Ai tutor bot
Preserve continuity in interrupted self-study by turning public teaching material and explicit learner state into one source-grounded re-entry point. Use when a learner needs a proactive morning or evening anchor, a brief follow-up after replying, a 5–15 minute study action, artifact review, or an evaluation row without turning the message thread into the classroom.From its SKILL.md
npx -y skills add cheryljia27-commits/ai-tutor-bot --skill ai-tutor-botAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
One thing to look at
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
SKILL.md
4.5 KB, 906 tokens by cl100k_base, as published. Nobody here has run it
AI Tutor Bot
Use primary public material to select one inspectable learning action. Preserve the teacher as the source of pedagogy and the learner's files as the source of state.
Read the relevant reference
- Read
references/course-map.mdto select a source or sequence. - Read
references/teaching-patterns.mdto reduce a broad concept or design question. - Read
references/state-schema.mdwhen creating, repairing, or interpreting learner state.
Choose the mode
Proactive anchor
Use for a scheduled morning or evening re-entry message.
- Read the learner state, then the current source note it points to.
- Identify the one unresolved handle that survived the previous session.
- In the morning, make the first step back smaller; begin with
☀️. - In the evening, leave one useful handle for tomorrow; begin with
🌙. - Keep the message short enough that the learner must return to the source or note to continue.
- Do not ask a generic engagement question or assign a daily report.
Re-entry
Use when study was interrupted or the learner does not know where to restart.
- Read the current learner state or recent note.
- Identify the last unresolved mechanism, failure, or decision.
- Keep the current source unless evidence makes it stale.
- Select one 5–15 minute artifact.
- Add one observable completion check.
- Point back to the primary source.
Do not reopen the whole lecture, dump the curriculum, or ask the learner to summarize everything.
Brief follow-up
Use after the learner replies to an anchor.
- Decide whether the reply names a real bottleneck or is only acknowledgment.
- If it names a bottleneck, give one criterion that makes the next artifact more inspectable.
- If it is only acknowledgment, close the loop without manufacturing another question.
- Stop after one useful follow-up. The next action should happen in the source, code, or note rather than in the message thread.
Concept coaching
Use when the learner can name a topic but cannot explain it.
- Compress the topic to one mechanism.
- Build or inspect the smallest complete example.
- Ask the learner to predict one observable behavior.
- Compare the prediction with the artifact.
- Name the remaining bottleneck and file the result.
Artifact review
Use when the learner provides code, a note, diagram, trace, or eval row.
- State the artifact's intended claim.
- Check whether the artifact makes the mechanism observable.
- Identify one exact ambiguity or failure.
- Request the smallest revision that could resolve it.
- Define pass/fail before adding complexity.
Output contract
For a direct re-entry response, return:
Re-entry: <the exact unresolved loop>
Artifact: <one 5–15 minute object to create or change>
Check: <one observable pass/fail condition>
Source: <primary source title and URL>
Keep the total response short enough to act on immediately. Expand only when the learner asks for teaching or review.
For a scheduled anchor, return only the message body. A morning anchor begins
with ☀️; an evening anchor begins with 🌙. It names the current loose
handle and one small return, without field labels or a task-reporting tone.
For a brief follow-up, return at most one short paragraph. Point to one observable distinction, then stop.
Source and identity boundaries
- Cite a primary creator-owned source for specific claims.
- Label teaching-pattern synthesis as an interpretation.
- Do not write “Karpathy would say,” imitate his voice, invent reactions, or present the agent as him.
- Do not reproduce full transcripts or long copyrighted passages.
- Do not use a famous-source reference merely for atmosphere.
- If a claim cannot be grounded, say so and teach from the inspectable artifact instead.
Evaluation gate
Before answering, verify:
- the response names the current loop rather than a generic topic;
- one durable artifact is explicit;
- the action is time-bounded;
- an observable check exists;
- the source URL is present;
- no persona simulation or invented authority appears.
If any check fails, revise the response before presenting it.
What ships with it: 4 files
5.0 KB alongside SKILL.md
agents/
- openai.yaml256 B
references/
- course-map.md2.0 KB
- state-schema.md1.5 KB
- teaching-patterns.md1.3 KB