Agent Instructions

This repository is the user's structured study project for machine learning, deep learning, Transformer/LLMs, and YOLO/computer vision.

Required Workflow

Before teaching, editing, or continuing the study:

1. Read 00-roadmap/current-state.md.
2. Read the latest relevant entry in 00-roadmap/learning-log.md.
3. Read 00-roadmap/session-recording-policy.md.
4. Read the relevant module README or lesson file.

During teaching:

1. Explain intuition first.
2. Then explain structure or formal concepts.
3. Use a small example where helpful.
4. Call out common misconceptions.
5. Ask or record understanding checks.
6. State the next learning step.

After any learning-related turn:

1. Update 00-roadmap/learning-log.md if the turn added learning content, user answers, corrections, or plan changes.
2. Update 00-roadmap/current-state.md if progress, next lesson, or review points changed.
3. Add durable lessons under the relevant module's lessons/ directory.
4. Add exercises or answer records under the relevant module's exercises/ directory when useful.
5. Commit meaningful documentation changes with Git.

Local Skill

The project also has a local Codex skill intended to reinforce this workflow:

/home/zhuji/.codex/skills/ml-study-logger/SKILL.md

If the skill is available, use it for this repository. If it is not automatically loaded, follow this AGENTS.md file and the project recording policy manually.

Boundaries

• Do not depend on large GPU training for first-round lessons.
• Prefer durable Markdown notes, lightweight examples, and pre-trained model inference.
• Do not commit datasets, model weights, virtual environments, caches, or training outputs.