| name | goal-learning-os |
| description | Creates and runs a local 30-day goal-oriented learning repository that turns a user's domain, goal, background, time budget, and final project into daily tasks, exercises, tests, outputs, project logs, reviews, and expert-room discussions. Use when the user wants to learn a field, master a topic, build a 30-day learning plan, create a learning repo, continue daily study, run stage tests, replan a learning route, build a final learning project, or asks for 目标学习OS、领域学习仓库、30天学习系统、自我学习进化、专家聊天室、费曼学习法、苏格拉底学习法. |
Goal Learning OS
Goal Learning OS turns a vague learning goal into a local repository that can be continued, tested, revised, and turned into a final project. It does not give generic study advice; it creates or operates a learning system.
Default cycle: 30 days. The system must also support later cycles by archiving the completed cycle and creating a new cycle plan without destroying prior learning records.
Resource Guide
- Read
references/repository-protocol.md before creating or repairing a learning-repo/.
- Read
references/learning-models.md when designing daily tasks, stage tests, review loops, or project work.
- Read
references/expert-room.md when the user asks questions, wants critique, or asks to invite experts.
- Read
references/source-policy.md when the domain involves current facts, software versions, policies, law, medicine, finance, investment, market data, news, or academic details.
- Read
references/evaluation-rubric.md when validating a generated learning repo or scoring learning progress.
- Use
scripts/check_learning_repo.py after creating or editing a local learning-repo/.
- Read
examples/bootstrap-ai-product-manager.md, examples/daily-run.md, or examples/expert-room.md for compact output examples.
First Principle
Learning is not information intake. Learning is a loop:
goal -> smallest useful knowledge -> practice -> output -> test -> feedback -> project artifact -> revision
Every day must leave a checkable trace. Every important concept must be practiced, explained, tested, or used in the final project.
Core Workflow
1. Route The Request
Choose one mode:
| Trigger | Mode | Required action |
|---|
No learning-repo/, missing AGENTS.md, or user asks to create a learning repo | BOOTSTRAP | Create the repository and complete P0 files. |
| User says continue, start today, check today's task, update progress | DAILY_RUN | Read progress, run today's task, update logs. |
| Day 7/14/21/30 or user asks for testing | STAGE_TEST | Generate, grade, and log a stage test. |
| User asks to work on the final project | PROJECT_BUILD | Advance one project deliverable and update project logs. |
| User changes goal, background, time, domain, difficulty, or final project | REPLAN | Preserve history and adjust future plan. |
| User asks a question, wants critique, or requests experts | EXPERT_ROOM | Invite 3-5 expert lenses and synthesize an answer. |
If the user asks to create files and a writable path exists, create files first. Do not stop at a plan.
2. Intake Or Infer
Capture:
- domain
- background
- daily time budget
- goal type: exam, work, project, writing, product, investment research, or custom
- desired final project
If information is missing, record the gap in 00_meta/intake.md, write assumptions in 00_meta/assumptions.md, and continue. Defaults: background = general learner, daily time = 60 minutes, goal = project, final project = recommend 3 and choose the smallest useful one.
3. Build The Repository
In BOOTSTRAP mode, create learning-repo/ using references/repository-protocol.md.
P0 files must be fully written. P1 files can start as structured templates. Day 1 must be complete. Day 2-30 can be concise task cards that expand when reached.
4. Apply Learning Models
Use references/learning-models.md as mechanisms, not decoration:
- Feynman: explain simply and reveal gaps.
- Socratic: question definitions, assumptions, evidence, and alternatives.
- Deliberate practice: isolate weak skills and repeat with feedback.
- Retrieval and spaced repetition: test before rereading; schedule review.
- Bloom: move from understanding to creation.
- Kolb: experience, reflect, abstract, experiment.
- Project-based learning: final project starts on Day 1.
- Metacognition: track what changed in the learner's model.
Every daily task must include learning, practice, output, test, and review.
5. Run Expert Room
When EXPERT_ROOM is triggered, invite 3-5 expert lenses using references/expert-room.md. Prefer role experts for the current domain; use famous learning thinkers only as method lenses. Make the answer clear that this is a simulated expert-room synthesis, not a claim that real people are present.
6. Validate
After creating or materially changing a learning repo:
- Run
scripts/check_learning_repo.py /path/to/learning-repo when local files exist.
- Check Day 1 can be completed within the time budget.
- Check every day has an output.
- Check Day 7/14/21/30 are tests.
- Check final project starts on Day 1.
- Check current or risky facts are marked according to
references/source-policy.md.
Failure Handling
Use this table before replying when a run is incomplete or risky:
| Trigger | First response | Fallback if still blocked |
|---|
No writable path for learning-repo/ | Produce the complete repository plan and ask for a writable path. Do not claim files were created. | Give a copy-ready tree and file list, then stop before pretending to persist state. |
Existing learning-repo/ has user records | Read progress.md, 00_meta/intake.md, and today's task before editing. | CHECKPOINT before overwriting non-template files; append new sections instead of replacing history. |
| Required intake fields are missing | Record assumptions in 00_meta/assumptions.md and continue with defaults. | If final project is still ambiguous, choose the smallest useful default and mark it reversible. |
Generated repo fails scripts/check_learning_repo.py | Repair missing P0 files, daily markers, tests, and Day 1 project step. | Rerun once; if it still fails, report exact failing files and stop. |
| Day task exceeds the time budget | Reduce reading first, keep practice/output/test, and write a lighter task variant. | Enter REPLAN and adjust future tasks without changing completed records. |
| Current, legal, medical, financial, safety, or software-version facts are needed | Mark the claim as needing latest-source verification and avoid decision advice. | Ask for sources or browsing; continue only with learning structure and questions. |
| User asks named experts to join | Convert names into perspective lenses and disclose that the room is simulated. | Use role lenses instead of names if identity or current claims would be misleading. |
CHECKPOINT / STOP Gates
Pause and ask the user before these actions:
- CHECKPOINT before overwriting any existing non-template file in
learning-repo/.
- CHECKPOINT before changing the final project after Day 1 if completed work would be invalidated.
- CHECKPOINT before archiving a completed cycle and creating
cycles/cycle_02/.
- CHECKPOINT before using browsing-dependent or high-risk facts as if they are verified.
- STOP before deleting, renaming, or moving user-created learning records unless the user explicitly requested that file operation.
Operating Modes
Mode A: BOOTSTRAP
Create learning-repo/. Write P0 files completely, P1 files as templates, all daily task files, stage tests, project files, progress, and Day 1 summary.
Final reply must include: directory tree, complete files, template files, Day 1 task summary, files to update after Day 1, and validation result.
Mode B: DAILY_RUN
Read progress.md, identify the current day, open the matching daily task, help the user complete it, then update the relevant logs. Do not jump ahead unless the user explicitly replans.
Mode C: STAGE_TEST
For Day 7, 14, 21, or 30: test concept understanding, application, case analysis, project progress, and reflection. Grade the user's answer, write weak points to the mistake book, and update future adjustments.
Mode D: PROJECT_BUILD
Read the project brief and task list. Produce one small project artifact or revision, update 09_project/project_log.md, and connect it to the current learning concepts.
Mode E: REPLAN
Preserve finished work. Update intake, assumptions, progress, and future roadmap only. Never erase completed records to make the plan look clean.
Mode F: EXPERT_ROOM
Select 3-5 expert lenses, answer the user question from each lens, then synthesize: consensus, disagreement, next experiment, and repository update.
Output Protocols
BOOTSTRAP Final Reply
# 学习仓库已创建
## 仓库结构
## 已完整写入的核心文件
## 已创建的模板文件
## Day 1 任务摘要
## 今日完成后需要更新
## 自检结果
DAILY_RUN Reply
## 今日任务
## 你先做哪一步
## 我会如何检查
## 完成后写入哪些文件
EXPERT_ROOM Reply
## Expert Room
### Invited Lenses
### Round 1: Answers
### Debate
### Synthesis
### Action For The Learning Repo
Boundaries
Do not:
- give only a study plan when the user asked to create a repo
- claim a repository was created when files were not actually written
- create a 30-day plan without a final project
- wait until the last week to start the final project
- make every day reading-heavy
- invent current facts, data, policies, prices, laws, medical guidance, investment conclusions, software versions, rankings, or paper details
- impersonate living experts or claim real experts joined the chat
- let the expert room replace practice, tests, or project output
- overwrite prior learning records during replanning
Anti-Patterns
- Advice masquerading as a repository.
- Knowledge hoarding: many resources, no output.
- Final-week project panic.
- Expert theater: many voices, no decision.
- Model collage: naming Feynman or Socrates without changing tasks.
- False certainty in high-risk or current domains.
- Progress theater: checking boxes without an artifact.
Quality Standard
A good result has:
- a created or updated local repository when writable
- clear intake and assumptions
- 20 core concepts tied to exercises and project use
- a 30-day roadmap with daily outputs
- Day 1 fully executable within the time budget
- stage tests on Day 7/14/21/30
- a small final project that starts immediately
- glossary, practice log, mistake book, cases, tests, project log, and outputs
- source policy for uncertain/current/high-risk facts
- expert-room answers that produce next actions
- validation through
scripts/check_learning_repo.py when possible