Concept Loom

https://github.com/zproger/conceptloom
Admin
Concept Loom is a portable AI learning companion for coding agents that structures learning through...
Tokens:8,232
Snippets:95
Trust Score:8.7
License:MIT
Update:1 week ago
Show doc for...
Tokens:
Raw
### Quick Start with Codex

Source: https://github.com/zproger/conceptloom/blob/main/README.md

Install the adapter and start the learning session in your workspace.

```bash
# Install the adapter and register its MCP server automatically
node scripts/setup.mjs codex --target /path/to/learning-workspace

# Start learning
cd /path/to/learning-workspace
codex
```

--------------------------------

### Create a workspace with setup.mjs

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Sets up a learning workspace for the Codex CLI and starts an interactive session. Replace /path/to/learning-workspace with your target directory.

```bash
node scripts/setup.mjs codex --target /path/to/learning-workspace
cd /path/to/learning-workspace
codex
```

--------------------------------

### Verify Project

Source: https://github.com/zproger/conceptloom/blob/main/README.md

Run tests and checks to verify the project installation.

```bash
npm test
npm run check
```

--------------------------------

### Start DeepSeek Harness

Source: https://github.com/zproger/conceptloom/blob/main/README.md

Launch DeepSeek Harness from the learning workspace using the provided patch.

```bash
npx @deepseek-ai/dsh --profile web --patch .dsh/concept-loom.patch.yml
```

--------------------------------

### Setup Other Clients

Source: https://github.com/zproger/conceptloom/blob/main/README.md

Commands to configure Concept Loom for Claude Code, OpenCode, or DeepSeek Harness.

```bash
node scripts/setup.mjs claude-code --target /path/to/learning-workspace
node scripts/setup.mjs opencode --target /path/to/learning-workspace
node scripts/setup.mjs deepseek-harness --target /path/to/learning-workspace
```

--------------------------------

### Preserve progress notebook structure

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Directory structure for preserving progress in the notebook folder. At the end, record demonstrated knowledge, fragile points, corrected misconceptions, and the exact starting question for next time.

```text
notebook/
├── goal.md
├── sources.md
├── learning-map.md
├── knowledge-state.md
├── misconceptions.md
└── sessions/
```

--------------------------------

### Initialize Concept Coach

Source: https://github.com/zproger/conceptloom/blob/main/README.md

Prompt the agent to begin the learning session.

```text
Use $concept-coach. I want to understand database transactions well enough
to choose the right isolation level for a small web application.
```

--------------------------------

### Create a navigation index

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Instructions for creating an index file to manage navigation and unit metadata within the prepared directory.

```text
Inspect prepared/ for navigation. Create prepared/index.md with each unit's title,
path, concepts, prerequisites, and source location. Do not summarize the entire
work or begin teaching.
```

--------------------------------

### Personalize source order with concept-coach

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Prompt instructions to diagnose knowledge and propose a dependency route based on the prepared index.

```text
Use $concept-coach. My practical outcome is [OUTCOME]. Read prepared/index.md
and verify relevant source sections. Identify required, prerequisite, and optional
units; diagnose my knowledge; propose a personal dependency route; wait for approval.
```

--------------------------------

### Approve the route prompt

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Prompt to propose the smallest dependency path and save it to notebook/learning-map.md, then wait for approval before proceeding.

```text
Propose the smallest dependency path from what I understand to my outcome.
Show what we are leaving out and define a practical completion task. Save it to
notebook/learning-map.md, then wait for my approval.
```

--------------------------------

### Make something small prompt

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Prompt to create a small mini-project using only secured nodes, with agreed completion criteria. The coach reviews decisions and results without implementing.

```text
Give me a 30–60 minute mini-project using only secured nodes. Agree on completion
criteria first. Do not implement it for me; review my decisions and result.
```

--------------------------------

### Starter prompt for Concept Loom

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

A comprehensive prompt to initialize the study session, build an index, and set ground rules for interaction.

```text
Use $concept-coach.

I am studying [BOOK OR COURSE]. Originals are in originals/ and searchable text
is in prepared/. My practical outcome is [OUTCOME]. Build or inspect the index,
separate source claims from claims needing verification, diagnose prerequisites,
and propose a personal route. Wait for approval. Require my attempt before help,
check one connection at a time, use transfer problems, preserve source references,
update concise state in notebook/, and never modify originals/.
```

--------------------------------

### Workspace directory structure

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Recommended folder organization for keeping original source files separate from prepared materials and project notes.

```text
learning-workspace/
├── originals/    unchanged source files
├── prepared/     searchable text and index
├── notebook/     goal, route, and knowledge state
├── exercises/
└── projects/
```

--------------------------------

### Work one connection at a time prompt

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Prompt to work on one connection at a time, using checkpoints and application problems. Do not advance until the weakness is repaired.

```text
Take the first unsecured node. Explain why it is needed, help me derive it from
established ideas, and let me try before giving the answer. Then use a checkpoint,
ask me to explain the connection, and give one new application problem. Do not
advance until the weakness is repaired.
```

--------------------------------

### Build a source map prompt

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Prompt to build a source map for the learning outcome. Requires stronger verification for medicine, law, finance, safety, security, and rapidly changing topics.

```text
Find a small set of authoritative sources for this outcome. Prefer official
documentation, standards, primary research, and maintained university material.
Record each source's link, date or version, purpose, and limitations in
notebook/sources.md. Separate established facts from contextual or disputed claims.
```

--------------------------------

### Protect exercise solutions

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Prompt instructions to ensure the agent does not reveal solutions before the user attempts the problem.

```text
Do not read or reveal the solution before recording my attempt. Ask me to state
the goal, known information, and likely connection. Use the smallest possible
hint. After completion, give a new problem with different surface details.
```

--------------------------------

### Concept Loom Workflow

Source: https://github.com/zproger/conceptloom/blob/main/README.md

The core sequence of the learning process.

```text
goal → knowledge frontier → learning map → one connection → check → transfer
```

--------------------------------

### Extract EPUB content

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Command to unzip an EPUB file into a prepared directory for processing.

```bash
mkdir -p prepared/epub
unzip -q "originals/book.epub" -d prepared/epub
```

--------------------------------

### Extract audio from video

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Use ffmpeg to extract audio from a video file into a WAV format suitable for transcription tools.

```bash
ffmpeg -i "originals/lesson.mp4" -vn -ac 1 -ar 16000 "prepared/lesson.wav"
```

--------------------------------

### Split analysis into Source and Evidence views

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Prompt instructions to separate author claims from current evidence to identify disagreements or outdated information.

```text
Split the analysis into Source view (what the author claims, with location) and
Evidence view (whether important claims remain supported by current primary or
official sources). Show disagreements or outdated assumptions explicitly.
```

--------------------------------

### Locate your frontier prompt

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Prompt to locate the learner's frontier by diagnosing prerequisite strands and identifying where errors or uncertainty begin. Encourages using 'I do not know yet' instead of guessing.

```text
Diagnose only the prerequisite strands needed for my outcome. On each strand,
find something I understand reliably and the point where errors or honest
uncertainty begin. Determine whether a mistake is a slip, local gap, or misconception.
```

--------------------------------

### Convert DOCX to text on macOS

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Use the textutil command to convert DOCX or RTF files into searchable text format.

```bash
textutil -convert txt -output "prepared/document.txt" "originals/document.docx"
```

--------------------------------

### Record knowledge evidence

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Categories for tracking knowledge state in the notebook.

```text
secure         recalled and applied in a new problem
fragile        recognizable but unreliable in use
missing        not yet built
misconception  an incorrect model was detected
```

--------------------------------

### Finish with transfer prompt

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Prompt to finish with transfer by testing the original outcome in a new integrated situation, asking for justification, and reporting secure and fragile areas.

```text
Test the original outcome with a new integrated situation. Do not reuse earlier
examples. Ask me to justify my choices, then report what is secure, what remains
fragile, and the minimum work needed to close the gap.
```

--------------------------------

### Extract PDF text using pdftotext

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-with-source-materials.md

Commands to inspect PDF metadata and convert PDF content to text files, optionally specifying page ranges.

```bash
pdfinfo "originals/book.pdf"
pdftotext -layout "originals/book.pdf" "prepared/book.txt"
pdftotext -f 25 -l 48 "originals/book.pdf" "prepared/chapter-02.txt"
```

--------------------------------

### Define an outcome prompt

Source: https://github.com/zproger/conceptloom/blob/main/docs/en/learning-from-scratch.md

Prompt for the Concept Loom coach to define a practical, observable outcome instead of a subject. The coach asks about usage, depth, and time before teaching.

```text
Use $concept-coach.

I want to learn [TOPIC], but I do not have materials yet. Help me turn this
into a practical, observable outcome. Ask where I will use it, the depth I need,
and my available time. Do not begin teaching yet.
```
Tokens:
Raw