Backlog.md
A Markdown-native task board that makes AI-assisted Git work reviewable before, during, and after implementation.
Per-dimension scores and reasoning
Evidence shows: tool is local-first, no telemetry, but least privilege not explicitly addressed; user confirmation exists (e.g., review checkpoints) but not systematic; data flow transparency limited; sensitive data handling not specified; dependency security not audited; external effects (e.g., Git operations) have config options but auto-commit may be default; rollback not specified; source attribution has author info but unverified. Deductions: lack of explicit least privilege, data flow transparency, and sensitive data handling details.
Evidence shows: extensive test coverage including unit and integration tests, but no test results provided; dependency availability not verified but multiple install methods; failure messages have concrete examples (e.g., error prompts). Deductions: dependency availability unverified, failure messages not comprehensive.
Evidence shows: clearly targets AI agents and humans, scenarios clear; capability boundaries described (CLI, Web, MCP); trigger precision high (task IDs, commands); environment fit good (cross-platform). Deductions: capability boundaries not fully exhaustive.
Evidence shows: clear information architecture (README, CLI reference, config docs); detailed install notes; stable naming (backlog command); rich examples and FAQ; known limitations mentioned (Apple Silicon); license clear (MIT); versioning and changelog not explicit; maintenance responsibility has author info but unverified. Deductions: versioning/changelog missing, maintenance responsibility unverified.
Evidence shows: high output usability (JSON, Markdown); high marginal value (addresses AI-era code review bottleneck); reasonable cost-benefit (free open source). Deductions: none.
Evidence shows: claims partially traceable (feature list), but lack independent verification; cross-source corroboration limited (community tools); fact-inference separation unclear. Deductions: claims lack independent verification, cross-source corroboration insufficient.
- Publisher identity unverified, treat with caution.
- Dependency security not audited, recommend checking for vulnerabilities.
- External operations like auto-commit need confirmation of default config.
- Versioning and changelog missing, affecting traceability.
What does this agent do, and when should you use it?
Backlog.md is a Markdown-native task manager and terminal Kanban visualizer for Git projects. It stores tasks as project-local Markdown files and manages them through a CLI, terminal board, local web interface, and optional MCP server. Its recommended AI workflow adds three review checkpoints: task specification, implementation plan, and code review, with one task intended for one agent session and one PR. `backlog browser` runs a drag-and-drop task interface bound only to 127.0.0.1 and keeps changes synchronized with the Markdown files. It can also be used as a standalone local task manager, including with `--no-git`.
backlog init "Project name" initializes a backlog in a repository, while backlog init "Project name" --no-git creates a filesystem-only project. Users create tasks with backlog task create, add descriptions, acceptance criteria, comments, and Definition of Done items with backlog task edit, and inspect work with backlog task list, backlog search, and backlog board. backlog board export produces shareable Markdown reports, and task list, task view, task <id>, and search provide stable machine-readable output through --json. backlog browser serves a local Kanban UI, while backlog mcp start exposes workflow guidance and project operations to configured MCP clients.
- An engineer using Codex in a Git repository who wants to approve acceptance criteria and an implementation plan before an agent starts coding.
- A technical lead maintaining a multi-contributor repository who needs tasks, dependencies, and milestones preserved as Git-trackable project records.
- A solo developer not using AI who wants to manage work from a terminal with
backlog task create,backlog search, andbacklog board. - A project team that wants to drag tasks across a local browser-based board and edit acceptance checklists while keeping Markdown files in sync.
- An integration developer who needs scriptable task listings or search results through the documented
--jsonread commands.
What are this agent's strengths and limitations?
- Tasks are ordinary Markdown files inside the repository, so task intent and history can be reviewed and committed alongside code rather than held in a hosted PM service.
- The workflow explicitly separates specification, implementation-plan, and code-review checkpoints, giving reviewers control before generated code expands the change.
- CLI commands, a terminal Kanban, exportable Markdown reports, and a local web UI all operate on the same task files.
- It documents CLI or MCP workflows for Claude Code, Codex, Gemini CLI, Kiro, and other MCP- or CLI-compatible assistants.
- Effective use depends on splitting work into tasks small enough for one session and having a human review each of the three checkpoints.
- The web UI is bound to 127.0.0.1, so it is not directly available to other devices on a LAN or VPN.
- MCP use requires client-side server configuration; manual setup also requires adding workflow-reading guidance to agent instruction files.
- Filesystem-only initialization disables active-branch checks, remote operations, and auto-commit, so those Git-oriented features are unavailable in that mode.
How do you install or deploy this agent?
No account or credentials are required for local use. Install with one of:
npm i -g backlog.md
bun add -g backlog.md
brew install backlog-mdOr run nix run github:MrLesk/Backlog.md -- --version. Then initialize in a Git repository with backlog init "My Awesome Project"; for a non-Git project, run backlog init "Personal Planning" --no-git.
How do you use this agent?
After initialization, an AI-assisted workflow can start with backlog instructions overview; ask the assistant to decompose a request into small tasks with descriptions and acceptance criteria, then write a plan into one selected task for review before implementation. For manual operation:
backlog task create "Render markdown as kanban"
backlog task edit BACK-1 -d "Detailed context" --ac "Clear acceptance criteria"
backlog boardRun backlog browser for the local web UI. For MCP, configure Codex with codex mcp add backlog -- backlog mcp start, or Claude Code with claude mcp add backlog --scope user -- backlog mcp start.
FAQ
Must I use an AI coding assistant?
Is Git required?
backlog init ... --no-git creates a filesystem-only project, but it disables active-branch checks, remote operations, and auto-commit.Will the web interface be exposed on my company network?
backlog browser listens on 127.0.0.1, and the README states that it is not reachable from other devices on a LAN or VPN.How do I connect Codex or Claude Code?
codex mcp add backlog -- backlog mcp start for Codex or claude mcp add backlog --scope user -- backlog mcp start for Claude Code.