aprende

---

⚡ 30-second pitch

You correct your agent. The session ends. Next week, you correct it again on the exact same thing. aprende fixes that. Type /aprende at the end of a session — it scans the conversation, surfaces every correction worth preserving, asks you which to keep, and writes them where your agent will find them next time. Reflexion-style anti-patterns, confirmation-first writes, cross-tool (Claude Code + Codex), bilingual.

You:  /aprende

aprende:
  Found 3 candidate learnings from this conversation.

   1. [project-doc] Use pnpm, not npm — user said "this repo uses pnpm"  (high)
   2. [lesson]      localStorage fails in purchase WebView — use IndexedDB  (high)
   3. [memory]      User prefers explicit error messages in tests  (medium)

  >

You:  1,2

aprende:
  ✓ project  → ./CLAUDE.md + ./AGENTS.md  (## Commands)
  ✓ lesson   → ~/.claude/projects/.../lesson_webview-localstorage.md
  Captured 2 learnings. Run /aprende --review next month to prune.

---

Why this exists

Coding agents repeat the same mistakes across sessions. The corrections you make in one chat — "no, this repo uses pnpm", "always add using and with check to RLS policies" — evaporate when the session ends. You correct the agent. Next week, you correct it again. And again. aprende fixes that. After each session, you run /aprende. It surfaces what was worth learning. You pick what to keep. It writes it where the agent will find it next time.

---

The four categories

Category	Lives at	When
memory	~/.claude/projects/<slug>/memory/<name>.md	Durable facts, preferences, project context
lesson (Reflexion-style anti-pattern)	Same folder, type: lesson	A mistake happened. Captures what / why / how-to-avoid / detection-signal
skill (stub)	~/.claude/skills/<slug>/SKILL.md	A reusable multi-step workflow appeared
project-doc	./CLAUDE.md + ./AGENTS.md (dual-write)	Build commands, repo conventions, gotchas for every future agent

The lesson category is the one that compounds. Following the Reflexion paper, every lesson must answer four questions:

**What happened:** ...
**Why it happened:** ...
**How to avoid:** ...
**Detection signal:** ...

---

Install

Option 1 — Plugin (recommended)

From inside Claude Code:

/plugin marketplace add Hainrixz/aprende-skill
/plugin install aprende@aprende-skill

That's it. The skill and its hooks are wired up automatically. Hooks are on by default (see disclosure below).

Option 2 — Manual

Clone the repo and copy the skill folder:

git clone https://github.com/Hainrixz/aprende-skill.git
cp -r aprende-skill/skills/aprende ~/.claude/skills/
cp -r aprende-skill/skills/learn ~/.claude/skills/   # English alias (optional)

To enable the hooks (optional):

# Copy the handlers somewhere stable:
mkdir -p ~/.claude/aprende-hooks
cp aprende-skill/hooks-handlers/*.sh ~/.claude/aprende-hooks/
chmod +x ~/.claude/aprende-hooks/*.sh

Then in a Claude Code session, run /aprende-enable-hooks — it will splice the hook entries into ~/.claude/settings.json. You'll need to replace ${CLAUDE_PLUGIN_ROOT} with ~/.claude/aprende-hooks in those entries since manual install doesn't set that env var. The README walks through the exact JSON to paste.

---

Usage