Learned Lessons

Learned Lessons: Knowledge Base

Purpose

This skill accumulates knowledge from solved problems so you never search for the same thing twice.

When it activates

1. After web search for problem solving: offer to record the lesson
2. After using MCP tools: if a technical problem was solved
3. During debugging: check the knowledge base first
4. Explicit invocation: user says "/til" or "record a lesson"

Workflow after search/MCP

After successfully using web search or MCP tools to solve a technical problem:

1. Ask the user briefly:

`` Record this lesson to the knowledge base? (yes/no) ``

1. If yes, add an entry to the knowledge base file (default: ~/.claude/knowledge-base.md):

```markdown ## [YYYY-MM-DD] Brief title

Problem: What was not working

Symptoms: How it manifested (errors, behavior)

Solution: What helped

Tags: #python #asyncio #networking ```

1. Add entries at the TOP of the file (newest lessons first)

Workflow during debugging

When systematic-debugging activates or during explicit debugging:

1. Read ~/.claude/knowledge-base.md first
2. Search for similar problems by keywords and tags
3. If relevant match found, show the user
4. If nothing found, continue normal debugging

Entry Format

Each entry contains:

| Field | Description | |-------|-------------| | Date | When it was learned | | Title | Gist in 5-7 words | | Problem | What you were trying to do | | Symptoms | Errors, traceback, behavior | | Solution | Specific steps or code | | Tags | For search (#python, #docker, #api) |

Commands

• /til: record a lesson manually (asks for details)
• /lessons: show last 5 lessons
• /search-lessons : search the knowledge base

Important

• Do NOT record trivial things (typos, forgotten imports)
• Record what was NON-OBVIOUS and required research
• Record SPECIFIC solutions, not general advice
• Tags help search. Use technologies and concepts

Example

User: why does requests hang in asyncio?

Claude: [uses web search]
Claude: Found it! requests blocks the event loop. Use
        aiohttp or run_in_executor instead.

        Record this lesson to the knowledge base? (yes/no)

User: yes

Claude: Recorded with tags #python #asyncio #requests

Knowledge Base

The file ~/.claude/knowledge-base.md contains all accumulated lessons. Read this file during debugging to check for known solutions. Create it on first use if it does not exist.

Source & license

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

• Author: vasilievyakov
• Source: vasilievyakov/miracle-infrastructure
• License: MIT

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