nodecode-good-student

Good Student: wrong questions a parent confirms, analysed into weak points and a week of study actions. A Nodecode rewrite of harrylabsj/hermes-plugin-good-student by harrylabsj.

The model gets twelve good_student_* tools that keep one child's learning record in a single JSON file under this home: profiles, exam and homework scores, wrong questions waiting for a parent, the analysis they support, this week's actions, reassessments and a weekly brief. Plus /good-student to see the profiles, what waits, or one child's week. Nothing leaves the machine and there is no key.

The gate the original is built on is kept. Recognition output is a candidate: it sits in the waiting area, expires unconfirmed, and never touches the profile until a parent confirms it. One wrong question is suspected and never mastery; only an independent reassessment — a new variant, no hints — moves a knowledge point, past the pass line.

Get it

Tell the agent "add the good-student cell", or run nodecode add nc://good-student.

Set it up

Nothing is required. The records live in good-student/records.json under this home unless data_dir names another folder.

{
  "good-student": {
    "data_dir": "",          // empty: this home's good-student folder
    "pass_percent": 80,      // what an independent reassessment must reach
    "review_days": 14,       // how long a mastered point stands
    "candidate_hours": 24    // how long an unconfirmed candidate waits
  }
}

Member	Default	What it does
`data_dir`	none (this home)	the folder `records.json` is kept in
`pass_percent`	`80`	the accuracy an independent reassessment must reach
`review_days`	`14`	how long a mastered point stands before its review is due
`candidate_hours`	`24`	how long an unconfirmed candidate waits

Use it

Record Sam's maths test: 18 out of 20, and reading 8 out of 10.

Here are three wrong questions from this week's worksheet — put them in for me to confirm.

What should we work on this week?

He redid the fraction exercise with no hints: 5 out of 5.

The tools, as the model calls them:

Tool	Argument	What it does
`good_student_list_students`	—	every profile, with counts
`good_student_create_student`	`display_name`, `grade?`, `active_subjects?`, `goals?`	a new profile
`good_student_record_scores`	`student_id`, `records`	a batch of scores, whole or not at all
`good_student_list_scores`	`student_id`, filters	scores with a per-subject summary
`good_student_ingest_candidates`	`student_id`, `batch`	recognized wrong questions into the waiting area
`good_student_list_pending`	`student_id`	what waits for a parent
`good_student_confirm_questions`	`student_id`, `items`	a parent's yes or no, all or none
`good_student_analyze`	`student_id`	weak points with status, risk, confidence, evidence
`good_student_create_plan`	`student_id`	this week's actions, each with its reassessment day
`good_student_record_reassessment`	`student_id`, `kc_id`, `correct_count`, `total_count`	one reassessment, and the status it moves
`good_student_weekly_brief`	`student_id`	the parent's week, read-only
`good_student_delete_student`	`student_id`, `confirm_phrase?`	the scope, then the delete once the display name is given

/good-student shows the profiles and what waits; /good-student pending lists every waiting candidate; /good-student Ada is one child's week.

A knowledge point is named by its kc_id — the knowledge point's name lowercased with dashes — and that is what a reassessment is recorded against. A mastered point whose review date has passed goes back to review_due, and the plan gives it an action again.

The records are one records.json in data_dir, written atomically under the folder's lock, so two Nodecode processes can share one folder. Nothing else is written there.