kanly / learn

Learn

Mistakes are data. Capture what went wrong, why it happened, and how to prevent it next time. Build institutional knowledge instead of repeating failures.

What

Learn protocol formalizes learning from mistakes. When something breaks, gets reverted, or causes unexpected behavior:

  1. Log what happened
  2. Identify root cause
  3. Define correct behavior
  4. Store it for future sessions

When

Use Learn after:

  • A bug is fixed (especially recurring bugs)
  • A decision is reversed
  • Unexpected behavior is discovered
  • A pattern failure is identified
  • User says “remember this” or “don’t do this again”

How

  1. Describe what happened — The symptom/failure
  2. Identify root cause — Why it happened (not just what broke)
  3. Define correct behavior — How it should work instead
  4. Log to memory — Store in project memory or documentation
  5. Apply immediately — Fix the current instance using the new pattern

Example

User: "The API keys are exposed in the git logs again"

AI with Learn:
LEARN — Pattern failure detected

## What Happened
API keys committed to git (3rd occurrence this month)
Files: .env, src/config/secrets.ts

## Root Cause
.env file not in .gitignore, developers editing config files with inline keys
instead of referencing environment variables.

## Correct Behavior
1. NEVER commit .env files (add to .gitignore)
2. Config files should ONLY reference process.env.X, never hardcode keys
3. Use .env.example with dummy values for documentation
4. Pre-commit hook: fail if files contain patterns like "sk-", "Bearer ", etc.

## Logged To
- Updated .gitignore
- Added pre-commit hook (scripts/pre-commit-check.sh)
- Documented in CONTRIBUTING.md

## Immediate Action
- Removing keys from current commit
- Rotating exposed API keys (Stripe, OpenAI)
- Running git filter-repo to purge from history

This pattern is now in project memory. Won't repeat.
← back to protocols