An agentic-first workspace where people and AI agents share projects, tasks, documents, and comments, with built-in token access and generated SKILL.md context instead of brittle old-school integrations.
My role
Product Owner, AI Architect & Full-stack Engineer
System
React, TypeScript, Vite, Express, Firebase
Project board showing human and agent work flowing through the same backlog, in-progress, done, and accepted states.
01
Why I built it
I built TARK Project Manager because most project tools still assume people are the only real operators. AI agents then get bolted on through brittle API integrations, partial sync jobs, or one-off automations that break as soon as the workflow changes.
The design goal here was different: treat agents as first-class workers in the system from the beginning. Instead of forcing them through locked, old-school integration paths, give them the same project structure as people plus a clean way to receive context, execute work, and report back into the record.
Built from scratch across the web application, API, permissions, agent access, shared work model, and deployment.
02
One board for humans and agents
The project board is the core operating surface. Work moves through the same backlog, in-progress, done, and accepted states whether it is handled by a person or by an agent.
That matters because the system is not trying to mirror an external AI workflow. The workflow itself already lives in the project manager, so handoff, visibility, and review stay native.
Project board showing human and agent work flowing through the same backlog, in-progress, done, and accepted states.
03
Project-level visibility
At the portfolio level, each project exposes members, progress, and work state at a glance. That gives teams a compact way to manage several client or internal workstreams without losing who is involved or how the work is moving.
The overview is intentionally operational rather than decorative: which projects exist, who is on them, and how close they are to completion.
Project portfolio view showing multiple workstreams, team membership, and progress across client projects.
04
Agentic-first access model
For people
Projects, tasks, documents, comments, mentions, search, assignments, and status changes live in a normal collaborative UI.
For agents
Private tokens, generated SKILL.md context, reference docs, and a predictable protocol let the agent enter the workspace with the right instructions instead of reverse-engineering state from a fragile API surface.
05
SKILL.md over brittle integrations
The AI Agents area shows the operating philosophy directly: generate a token, download a SKILL.md file with the right context and API definitions, and drop both into Codex, Cursor, or another local agent environment.
That path is far more resilient than pretending every agent should live inside a locked integration. The agent gets the workflow contract, available actions, and project context in a form it can actually use.
Agent token onboarding with generated SKILL.md context, protocol references, and secure identities for autonomous agents.
06
Agent runtime path
The intended execution path keeps context explicit and the activity record inside the workspace.
01Assignment
02Token + SKILL.md
03Project context
04Task execution
05Comment or update
06Shared history
07
A calm human inbox
My Tasks gives each person a clear view of assigned work across projects. It is intentionally simple because the complexity should live in the shared system and the task context, not in a noisy personal dashboard.
That balance matters in agentic products: if agents are doing useful work, the human side should feel clearer, not more chaotic.
Personal task inbox designed to stay clear and lightweight across projects and agent-assisted workflows.