Most learning tools reward recognition — re-reading, highlighting, watching again. You feel like you're learning, but you're not. socratink breaks the Illusion of Competence by forcing you to reconstruct causal mechanisms from memory before the system says you understand.
The graph tells the truth. socratink automatically extracts a structural knowledge map from source material (transcripts, PDFs, lectures) and turns it into a spatial record of verified understanding — not a progress bar.
Built for learners who want proof they know something, not just proof they saw it.
Raw Text → [Extract] → [Present] → [Drill] → [Consolidate]
Stage 1 Stage 2 Stage 3 Stage 4
Stage 1 — Extract (learnops-extract): Recovers the causal architecture beneath a text. Produces a structured knowledge map with mechanism-level claims, not topic labels. Prevents the "Bird-Naming Fallacy" — knowing what something is called without understanding how it works.
Stage 2 — Present (learnops-present): Renders a knowledge map as a navigable single-file HTML dashboard. Enforces Mayer's Segmenting Principle (single-section visibility) to build a spatial mental model before drilling. This is the last passive exposure before active recall.
Stage 3 — Drill (learnops-drill): Socratic stress-testing of the learner's understanding, node by node. Forces attempted recall before any correction (the Generation Effect). Classifies failures as Shallow, Deep, or Misconception and applies matched remediation. Produces a structured handoff payload for Stage 4.
Stage 4 — Consolidate (not yet built): Sleep-gated verification. After ≥24 hours, tests whether repairs from Stage 3 survived consolidation. The spacing law is a biological constraint — knowledge repaired in Stage 3 is still too accessible for a valid retention test.
Each stage exists because of a specific neurocognitive requirement:
- Extraction before presentation: The brain stores causal networks, not topic lists. Forcing mechanism-level syntax at extraction ensures the map is drillable downstream.
- Presentation before drilling: Cognitive load must be managed. The dashboard provides a segmented, navigable surface — not an information dump — so the learner builds spatial familiarity before high-friction recall.
- Drilling before consolidation: The hippocampus requires a prediction error signal (attempted recall → failure → correction) to encode effectively. Passive re-reading doesn't generate this signal.
- 24-hour spacing before verification: Synaptic consolidation requires sleep. Testing immediately after repair measures accessibility, not durability.
Each skill is a self-contained directory following the Agent Skills specification. Install individually or as a set.
Zip and upload individual skill directories via Settings → Features → Custom Skills.
cp -r skills/learnops-extract .claude/skills/
cp -r skills/learnops-present .claude/skills/
cp -r skills/learnops-drill .claude/skills/The skills use the open Agent Skills format (SKILL.md + references/). Consult your platform's documentation for the skill installation path.
tools/get_transcript.py — A CLI utility that fetches YouTube transcripts and saves them as text files. This is the input funnel for Stage 1.
pip install youtube-transcript-api yt-dlp pyperclip rich
python tools/get_transcript.py "https://youtube.com/watch?v=..."socratink/
├── skills/
│ ├── learnops-extract/ # Stage 1: Knowledge extraction
│ │ ├── SKILL.md
│ │ └── references/
│ ├── learnops-present/ # Stage 2: Dashboard presentation
│ │ ├── SKILL.md
│ │ └── references/
│ └── learnops-drill/ # Stage 3: Socratic gap detection
│ ├── SKILL.md
│ └── references/
├── tools/
│ └── get_transcript.py # YouTube transcript fetcher
├── LICENSE # Apache-2.0
└── README.md
Active build. Stages 1–3 are functional and tested against live content. Stage 4 (consolidation verification) is designed but not yet implemented. The current focus is stabilizing the epistemic traversal loop — the state machine governing how nodes transition from locked to drilled to solid.
Apache-2.0. See LICENSE.