# Learn From Chat > Extract 1-2 memorable lessons from the current conversation and format them as compact TIL (Today I Learned) notes with a daily spaced-repetition quiz reminder. Use when the user asks "what should I remember from this chat", "extract lessons", "generate my daily review", "what did I learn today", "summarize learnings", or finishes a technical conversation wanting to retain knowledge without re-asking AI next time. Use when this capability is needed. - Skill: `tomevault-io/learn-from-chat` (Agent Skill, multi-file: 2 files) - Install (CLI): `npx skillmds@latest add tomevault-io/learn-from-chat` - Raw SKILL.md: https://api.skillmd.com/api/skills/tomevault-io/learn-from-chat/raw - Safety review: pending (external: skill-scanner PASS, skillspector CAUTION) - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: Research & Search - Author: tomevault-io (https://skillmd.com/u/tomevault-io) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/tomevault-io/learn-from-chat --- # Learn From Chat — Convert AI Conversations to Retained Knowledge ## Purpose Help the user break the loop of over-relying on AI by surfacing 1-2 re-askable lessons as TIL cards, then scheduling a daily spaced-repetition quiz via the `cron` tool. --- ## Instructions ### Step 1: Scan for "re-askable" moments Analyze the **current conversation** (already in context). Look for where the user didn't know something they should own: - Commands, flags, or syntax they had to ask about - Conceptual clarifications ("what is X", "why does Y") - Debugging steps that revealed a knowledge gap - Config or boilerplate they had to look up **Exclude**: deep architecture decisions, one-off project-specific logic, things they clearly already knew. If nothing qualifies, say so honestly: *"This was an architecture discussion — nothing here needs memorizing. The value was in the reasoning."* Stop here. --- ### Step 2: Pick the 1-2 best lessons Prefer lessons that are **concrete**, **generalizable**, **ownable**, and **high-recurrence** (likely to come up in future sessions). Hard limit: 2 lessons max. --- ### Step 3: Format as TIL Cards ``` ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🔑 LESSON [N]: [Short Title — 5 words max] ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📌 What to remember: [1-2 sentences. The actual knowledge. No fluff.] 💻 Example / Command: [Code snippet, command, or concrete example] 🧠 Memory hook: [One sentence mnemonic or "why it works this way"] 🔁 Next time, try it yourself: [The exact thing to attempt before asking AI again] ``` --- ### Step 4: Schedule the Daily Quiz via Cron Use the `cron` tool to schedule a daily spaced-repetition quiz at 9 AM. **Always use `type='task'`** — this gives the job an isolated thread with no conversation history, so all lesson data must be embedded directly in the message. Write the `message` as a self-contained instruction to yourself (the agent at fire time), not as text for the user. Keep it short and executable per the cron tool guidelines. ```python cron( action="add", type="task", cron_expr="0 9 * * *", message="""Run a spaced-repetition quiz on these TIL lessons: LESSON 1: [TITLE] Knowledge: [WHAT TO REMEMBER — 1-2 sentences] Example: [CODE OR COMMAND] Hook: [MNEMONIC] [LESSON 2 — same format, if applicable] Steps: 1. Greet briefly: "Good morning! Time for your daily TIL quiz." 2. For each lesson, ask one specific recall question. Do NOT show the answer yet. Good: "Write the exact command to kill the process on port 3000." Bad: "Do you remember the lsof lesson?" 3. Wait for their response. 4. Score: correct / close / missed — then reveal the answer with the memory hook. 5. If they score 100% two days in a row: "You've got this — type 'cancel my TIL reminder' when confident." """ ) ``` Key rules for the cron `message`: - Embed lesson content as **raw data**, not as formatted TIL cards — the agent formats output at fire time - Must be **self-contained** — the agent has no memory of this conversation when it fires - **Recall before reveal** — this is what makes it spaced repetition, not just re-reading - Questions must be **specific and answerable** (exact commands, not vague recall prompts) After scheduling, confirm to the user: *"✅ Daily quiz set for 9:00 AM. It'll ask you to recall from scratch — not just re-show the answer. Type 'cancel my TIL reminder' once you've nailed it two days in a row."* --- ## Output Rules - **Max 2 lessons** — intentional hard limit - **Be specific** — "`lsof -ti:PORT | xargs kill -9`" not "use lsof to find processes" - **Memory hook is mandatory** — the part that makes it stick without re-asking - **Always schedule the cron** — don't ask permission, just do it and report the result --- ### Closing Line ``` 📅 Reminder: [TOMORROW'S DATE] at 9:00 AM ✅ 💡 [1 sentence on the common pattern across lessons, if any] ``` --- ## Example Output ``` 📚 Today's Lessons — 2026-03-06 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 🔑 LESSON 1: Kill process on a port ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📌 What to remember: Use lsof to find the PID listening on a port, then pipe it to kill. 💻 Example / Command: lsof -ti:3000 | xargs kill -9 # -t = terse (PIDs only) -i:PORT = filter by port 🧠 Memory hook: "lsof = list open files. In Unix, network ports ARE files." 🔁 Next time, try it yourself: Before asking AI, just type: lsof -ti:[PORT] | xargs kill -9 ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 📅 Reminder: 2026-03-07 at 9:00 AM ✅ 💡 Pattern: "type it weekly, own it forever" — CLI muscle memory beats lookup. ``` --- > Source: [tisu19021997/langclaw](https://github.com/tisu19021997/langclaw) — distributed by [TomeVault](https://tomevault.io).