You decide what gets built and what “done” means. Your agents do the work. Taskaat holds the state across sessions, agents and machines, and keeps the evidence of whether the work actually met your bar.
Claude Code · Cursor · Windsurf · Roo Code · Codex · Antigravity
Add MCP tools to archive and unarchive sections: write parity with the web app
The agent says it’s done. It hasn’t shown you the check passing.
The todo list it keeps is excellent, and it dies with the context window. The next session doesn’t know what was decided, what was already tried and rejected, or what “done” means here. The state that mattered is in a transcript nobody can review and nothing can query.
Close the window, switch machines, switch agents, come back in a week. The project, the decisions and the reasons they were made are still there, and still queryable.
Write down what “done” means before the work starts. Your agent runs that check and records the raw result, not a claim that it passed. Close it without that evidence and the task says so, permanently.
Over MCP, not a tab you remember to update. Your agent reads its next task and writes the result back. You never copy anything between a chat window and a tracker.
You write down what “done” means before the work starts. Your agent runs that check and submits what it actually observed, not an assertion that it passed.
Nothing is forced. Taskaat records rather than blocks, because a tool that blocks gets worked around. A task closed without passing evidence carries that mark for good, where you and anyone else can see it.
Fetch the live page and confirm the new copy is present.
HTTP 200 · “durable state” found at line 41
Close it without that, and it keeps this.
DONE (UNVERIFIED)Run the universal setup runner to connect Claude Code or Google Antigravity in seconds. 1-click browser authorization, zero API keys to copy, zero config files to open.
npx github:Xardoxis/taskaat-plugins # Or inside Claude Code: /plugin marketplace add Xardoxis/taskaat-plugins /plugin install taskaat@taskaat