# Learnings

This folder captures learnings from retrospectives and day-to-day work. Learnings serve three purposes:

1. **Historical record**: What we learned and when
2. **Governance reference**: Why we work the way we do
3. **Encoding source**: Input that gets encoded into skills and agents

## The Learning Flow

```
Experience → Learning captured → Encoded into system → Knowledge is actionable
                   ↓
            Stays here for:
            - Historical reference
            - Governance validation
            - Periodic review
```

Learnings are **not** the final destination. They are inputs that get encoded into skills and agents where Claude can actually use them. But we keep the learning file as a record of *why* we encoded what we did.

## Writing a Learning

Create a new file: `YYYY-MM-DD-short-title.md`

Use this template:

```markdown
# [Title]

**Date**: YYYY-MM-DD
**Context**: What triggered this learning (task, incident, observation)

## Learning

The insight we gained. Be specific and actionable.

## Encoded In

Where this learning has been (or will be) encoded:

- `skills/xxx/SKILL.md` - What was added/changed
- `agents/xxx/AGENT.md` - What was added/changed

If not yet encoded, note: "Pending: Issue #XX"

## Governance

What this learning means for how we work going forward. This is the "why" that justifies the encoding.
```

## Encoding Process

1. **Capture the learning** in this folder
2. **Create an issue** to encode it into the appropriate location
3. **Update the skill/agent** with the encoded knowledge
4. **Update the learning file** with the "Encoded In" references

The goal: Claude should be able to *use* the learning, not just *read* about it.

## What Gets Encoded Where

| Learning Type | Encode In |
|---------------|-----------|
| How to use a tool | `skills/` (background skill) |
| Workflow improvement | `skills/` (user-invocable skill) |
| Subtask behavior | `agents/` |
| Organization belief | `manifesto.md` |
| Product direction | `vision.md` (in product repo) |

## Periodic Review

Periodically review learnings to:

- Verify encoded locations still reflect the learning
- Check if governance is still being followed
- Identify patterns across multiple learnings
- Archive or update outdated learnings

## Naming Convention

Files follow the pattern: `YYYY-MM-DD-short-kebab-title.md`

Examples:
- `2024-01-15-always-use-comments-flag.md`
- `2024-01-20-verify-before-cleanup.md`
- `2024-02-01-small-prs-merge-faster.md`