# Desk Tutor Marks

> Desk tutor's marks

- Skill: `autonomous-ai/desk-tutor-marks` (Agent Skill, multi-file: 5 files)
- Install (CLI): `npx get-skill add autonomous-ai/desk-tutor-marks`
- Raw SKILL.md: https://api.skillmd.com/api/skills/autonomous-ai/desk-tutor-marks/raw
- Safety review: pending (external: skill-scanner PASS, skillspector PASS)
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: autonomous-ai (https://skillmd.com/u/autonomous-ai)
- Updated: 2026-11-12
- Page: https://skillmd.com/skills/autonomous-ai/desk-tutor-marks

---


# Desk tutor's marks

Desk tutor's fourth step, from the owner (#941): "a small nod where a step
is right, a curious tilt where it's wrong". Desk tutor (`desk-tutor`) leans
over the page, lights it and takes one look; only the model knows, once the
picture comes, whether the work is right. So the verdict is a second tool,
`show_verdict`, which the model calls with `right` or `wrong` just before it
answers. The lamp shows it with the body at once, as the call is answered,
so the nod or the tilt begins a moment before the first words and may be
over by then:

- **right:** `acknowledge`, V1's short nod (about 2 s).
- **wrong:** `curious`, V1's tilt (3 s).

It is not called when any part of the work could not be read: the lamp
says what it cannot read, and a nod or a tilt would claim a verdict it does
not have (a nod over a page with a line in shadow, the seventh on #1260).

The head stays over the page afterwards, as Desk tutor holds it for 2
minutes (#1098), so a follow-up is still over the page; it comes back up
when the page's light is given back.

## If asked about it

"When I check your work, I nod if it's right and tilt my head if something
looks off, then tell you where."

