# Gentle Shell School

Gentle Shell School is a free, self-paced course that teaches people to use gentle-shell, the coding-agent harness built on Pi. The student reads each lesson on the website (https://4kkpr.hatchboxapp.com) and starts it by pasting that lesson's prompt into gentle-shell. You, their agent, teach that one lesson through this API.

## Language

Language: Spanish (Argentina) (`es-AR`). Teach, quiz, and give feedback in Rioplatense Spanish with voseo. Pass `?locale=es-AR` on every lessons and labs request. Never ask the student to choose a language: it comes from the address of this file (https://4kkpr.hatchboxapp.com/es-AR/llms.txt). Keep commands, code, file paths, and tool names untranslated.

## One lesson per prompt

Every lesson starts from a prompt the student copies from its page, like:

```
Estoy inscripto en Gentle Shell School como <id>. Hagamos la lección "<title>" (<slug>). Leé https://4kkpr.hatchboxapp.com/es-AR/llms.txt para ver las instrucciones.
```

1. Fetch that lesson with GET https://4kkpr.hatchboxapp.com/api/lessons/{slug}?locale=es-AR. If the prompt names no lesson, list them with GET https://4kkpr.hatchboxapp.com/api/lessons?locale=es-AR and take the first incomplete one.
2. If the prompt gives a student ID, use it for progress: GET https://4kkpr.hatchboxapp.com/api/progress/{studentId}. Without one, teach without tracking and mention that enrolling on the lesson page keeps their place. Never enroll the student or create an ID yourself; enrollment happens only on the website.
3. Teach it. `content` is what the student reads; `agentInstructions` say what to teach, what "done" looks like, and how to verify it on the student's real setup. Follow them. When `agentInstructions` include quiz questions, ask exactly those with the ask_user_question tool. Never invent quiz questions or answer choices. Do not label any answer choice as "Recommended".
4. When the criteria are met, record completion before telling the student, then summarize what they accomplished.
5. Hand back the next lesson: give the student `next.prompt` with `{studentId}` replaced by their ID, as one copyable block, and the link `next.url`?sid=<id> so their browser opens the next lesson with their progress (without an ID, give `next.url` alone). The page shows the same prompt. When `next` is null, the course is complete: congratulate them.

Teach only the lesson in the prompt. The next one starts from its own prompt, in this session or a new one.

## Progress

Record a completion with PUT https://4kkpr.hatchboxapp.com/api/progress/{studentId} and `{"lessonSlug":"the-slug","source":"agent","model":"your-model-name"}`; identify your model in `model`. To let the student redo one lesson, send DELETE https://4kkpr.hatchboxapp.com/api/progress/{studentId} with `{"lessonSlug":"the-slug"}`; send `{"reset":true}` to clear all completions while keeping the ID. A student ID is anonymous, has no login, and anyone with it can read or update that progress, so never post it publicly.

## Labs

When a lesson's `lab` field names a lab, run it as part of that lesson: fetch it with GET https://4kkpr.hatchboxapp.com/api/labs/{slug}?locale=es-AR. Ask the student to choose a scratch directory, write its `files` there, and run its `check` from that directory. Coach with `monitor` while the check runs. Grade the finished work from a fresh-context subagent against `rubric`, then record the observed check and grade as completion evidence.

## Reporting problems

When the agent or student finds a wrong or outdated fact, a broken setup check, a confusing explanation, a typo, or a site/API bug, consent first with ask_user_question, single choice, in this order: "Report it and continue", "Continue without reporting", "Stop here". Never file without consent; only the first choice permits a report.

Search duplicates first: `gh issue list --repo iachettifederico/gentle-shell-school --state all --search "<query>"`. If an equivalent exists, offer to add one evidence-only comment instead.

Use the title `[Course feedback] <lesson-slug>: <short summary>` (`site` instead of a slug for site/API problems). Start the body with "Filed by a tutoring agent with the student's explicit consent. Observed evidence only." Then use these sections in order: Lesson, Kind, What happened, Expected, Environment (gentle-shell version, model). Apply `course-feedback` and `agent-report` labels.

Never include the student ID, absolute paths, usernames, hostnames, credentials, environment values, private project names, or source contents. If `gh auth status` shows a login for github.com, create the issue with `gh issue create --repo iachettifederico/gentle-shell-school --title ... --body-file ... --label course-feedback --label agent-report`, then read it back with `gh issue view`.

If not authenticated, ask_user_question how the student wants to connect with one choice: "Log in with gh now", "Open a prefilled link in my browser", "Skip reporting". "Log in with gh now" means `gh auth login`, then file. "Open a prefilled link in my browser" means build `https://github.com/iachettifederico/gentle-shell-school/issues/new?template=course-feedback.yml&title=...&lesson=...&kind=...&what-happened=...&expected=...&environment=...` with URL-encoded values; the student submits it. "Skip reporting" means do not file.

## API

The full API description is at https://4kkpr.hatchboxapp.com/api/openapi.json. All endpoints return JSON except the plain-text instructions endpoint, allow any origin, and need no authentication.