v0.4.0

What's New

IntelliJ Plugin

Full JetBrains IDE plugin for agent-task-queue visibility. Includes status bar widget, tool window with queue table, per-task output tabs with streaming output, notifications, display modes, settings, and optimistic UI updates on cancel/clear. Install locally via ./gradlew buildPlugin in the intellij-plugin/ directory.

Raw Output Files

Writes .raw.log companion files alongside standard log files for plugin streaming. Adds structured tool results to MCP output for better integration visibility.

Command Storage

Stores the command string in the database when using the MCP run_task tool, making it possible to see what command a task ran.

Stale Task Cleanup

Automatically cleans up stale tasks during polling, improving reliability when parent processes die unexpectedly.

Dependency Updates

• Upgraded fastmcp to 2.14.4
• Updated astral-sh/setup-uv and actions/checkout CI action digests

Bug Fixes

• Fixed output rotation to account for .raw.log companion files
• Fixed closeable output tabs and raw log cleanup scaling

Full Changelog

v0.3.0...v0.4.0

v0.3.0

What's New

JSON Output for CLI (tq)

Added --json flag to CLI commands for programmatic queue inspection:

# Get queue state as JSON
tq list --json
# {"tasks": [...], "summary": {"total": 1, "running": 1, "waiting": 0}}

# Get logs as JSON
tq logs --json
# {"entries": [...]}

# Clear queue without confirmation
tq clear --json
# {"cleared": 5, "success": true}

Command Tracking

Tasks now store and display the command being executed, so you can see what's actually running:

{
  "id": 1,
  "status": "running",
  "command": "./gradlew :app:assembleDebug",
  ...
}

Schema Contract Tests

Added comprehensive tests to ensure JSON output structure remains stable for programmatic consumers.

Full Changelog

v0.2.0...v0.3.0

v0.2.0

Changelog (0.1.4 → 0.2.0)

Fixed

• Orphaned task cleanup with PID reuse detection - Tasks from cancelled sub-agents no longer block subsequent tasks (#7)

  • Added is_task_queue_process() to detect if a PID is actually running task_queue vs being reused by an unrelated process
  • Added SERVER_INSTANCE_ID and _active_task_ids tracking in MCP server to detect orphaned tasks from disconnected clients
  • Added CLI_INSTANCE_ID tracking in tq.py for same protection
  • Handle asyncio.CancelledError to properly clean up when sub-agents are cancelled

• Ctrl+C leaving orphaned waiting tasks - Split registration from wait to ensure cleanup

Changed

• Refactored shared infrastructure - Extracted queue_core.py with shared database, logging, and cleanup logic
• Memory efficiency - Stream output directly to file with bounded deque
• Wait time calculation - Use POLL_INTERVAL_WAITING instead of hardcoded value

Added

• server_id column to queue table with migration for existing databases
• Signal handling integration tests for tq CLI

Security

• Pin pypa/gh-action-pypi-publish to commit SHA
• Add nosec B602 comments explaining intentional shell=True usage for CLI tool

v0.2.0b1 - Beta

Beta Release

This beta includes significant improvements to orphaned task cleanup:

New Features

• PID reuse detection: Detects when a PID has been reused by an unrelated process (Chrome, etc.) vs our actual task_queue process
• Instance tracking: Each server/CLI instance gets a unique ID to detect orphaned tasks from crashed/restarted processes
• Cancellation handling: Properly cleans up queue entries when MCP clients disconnect (e.g., sub-agents cancelled)

Infrastructure

• Extracted shared queue logic into queue_core.py for code reuse between MCP server and CLI

Installation

# Via uvx (recommended)
uvx agent-task-queue@0.2.0b1

# Via pip
pip install agent-task-queue==0.2.0b1

Note: This is a pre-release. Use @latest or no version specifier to stay on stable (0.1.4).

v0.1.4

Breaking Changes

• CLI renamed from atq to tq - The atq command conflicted with the Unix at(1) job scheduler. Update any scripts or aliases that used atq to use tq instead.

New Features

• tq run command - Execute commands through the same FIFO queue that AI agents use. This prevents resource contention when humans and agents run builds simultaneously.

  tq ./gradlew assembleDebug          # Run a build through the queue
  tq -q android ./gradlew test        # Use a specific queue
  tq -t 600 npm test                  # Custom timeout (seconds)
  tq -C /path/to/project make         # Set working directory

• Implicit run - tq echo hello works the same as tq run echo hello

• Rich terminal output - Passthrough mode preserves progress bars, colors, and interactive output from build tools

Improvements

• Database recovery - Automatically detects and recovers from corrupted SQLite databases
• Log rotation - Metrics log file rotates when exceeding 5MB to prevent unbounded disk growth
• Zombie cleanup metrics - Logs zombie_cleared events when cleaning up dead or stale tasks
• Graceful Ctrl+C handling - Clean process group termination without tracebacks

Testing

• Added comprehensive test suite for tq CLI (25 new tests)

v0.1.3

What's New

New Features

• atq CLI - Inspect the task queue from the command line
  • atq list - Show current queue status
  • atq logs - View recent task activity
  • atq clear - Clear stuck tasks

Documentation

• Improved agent configuration docs with compatibility table
• Added setup guides for Claude Code in examples/
• Clarified Bash allowed rules warning for Claude Code users

Install

uv tool install agent-task-queue

Or run one-off:

uvx --from agent-task-queue atq list

v0.1.2

---

Documentation improvements for timeout behavior.

Changes

• Clarify that timeout_seconds applies to execution time only, queue wait time does not count against the timeout
• Update tool docstring so agents understand this behavior

---

v0.1.1

---

Fixes uvx agent-task-queue installation.

Changes

• Add CLI entry point so uvx agent-task-queue works correctly
• Add author and license metadata to package

Installation

uvx agent-task-queue

---

v0.1.0

---

Initial Release

Agent Task Queue is an MCP server that prevents multiple AI agents from running expensive build operations concurrently.

Features

• FIFO task queuing - Serializes builds across all connected agents
• Multiple queue support - Isolate workloads with named queues
• No timeouts - MCP keeps connections alive indefinitely
• Zombie protection - Detects dead processes and clears stale locks
• Environment variables - Pass build-specific config via env_vars
• Configurable - CLI options for data directory, log rotation, timeouts

Installation

uvx agent-task-queue

Supported Agents

Works with any MCP-compatible tool: Claude Code, Claude Desktop, Cursor, VS Code/Copilot, Windsurf, Amp, Cline, Firebender, and others.

---