Matrix root is claimed atomically to avoid concurrent overwrite races.
- Blank or unknown launch requests warn without failing the creation.
- Generated Codex and agent context now allows explicit AgentsCommander CLI operations through
AGENTSCOMMANDER_BINARY_PATH, while keeping arbitrary shell writes and hand-written scripts outside the normal write zones forbidden. - Successful CLI-created matrices write a project refresh request for the running app to consume.
Known Caveat
Sidebar refresh after CLI-created agents has backend and frontend support in this release, and the shipped artifacts include that support. The event contract and store/listener coverage were verified, but the visual sidebar refresh was not fully verified in a live branch GUI before release. If a newly CLI-created agent does not appear immediately in the sidebar, reload the app or project view.
Downloads
Choose the asset for your platform from the files attached to this release.
Windows
Agents.Commander.New_0.8.43_x64-setup.exefor the standard installer.Agents.Commander.New_0.8.43_x64_en-US.msifor MSI-based installation or deployment.
Linux
Agents.Commander.New_0.8.43_amd64.AppImagefor a portable Linux app.Agents.Commander.New_0.8.43_amd64.debfor Debian and Ubuntu-based systems.Agents.Commander.New-0.8.43-1.x86_64.rpmfor RPM-based systems.
macOS Apple Silicon
Agents.Commander.New_0.8.43_aarch64.dmgfor Apple Silicon Macs.Agents.Commander.New_aarch64.app.tar.gzif you prefer the archived app bundle.
macOS Intel
Agents.Commander.New_0.8.43_x64.dmgfor Intel Macs.Agents.Commander.New_x64.app.tar.gzif you prefer the archived app bundle.
Agents Commander v0.4.25
What's New in v0.4.25
Bug Fixes
- Fixed false-positive credential injection: the PTY read loop was triggering on rendered text containing
%%ACRC%%(tool calls, prose). Now uses line-based matching to only detect the marker when it appears as an independent line - Fixed cross-buffer ACRC detection: retained partial lines from the last newline to prevent the marker from being split across read buffers
- App now launches the GUI by default and enforces a single-instance guard to prevent duplicate windows
Full Changelog: v0.4.24...v0.4.25