AI Agents Lose Track—This Free Claude Skill Fixes It

Someone reverse-engineered the core pattern behind Manus (the $2B Meta acquisition) and packaged it as a free Claude skill.

Turns out AI agents lose track of goals after multiple tool calls. Manus solves this with three markdown files that keep context clean:

• task_plan.md - checkboxes for tracking progress
• notes.md - research storage
• deliverable.md - final output

The agent reads the plan before each decision, so goals stay visible instead of getting buried in bloated context.

Install it in 10 seconds:

cd ~/.claude/skills git clone https://github.com/OthmanAdi/planning-with-files.git

Works with Claude’s CLI. MIT licensed. Pretty clever way to prevent task drift without complicated prompt engineering - just forces the agent to check a simple todo list before doing anything.