dex CLI
A task tracker for agents: tasks with context and result live in JSONL beside the code and outlive sessions
- 405 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The CLI creates and edits task files in the project directory
- Anyone with repository access can read the tasks, so keep secrets out of descriptions
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/dexDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The CLI creates and edits task files in the project directory
- Anyone with repository access can read the tasks, so keep secrets out of descriptions
This is third-party code. Review the repository files before installing.
What it does
dex is a command-line task tracker for AI agents that keeps the work plan between sessions. Each task is shaped like a ticket: a short name like an issue title, a description with context and approach like an issue body, and a result like a pull request description. The agent splits big work into subtasks, marks progress and records results and decisions. Tasks live in a JSONL file, one per line, so changes show up cleanly in Git and rarely cause merge conflicts. Storage is configured in `.dex/config.toml`, with a sample in `dex.toml.example`. The CLI installs as the `@zeeg/dex` package, and the `dex` and `dex-plan` skills connect it to an agent. The `/dex` command manages tasks in plain language, and `/dex-plan` turns a planning document, such as a spec or a roadmap, into a set of tasks. The author distinguishes dex from agents' built-in task lists, such as Claude Code's, which live in one session only.
Who it is for. Developers who run multi-session work with an agent and want the plan, decisions and results kept in the repository instead of vanishing when the chat closes.
Good fit when
- A feature will span several days and the next session must know what is already done
- A spec or roadmap has to become a task list for an agent
- Decisions and results need to reach a colleague who opens the same repository
- The agent should pick the next logical task by itself and commit after it
Not a fit when
- The task is one-off and fits a single session: ticket overhead outweighs the benefit
- You need task IDs in commits and pull requests: the skill forbids referencing them outside dex
- The team needs a shared board with a manager UI: dex works in the terminal and in files
- The agent cannot run shell commands: dex is driven through the CLI
Example request
Use dex: work on the next logical task, then run code-simplifier and commitLimitations
The repository was last changed in March 2026. Both the CLI and the skills are needed: the agent will not use the `@zeeg/dex` package without the skills. The skill expects to run via `dex` or `npx @zeeg/dex` when the command is not on `PATH`. Task IDs are short-lived: by the skill's rules they are not cited in commits and documents because they lose meaning once a task is done. The JSONL file lives in the repository, so anyone with access to the code can read it.
How to disable. Remove the CLI with `npm uninstall -g @zeeg/dex`, the `dex` and `dex-plan` skill folders from the agent's skills directory and, if you like, the `.dex/` directory from the project.
FAQ
How does dex differ from Claude Code's built-in tasks?
Built-in tasks live for one session, while dex keeps them in the project's JSONL file. The next session sees earlier descriptions, decisions and results.
What format does dex store tasks in?
JSONL, one task per line. That format diffs well in Git and merges without conflicts when several people add tasks at once.
How do I turn a plan into tasks?
Give the `/dex-plan` skill a document such as a spec or a roadmap and it will create a set of tasks with descriptions. The agent then picks them up through `/dex`.
Which agents are supported?
The description names Claude Code, OpenCode, Codex and Cursor. For Claude Code the repository has a plugin marketplace manifest and `/dex` and `/dex-plan` commands.
Where is the full documentation?
At dcramer.github.io/dex. It covers CLI commands and epic examples, and a short version sits in `docs/public/llms.txt`.
Discussion
No questions or reviews yet.
Related
CLI toolsA personal AI chief of staff on a notes folder: daily plan, meeting prep, tasks and people context in Claude Code and Cursor.
A SQLite issue tracker for AI agents: sessions with handoff notes, dependencies, templates and Claude Code hooks
A local task board with the taskctl CLI and a skill for Codex, Claude Code, and other agents: creates issues, moves statuses, and keeps comments under a fixed w
Yandex Tracker CLI
tracker
The tracker command-line tool for Yandex Tracker with JSON output and a ready SKILL.md workflow for agents
An MCP server and CLI that find in Claude Code and Codex sessions why a file became what it is: who asked the agent what, when and what changed
A Yandex Tracker CLI built for LLM agents: issues, comments, worklogs, checklists and bulk operations, with an agent skill
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| dex (this card) | no | 405 | ||
| Beads | no | 28,540 | the `bd` tracker with a dependency graph on Dolt, while dex stores tasks in plain JSONL | |
| Taskmaster | no | not required | 28,310 | splits a PRD into tasks and works through MCP, while dex manages tasks with skills and a CLI |
README in short
The short description explains why structured tasks beat plain lists and shows a quick start: CLI and skills install, a sample request and a slash command. The rest is a developer section with build commands, local plugin testing and releasing versions.