Set up a study project with an agent
Choose sources, access boundaries and checking rules before growing a learning workspace
Set up a study project with an agent by deciding what it may read and how the subject should be checked. The learning-init skill interviews you about trusted sources, conflicting accounts and what counts as a completed topic.
Setup has two phases. Initial decisions are tested through one real page using the companion writing capability, then revised in a second interview. Machine-assigned page labels reflect recorded sources and checks; they do not certify universal accuracy or your understanding. The complete package is required, rather than this interview file alone.
What it does
- Choose sources and an order of authority
- Set a clear file-reading boundary
- Define checks appropriate to the subject
- Distinguish independent sources from retellings
- Revisit settings after a real example
- Keep project decisions available for later sessions
Install and use
- Prepare the full tutor-skills plugin, companions and runtime in Claude Code, with Python3.10+ and a PyYAML environment. Select an authorised writable project and source boundary; the chosen agent service processes material it reads.
- State the subject and language. For example, request an introductory probability project from two permitted sources, with access limited to their folder. Accept or revise the recommendation attached to each setup question.
- Agree the sources and checking method. Executable checks need the subject’s tools; missing capabilities must remain visible. Scanned books may require separate text extraction or OCR tools.
- Use the companion capability to create one page, then revisit the initial settings. Complete the second interview and inspect checks and unresolved issues before expanding the project.
FAQ
Source
Made by Dmitry Lyapin. Original skill: learning-init, published on GitHub under the MIT licence.
View on GitHub