Turn messy homework, scans, and exam papers into an evidence-based learning plan your AI agent can run.
把一堆试卷、照片、讲义和批改痕迹，变成孩子今天该补什么、怎么补、下次怎么验证。
Most families do not lack learning materials. They lack a reliable loop.
After every exam or homework session, the same questions come back:
- Which mistakes are careless, and which are real knowledge gaps?
- What should the child do today instead of blindly doing another full paper?
- How do we know the problem is actually fixed next week?
- How can an AI agent help without turning everything into a messy folder system?
This repository gives Codex, Claude Code, OpenClaw, or any file-based AI agent a ready-to-run workflow for that loop.
It is not a question bank. It is not another folder template. It is a learning operations system for evidence, diagnosis, printable retests, and follow-up.
After you drop learning materials into 00_Inbox/, your agent can produce:
- A clear diagnosis of the child's top 1-3 learning gaps.
- Evidence links back to the real homework, exam paper, scan, or answer sheet.
- A short action plan for today: what to do, how much, and how long.
- A printable assessment or retest when evidence is missing.
- A knowledge gap ledger that tracks whether each gap is new, explained, retested, or stable.
- A next-action list with validation method, completion evidence, and write-back location.
The point is simple: do not collect more materials. Close the learning loop.
git clone https://github.com/Fuyera/education-learning-workflow.git
cd education-learning-workflowPut files into:
00_Inbox/
Then tell your agent:
Read START_HERE.md and AGENTS.md first.
Analyze the materials I put in 00_Inbox.
Tell me what the child does not understand, why the mistakes happened, and what to do today.
If the evidence is not enough, generate a printable assessment.
Chinese prompt:
请先阅读 START_HERE.md 和 AGENTS.md。
请分析我放进 00_Inbox 的资料，告诉我孩子哪里不会、为什么错、今天怎么补。
如果证据不足，请生成一份可打印评估卷。
You do not need to create folders, rename files, or understand the skills. Put the files in; let the agent run the workflow.
Project-level use works immediately after clone. If you also want the workflow skills installed globally:
./tools/install.sh --codex
./tools/install.sh --claude
./tools/install.sh --openclawInstall to a custom skills directory:
./tools/install.sh --target /path/to/skillsSee docs/INSTALL.md for details.
Before:
Downloads full of PDFs
Phone photos of answer sheets
AI comments copied from chat
Parents unsure what to print next
No record of whether a weak point was fixed
After:
00_Inbox receives all raw materials
Agent separates originals from student work
Analysis reports point to evidence
Knowledge gaps are tracked over time
Printable retests create new evidence
Next actions close the loop
This workflow is strict about evidence.
- It does not mark a child as "mastered" without real student work.
- It treats AI-generated error reports as clues, not proof.
- It tracks knowledge gaps instead of building an endless per-question mistake database.
- It separates short retest success from stable mastery.
- It forces every next action to have a validation method and write-back location.
That makes it more useful than a one-off "analyze this paper" prompt. The agent is not only explaining mistakes. It is maintaining the learning system.
Use it if you are:
- A parent with homework photos, exam scans, and no clear next step.
- A student who wants a repeatable way to turn mistakes into focused practice.
- A tutor or learning coach who needs a reusable diagnostic workflow.
- An AI agent user who wants Codex, Claude Code, or OpenClaw to manage learning records reliably.
Do not use it if you only want a large generated question bank. This workflow starts from evidence and closes gaps.
00_Inbox/ Drop files here
01_Originals/ Archived source materials
02_Student_Work/ Real student answers, scans, photos
03_Learning_Records/Analysis/ Evidence-based analysis reports
03_Learning_Records/Knowledge_Gaps/ Knowledge gap ledger
03_Learning_Records/Plans/ Printable assessments, retests, plans
60_Actions/ Next actions and validation records
80_System/skills/ Agent workflow skills
docs/ User, agent, workflow, and install guides
tools/ Install, validate, and package scripts
python3 tools/validate.pyThe validator checks required folders, skill files, skill prompts, and common public-package hygiene issues.
Student work is private.
This repository ignores user-uploaded materials and generated learning records by default. Before publishing or pushing changes, make sure you are not committing student names, school names, answer sheets, scans, photos, scores, or other private data.
MIT. See LICENSE.