The meta-skill that improves all your skills (including itself) and proposes new skill candidates from your work sessions.
Also known as “One skill to rule them all”, task-observer works for human-led AI sessions, as well as autonomous agents.
Over the first seven months of using it, task-observer logged 1200 observations across my 70 skills, the vast majority of which were turned into skill improvements.
Most of my skills were themselves created based on skill creation candidates that task-observer automatically logged during my work sessions.
Get started with task-observer to make your skills self-improving and to automate your skill creation process
The best way to get started with task-observer is to point the AI companion of your choice at the repo and ask it to guide you towards the ideal implementation for your specific setup:
Task-observer works in many AI environments
I’m a knowledge worker, and I use task-observer as the practical implementation of my Augmented Expertise methodology. My main environment is Claude Cowork.
However, I have also had feedback from users who integrated task-observer into their Hermes and Openclaw agent setups, or connected it to Obsidian. There is an adopted version for Codex that I link to from the main GitHub repo for task-observer. I will link to more versions as I discover them.
Your contributions are welcome
If you have feedback on task-observer, please feel free to submit an issue or a pull request to the repo. I regularly use reports from real users to further improve the project and task-observer wouldn’t be half as good as it is today without those contributions.
Let’s stay in touch
Please join my mailing list to receive updates about task-observer, Augmented Expertise and my open-source skill library.