This matrix compares the three task managers profiled in this section, feature by feature, so choosing between them does not require reading three notes. Everything below was verified against live sources on 2026-09-13.
Files versus database is the row that decides everything else: it determines whether your board survives multiple agents racing on it, and the one feature only the commercialized tool has, PRD ingestion, came bundled with a license change.
Legend: ✓ supported, ✗ not supported, ~ partial or conditional, ? not verified as of the date above. Each column links to the full research note; every cell below traces to a source cited there or in the references.
The matrix #
| Feature | Backlog.md | beads | Task Master |
|---|---|---|---|
| Kind | CLI, terminal and web kanban | Go CLI plus Dolt database | CLI plus MCP server |
| Storage | markdown task files in the repo | Dolt SQL in .beads/, JSONL export |
files under .taskmaster/ (tasks, config, state) |
| License | ✓ MIT | ✓ MIT | ~ MIT with Commons Clause |
| Task structure | tasks with acceptance criteria, DoD, milestones | hash IDs, epics, sub-tasks, message type | tasks with subtasks and dependency chains |
| Dependency graph | ~ milestones, no claim graph | ✓ graph plus auto-ready queue | ✓ dependency chains, move with them |
| Multi-agent concurrency | ✗ no atomic claims | ✓ atomic claim, hash IDs | ? not verified |
| PRD ingestion | ✗ manual task creation | ✗ manual task creation | ✓ parse-prd pipeline |
| Agent integrations | Claude Code, Codex, Gemini CLI, Kiro, Cursor, MCP | bd setup for Codex, Claude, Factory, Cursor, plus AGENTS.md and MCP |
MCP server and CLI, documented for Cursor, Claude Code, Windsurf, VS Code, Q CLI |
| Pricing | free | free | CLI free, Hamster $40 per creator per month |
| Current status | active, about 6.7k stars | active, about 27k stars | repo quiet since April 2026, product alive at Hamster |
Reading the matrix #
The storage row is the architecture decision: markdown files are diffable and agent-readable but race under concurrency, while Dolt gives beads cell-level merge and atomic claims at the cost of a database in your repo. Task Master’s file storage sits between the two but its concurrency story is unverified, which is the cell I would resolve first before adopting it for shared queues.
PRD ingestion is the pipeline feature, and its provenance is the warning: the only tool with it is the one whose license stopped being OSI open source and whose repo went quiet as the method moved into a paid product.
All three converge on meeting agents where they already are, MCP or AGENTS.md or editor config, so the integration row is nearly a tie and should not drive the choice.
Choosing from the matrix #
- Multiple agents sharing one queue: beads, for the atomic claims.
- Human review as the bottleneck: Backlog.md, for the three review gates.
- Want the PRD pipeline and accept the vendor: Task Master, pinned to a version you control.
Changes #
- 2026-08-27 - Created with three columns (Backlog.md, beads, Task Master) and ten rows, tracing every cell to the member notes.
See also #
- Spec Driven Development Feature Matrix - the process layer that fills these boards
- Control Planes Feature Matrix - what runs the agents once the queue exists
- Orchestration Feature Matrix - the parallel-session layer these trackers feed
- Agentic Coding Tools Landscape - the map these categories sit in
References #
https://github.com/MrLesk/Backlog.md - storage model, integrations, milestones for the Backlog.md column
https://tryhamster.com/pricing - the commercial tier behind the Task Master method