learn-from-session

You shipped it with AI. Now actually understand it.

A Claude Code skill that turns the coding session you just finished into a short, credibility-checked reading path, sized to what you already know.

The problem

You pair with an AI agent for an afternoon and ship a Chrome extension, a rate limiter, an SSE endpoint. It works. A week later someone asks why you picked a service worker over a content script, and you have nothing. The reasoning lived in the chat log, and the chat log is gone.

Reading random tutorials afterward doesn't fix this. Generic tutorials don't know what you built, which decisions mattered, or what you already understand. This skill does, because it reads your session.

What it does

Say "I want to learn what we just built" at the end of a session and the skill runs a six-step loop:

  1. Extracts 3 to 6 real topics from the conversation. Not everything mentioned, only the things where deeper understanding would have changed a decision. Each named as a searchable phrase ("Chrome MV3 service worker lifecycle", not "extensions").
  2. Finds 2 to 3 sources per topic. Practitioners who built the thing rank first. Interactive explainers with live playgrounds and explorable diagrams beat equally-correct walls of prose. SEO listicles and anonymous content farms get rejected outright.
  3. Runs a credibility pass. Every recommended piece gets searched on Hacker News, Reddit, and Twitter/X, and you see the evidence: "400-point HN thread, mostly agreeing" or "no independent mentions found, judged on content alone". Anything credibly debunked in comments gets dropped.
  4. Recaps what you built in plain language, including why each approach won over its alternative, with the rejected option and the tradeoff named. Then it pauses and asks what feels shaky before testing you.
  5. Places you with a conversational quiz. 5 to 8 questions, easy to hard, asked a few at a time with honest grading that adapts as you answer. At least one question comes straight from your session's code: predict the output, spot the bug.
  6. Orders the reading by what the quiz revealed: shaky-but-load-bearing topics first, new topics second, depth on solid ground last. The first pass stays under 2 hours, with a "going deeper" tier for later.

What you get

Example output: recap, quiz results, ordered reading path with credibility evidence

## What you built today (30-second recap)

## Quiz results
topic → level (solid / shaky / new)

## Reading path (in order)
1. <title> — <author/source> · <time to read> · <why this one, and what
   to look for while reading>
   Credibility: <HN/Reddit/Twitter evidence>

## Skip for now
- <topic you already know / rabbit hole not worth it yet>

Install

One command, via the skills CLI (detects Claude Code, Cursor, Codex, and other agents automatically):

npx skills add notacp/learn-from-session

Prefer no CLI? Copy the file directly (Claude Code, macOS/Linux):

mkdir -p ~/.claude/skills/learn-from-session
curl -fsSL -o ~/.claude/skills/learn-from-session/SKILL.md \
  https://raw.githubusercontent.com/notacp/learn-from-session/main/SKILL.md

The skill follows the Agent Skills format (a SKILL.md with YAML frontmatter), so it works in any agent that reads that spec. For a project-scoped install, put it in .claude/skills/ inside the repo instead.

How to trigger it

Any of these, at the end of a working session:

  • "I want to learn what we just built"
  • "Find blogs about what we did today"
  • "Go through our chat and get me reading material"
  • "Quiz me on this session"

Why it's built this way

  • Quiz before reading list. Recommending an intro article for a topic you aced wastes your evening. The placement quiz decides where each topic's reading starts.
  • Credibility over search rank. A large share of technical search results are content farms with no named author. Independent discussion on HN and Reddit is a cheap, honest filter, and the skill shows its evidence instead of asking you to trust it.
  • 2 to 3 sources per topic, not 10. A shortlist gets read. A dump gets bookmarked and forgotten.
  • Interactive sources rank up. For people who learn by doing, a blog with a live playground (think Josh Comeau or Bartosz Ciechanowski) beats an equally-correct essay. The skill says so in the reading path: "has a live playground, change the values and watch the cap trip".
  • Under 2 hours. This is the reading you do tonight, not a curriculum you abandon by Thursday.

License

MIT