Smart Donkey Learn

Smart Donkey Learn

Trigger the learning/distillation phase explicitly.

Steps

1. Read current brain file (if exists):

cat smart-donkey-brain.md 2>/dev/null || echo "NO_BRAIN_FILE"

1. Read session context:

• Read progress.md for what happened this session
• Read task_plan.md for decisions and errors
• Read findings.md for technical discoveries

1. Analyze the session for extractable lessons:

• What approaches worked well? (Save to "What Works Well")
• What mistakes were made and how were they fixed? (Save to "Mistakes to Avoid")
• Did the user express any preferences? (Save to "User Preferences")
• Any important codebase patterns discovered? (Save to "Codebase Patterns")
• Any debugging insights? (Save to "Debugging Playbook")
• Any architecture decisions made? (Save to "Architecture Notes")

1. Update smart-donkey-brain.md:

• If file doesn't exist, create it with the template
• If file exists, merge new lessons with existing ones
• Don't duplicate — update existing entries if refined
• Remove anything that's no longer accurate
• Keep total file under 200 lines

1. Present summary to user:

• Show what new lessons were extracted
• Show what existing lessons were updated
• Show total lesson count

Brain File Template

# Smart Donkey Brain
> Auto-generated learning file. Updated: [date]
> Sessions learned from: [N]

## User Preferences
(How the user likes to work)

## Codebase Patterns
(Key patterns, file paths, conventions)

## What Works Well
(Approaches that save time)

## Mistakes to Avoid
(Errors and their root causes)

## Architecture Notes
(Important design decisions)

## Debugging Playbook
(Issue pattern -> Solution approach)

Rules

• Be ruthlessly concise — 1-2 lines per entry
• Only save lessons that will be useful in FUTURE sessions
• Don't save task-specific details (they belong in progress.md)
• Facts > opinions. Mark uncertain items with "(uncertain)"
• If the brain file is over 200 lines, prune the least valuable entries

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: cloudyview
• Source: cloudyview/smart-donkey
• License: MIT

Install and usage instructions live in the source repository linked above.