academic-research-skills

A suite of Claude Code skills that covers research, writing, review, revision, and finalization, built on the rule that AI is a copilot, not the pilot.

The GitHub description summarizes the whole workflow in five words: research, write, review, revise, finalize. The README says the repository is a comprehensive suite covering the full pipeline from research to publication, and it highlights a plan command that walks through paper structure via Socratic dialogue. Python is the recorded language, which is how the skills themselves are packaged.

The README states the governing principle plainly: AI is a copilot, not the pilot, and the tool will not write the paper for the user. Instead it handles grunt work such as hunting down references, formatting citations, verifying data, and checking logical consistency. The framing is careful too: the tool is positioned as helping write better rather than helping hide the use of AI.

The README justifies the human in the loop design with research. It cites work on a fully autonomous AI research system that published a paper through peer review, and it describes the failure modes that fully autonomous pipelines inherit, including implementation bugs, hallucinated results, and citation hallucinations. The project is built on the premise that a human researcher augmented by AI avoids these failure modes better than either alone.

There are safeguards built in. The README mentions integrity gates that run a blocking checklist against known AI research failure modes, with an opt in calibration mode that measures its own error rates. It also cites an audit of millions of references that estimated hundreds of thousands of hallucinated citations, and notes the skills added trust chain frontmatter for source provenance and citation anchors. That attention to citation hygiene is what separates the project from a generic writing assistant.

The project's argument is that fully autonomous research pipelines have known failure modes, and the answer is to keep a human in the loop while using AI for the drudge work. The integrity gates and provenance tracking are worth borrowing on their own.

• Source 1 · Project repository
• Source 2 · README