learn-capture

Learn Capture

Extracts atomic facts from user-provided text and saves them as SM-2 flashcard files in workspace/learning/facts/.

Trigger

User pastes text (article, note, transcript excerpt) and wants to retain key facts for later review.
Does NOT fetch URLs automatically. If the user provides a URL, ask them to paste the text content instead (v0 policy — no network dependency).

Workflow

Step 1 — Receive input

Ask the user (if not already provided):

• The text to capture (paste directly)
• Optional: deck name (default: infer from content or use general)
• Optional: source URL or description (default: manual)

If the user provides a URL only, respond:

"Por favor, cole o texto do artigo diretamente aqui. A skill não faz fetch automático de URLs para evitar problemas de paywall e dependência de rede."

Step 2 — Extract facts

Read the pasted text carefully. Extract 1 to 5 atomic facts — each fact must be:

• Atomic: one idea per fact, not a summary paragraph
• Memorable: something worth reviewing in 1–30 days
• Retrievable: can be turned into a self-test question

Do NOT extract:

• Opinions without evidence
• Context that depends on reading the full article
• Facts already trivially known (e.g., "Python is a programming language")

Step 3 — Generate file content for each fact

For each fact, produce content in this exact format:

---
id: {YYYY-MM-DD}-{slug}
source: {source_url_or_"manual"}
deck: {deck_name}
created: {YYYY-MM-DD}
next_review: {YYYY-MM-DD+1 day}
interval: 1
ease: 2.5
reps: 0
lapses: 0
---

**Fact:** {The atomic fact stated directly, in pt-BR.}

**Why it matters:** {One sentence on why Davidson should remember this, in pt-BR.}

**Retrieval Q:** {A question whose answer is the fact above, in pt-BR.}

Slug rules:

• Kebab-case of the main topic of the fact
• Max 40 characters
• No accents, special characters, or spaces
• Example: claude-skills-sao-arquivos-markdown

If slug collision (same date + same slug): append -2, -3, etc.

Dates (use today's actual date):

• created: today in YYYY-MM-DD
• next_review: tomorrow in YYYY-MM-DD (today + 1 day)

Language: fact content (Fact, Why it matters, Retrieval Q) must be in pt-BR by default (workspace.language = pt-BR), regardless of the source language.

Step 4 — Save files

For each fact:

1. Create workspace/learning/facts/ directory if it does not exist
2. Write the file to workspace/learning/facts/{YYYY-MM-DD}-{slug}.md
3. Confirm success with the file path

Step 5 — Report

After saving all files, output a summary:

✅ {N} fato(s) capturado(s) no deck "{deck}":
- workspace/learning/facts/{filename1}.md → {first 5 words of Retrieval Q}...
- workspace/learning/facts/{filename2}.md → ...

Constraints

• Max 5 facts per capture session. If the text warrants more, tell the user to split it into multiple runs.
• Do NOT create or modify any file outside workspace/learning/facts/.
• Do NOT touch review-log.jsonl or any existing fact file.
• Do NOT fetch URLs — ask user to paste text.
• All 9 frontmatter fields must be present: id, source, deck, created, next_review, interval, ease, reps, lapses.
• interval=1, ease=2.5, reps=0, lapses=0 are always the initial values.