# potato-skill > potato-skill teaches Claude Code, Codex and Cursor to design, build, run and check human annotation studies in Potato, the self-hosted annotation tool. These pages are the reference material the skill loads while an agent builds a Potato annotation task. Each link below is a page's markdown source; the HTML version is the same path without `.md`. - Announcement: https://www.potatoannotator.com/blog/potato-skill-claude-code-annotation-studies - Source code: https://github.com/davidjurgens/potato-skill - Potato: https://www.potatoannotator.com/ ## Pages - [Agent Instructions](https://davidjurgens.github.io/potato-skill/agents-md.md): The AGENTS.md from potato-skill: short instructions for building a Potato annotation task that Codex and Cursor read from the project root. - [Designing an Annotation Task](https://davidjurgens.github.io/potato-skill/designing-a-task.md): The decisions behind a task, before any YAML gets written. - [What to Confirm With the Researcher](https://davidjurgens.github.io/potato-skill/asking-the-experimenter.md): Which design choices to make yourself and which to hand back. - [Interviewing the Researcher](https://davidjurgens.github.io/potato-skill/interviewing.md): The multiple-choice question bank for a task described from scratch. - [Writing the Annotation Guidelines](https://davidjurgens.github.io/potato-skill/writing-guidelines.md): Label definitions, tie-break rules and the codebook an annotator reads. - [Recording Design Decisions](https://davidjurgens.github.io/potato-skill/recording-decisions.md): DESIGN.md, QUESTIONS.md, and estimating what a task will cost. - [Starting From a Published Design](https://davidjurgens.github.io/potato-skill/finding-a-design.md): The Potato Showcase, and what to change before reusing a design. - [Building the Annotation Interface](https://davidjurgens.github.io/potato-skill/building-the-ui.md): instance_display, the display types, and what a screenshot can prove. - [Evaluating and Iterating on the Interface](https://davidjurgens.github.io/potato-skill/evaluating-the-ui.md): How to judge a rendered annotation page and what to change. - [Images, Video, Audio and 3D](https://davidjurgens.github.io/potato-skill/modalities.md): Routing a medium to its display type, its scheme, and its quirks. - [Agent Traces and Agent Evaluation](https://davidjurgens.github.io/potato-skill/agent-traces.md): The trace displays, the twenty evaluation schemes, and how they bind. - [Phases and Pages](https://davidjurgens.github.io/potato-skill/phases-and-pages.md): Consent, instructions, training, surveys, and the order they run in. - [Attention Checks and Gold Standards](https://davidjurgens.github.io/potato-skill/quality-control.md): The quality-control ladder, and what each rung costs. - [Assignment and Agreement](https://davidjurgens.github.io/potato-skill/assignment-and-agreement.md): Who annotates what, and which agreement measure follows from it. - [Modes and Subsystems](https://davidjurgens.github.io/potato-skill/modes-and-subsystems.md): The whole-task modes a brief can be asking for without naming. - [Model Assistance From a Self-Hosted Endpoint](https://davidjurgens.github.io/potato-skill/model-assistance.md): Pointing ai_support at your own vLLM or other OpenAI-compatible server. - [Letting Models Label and People Certify](https://davidjurgens.github.io/potato-skill/labeling-engine.md): The 2.11 labeling engine: blind samples, contracts, and what to ask first. - [Data Files and Admin Access](https://davidjurgens.github.io/potato-skill/data-and-access.md): Input formats, side files, and reading results back out. - [Importing Existing Annotations, Transcripts and Corpora](https://davidjurgens.github.io/potato-skill/importing-existing-work.md): What to do when the researcher already has annotated data. - [Getting the Annotations Out](https://davidjurgens.github.io/potato-skill/getting-the-data-out.md): The export formats, what a CSV holds, and where consent answers go. - [Running a Task](https://davidjurgens.github.io/potato-skill/running-a-task.md): Booting a server, reading the startup log, and handing it over. - [After Annotators Start](https://davidjurgens.github.io/potato-skill/after-annotators-start.md): Watching a live study, what is still safe to change, and what is not. - [Deploying a Task](https://davidjurgens.github.io/potato-skill/deploying.md): Putting a task somewhere annotators can reach it. - [When a Task Misbehaves](https://davidjurgens.github.io/potato-skill/troubleshooting.md): Symptoms that validate clean and then do something else. - [Reporting a Bug or Asking for a Feature](https://davidjurgens.github.io/potato-skill/reporting-upstream.md): Telling a config mistake from a defect, and filing the ones that are real. - [Interrogating Potato](https://davidjurgens.github.io/potato-skill/interrogating-potato.md): Recovering an undocumented config block from the validator and the log. - [A Worked Example](https://davidjurgens.github.io/potato-skill/worked-example.md): One complete task, annotated, and what to change first. ## Optional - [Build Potato annotation studies with Claude Code, Codex or Cursor](https://davidjurgens.github.io/potato-skill/index.md): potato-skill is an open-source skill that lets a coding agent design, build, run and check human annotation studies in Potato. Install it in Claude Code, or in Codex and Cursor with the skills CLI. - [Every page in one file](https://davidjurgens.github.io/potato-skill/llms-full.txt)