paulund/ai-learnt

Learnt — Session Teaching Loop

Review the current conversation and extract non-obvious lessons worth keeping. The goal is to make the next session smarter, not to record everything that happened.

What's worth capturing

Look for:

• Corrections — Claude did X, user said "no, do Y instead"
• Validated approaches — user confirmed or accepted a non-obvious choice without pushback (quiet wins count)
• Revealed preferences — how the user likes to work, what they care about, what annoys them
• Project facts — decisions, constraints, context that isn't in the code or git history

Ignore:

• Things derivable from reading the current code
• Standard best practices Claude already knows
• One-off task details with no future relevance
• Anything already in CLAUDE.md, MEMORY.md, or the global skill files

Decision tree — where does each lesson belong?

Is this a universal rule for a language, framework, or tool?
├── Yes → the relevant global skill file in ~/.claude/skills/
└── No  → Is it specific to this codebase's architecture or conventions?
           ├── Yes → {repo}/CLAUDE.md
           └── No  → Is it about the user, or project context?
                      ├── Working preference / correction → auto-memory (type: feedback)
                      ├── Project fact / decision → auto-memory (type: project)
                      └── Personal context about the user → auto-memory (type: user)

How to write each type

Auto-memory (feedback / user / project)

The project memory directory is shown in your system prompt context — look for the path ending in /memory/. This is project-specific and changes per repo.

File naming: feedback_<slug>.md, project_<slug>.md, user_<slug>.md

Frontmatter:

---
name: Short title
description: One-line hook used to decide relevance in future sessions — be specific
type: feedback | project | user
---

Body structure for feedback:

Lead with the rule. Then Why: (the reason given or inferred). Then How to apply: (when this kicks in).

Body structure for project:

Lead with the fact or decision. Then Why: (the motivation). Then How to apply: (how this should shape suggestions).

Body structure for user:

Describe what you learned about the user's role, knowledge, or preferences in plain terms.

After writing each memory file, add a one-line pointer to MEMORY.md in the same directory under the right section heading:

- [Title](filename.md) — one-line hook under 150 chars

Check MEMORY.md first — update an existing entry rather than creating a duplicate.

Global skill files

When a lesson is a universal rule (language, framework, or tooling pattern), append it to the relevant skill file at ~/.claude/skills/. Keep additions concise — one rule, one short explanation.

CLAUDE.md

When a lesson is project-specific (this codebase's architecture, conventions, or patterns), add it to the CLAUDE.md in the current repo root. Put it in the most relevant section, or add a new one if needed.

Output to user

After saving, give the user a short summary:

• What you found (if anything)
• Where each lesson was saved
• If nothing was worth saving, say so clearly

Keep it to a few bullet points. No waffle.