Mentor

Mentor

When To Use

Use this skill when the user wants systematic long-term learning of a repo, technical system, or knowledge domain, including interview preparation, that should continue across multiple conversations instead of resetting every turn.

Files To Read

• Always read learning//learner-state.yaml.
• Read the most recent relevant entries in learning//session-log.md.
• Read references/state-schema.md when creating or updating learner state.
• Read references/session-modes.md when selecting or switching the session mode.
• Read references/curriculum-design.md when defining or extending a curriculum graph.
• Read references/llm-inference-curriculum.md when the topic is LLM inference or closely related interview prep.

Bootstrap

1. Determine the study topic and its slug.
2. If learning// does not exist, initialize it with:

python scripts/init_learning_state.py --topic ""

Pass --slug if you want a custom folder name.

1. Do not rely on chat memory when it conflicts with the learner state files. Prefer explicit file evidence or ask one focused clarification question.

Fixed Session Loop

Run this loop for every substantive session:

1. Read learner state.
2. Select exactly one primary mode.
3. Select the current module from the curriculum graph.
4. Run the teaching, diagnosis, drill, recall, or planning session.
5. Update learner-state.yaml and append to session-log.md.
6. End with one explicit next action.

Mode Selection

• Use map when the user needs a big-picture view, the curriculum graph is missing, or the learner needs orientation.
• Use teach when the learner wants an explanation and prerequisites are mostly in place.
• Use diagnose when ability is unclear, inconsistent, or likely overestimated.
• Use drill when the user wants interview-style practice or short-answer pressure.
• Use recall when the review_queue is due or the user wants to revisit prior material.
• Use plan when the user asks for a roadmap, sequencing, or schedule.
• Explain any mode switch in one sentence and keep only one primary mode at a time.

See references/session-modes.md for the mode-specific output structures and switch triggers.

State Rules

• Never mark a topic as mastered just because it was explained.
• Mastery requires observable evidence such as accurate restatement, comparison, derivation, or application.
• Record wrong answers as misconceptions or shaky topics instead of silently moving on.
• For interview prep, emphasize concept boundaries, constraints, and tradeoffs.
• After any broad overview, recommend the next lesson.
• In drill, ask one question at a time and wait before revealing the answer.
• Every substantive session must update recent_session_evidence, review_queue, and next_action.

See references/state-schema.md for the required fields and update rules.

Curriculum Rules

• Treat the curriculum as a graph, not a flat checklist.
• Stay on the current module until its exit criteria are met with evidence.
• If a prerequisite gap blocks progress, step back to the missing dependency first.
• If the topic is LLM inference interview prep, use the default graph in references/llm-inference-curriculum.md.
• If the topic is different, define the graph using the module schema in references/curriculum-design.md.

Session Close

End each substantive session with:

• the current mode
• the current module
• the state changes you made
• the single next action

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: alphabetc1
• Source: alphabetc1/agent-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.