TaskTrack is a Specify spec for managing structured work plans executed by AI agents.
This specification is a draft. It may still change in backwards-incompatible ways between revisions.
TaskTrack builds on the upstream standards contained in SpecPack:
- MiniMark — the restricted markdown subset used for all TaskTrack files.
- riVer — the versioning scheme used for the specification
- Specify — conventions for AI-consumed specifications.
The TaskTrack tracked resource resides in the spec folder.
The golden source lives at spec/TaskTrack.md.
The filename of a resource publications follows the pattern TaskTrack-[REVISION_IDENTIFIER].md.
A TaskTrack plan is a collection of tasks grouped into ordered phases.
Phases are executed in sequence such that all tasks in one phase must have been resolved before the next phase can start. Furthermore, tasks in one phase can have dependencies on each other informing the task execution sequence within a phase.
Each TaskTrack plan is stored as a collection of plain markdown files in a single folder.
For more details and additional information please refer to the TaskTrack specification.
A minimal plan folder looks like this:
my-plan/
├── TaskTrack-v1_draft_20260527T120000Z.md # publication of the spec
├── MYPL-TaskTrack.md # overview file
├── MYPL-1.md # one task file per task
├── MYPL-2.md
└── MYPL-3.md
The overview file (MYPL-TaskTrack.md) lists every task grouped by phase together with its current status:
# MYPL - My Plan
**Objective**: Demonstrate a minimal TaskTrack plan.
**Created**: 2026-05-27T12:00:00Z
**Resources**: none
## 1: Foundations
* [DONE] MYPL-1: Set up the project skeleton
* [BUSY] MYPL-2: Add the initial data model
## 2: Features
* [OPEN] MYPL-3: Wire up the first end-to-end flowEach task file (MYPL-1.md, MYPL-2.md, …) carries the task's status, dependencies, description, acceptance criteria, and — once finished — a resolution or failure note. See the specification for the exact schema.
Point an AI agent to
- a TaskTrack specification file,
- a PRD or any other document defining the work,
- a file system directory,
and instruct the agent to author a TaskTrack plan for the given work in the given directory.
With an authored TaskTrack plan in place, instruct an AI agent to execute the plan following the rules of the TaskTrack specification.
TaskTrack defines a plan execution run as a loop executing one task processing run at each pass. This opens up the possibility to assign one agent to the overall plan execution and a separate agent to each task processing for advanced context management.
Issues and pull requests are welcome. For substantive changes to the specification, please open an issue first to discuss the proposal.
Published specifications are licensed under CC BY 4.0. See LICENSE for the full text.
Copyright (c) 2026 New Adventures in IT and TaskTrack contributors.