learn-from-your-ask

Imported from my1057739543-max/learn-from-your-ask (SKILL.md). Install upstream with npx skills add my1057739543-max/learn-from-your-ask. Copyright stays with the author.

learn-from-your-ask

Capture the most recent valuable technical Q&A from the current conversation and add it to a local learning knowledge graph.

Output Locations

Use paths relative to the current workspace unless the user gives another target:

• Knowledge data: knowledge/knowledge.json
• Generated visualization: knowledge/index.html
• HTML template: templates/graph.html

Create knowledge/ and knowledge/knowledge.json if they do not exist.

Filtering Rules

Skip saving when any rule applies:

• The conversation is casual chat, greetings, status updates, or off-topic.
• There is no technical concept, code, tool, framework, architecture, debugging, or workflow knowledge.
• The user question is too short and contains no technical terms.
• The topic is only about operating Claude Code, agent settings, slash commands, or this skill itself.
• The assistant answer is a refusal or does not contain useful explanation.
• The point is a near-duplicate of an existing record in knowledge/knowledge.json.

Save when the exchange teaches or clarifies a reusable technical concept, workflow, implementation detail, debugging pattern, or design decision.

Knowledge Schema

Maintain this shape:

{
  "version": "1.0",
  "profession": "",
  "domains": {
    "Backend": {
      "FastAPI": [
        {
          "topic": "Dependency injection",
          "tags": ["python", "fastapi", "dependency-injection"],
          "date": "2026-06-29",
          "question": "Concise summary of the user's question.",
          "answer": "Concise summary of the assistant's answer.",
          "conversation": [
            {"role": "user", "content": "..."},
            {"role": "assistant", "content": "..."}
          ]
        }
      ]
    }
  }
}

Use English domain names by default unless the existing knowledge base already uses another language.

Common domains:

• Backend: APIs, server frameworks, databases, queues, auth, services
• Frontend: React, Vue, CSS, browser APIs, UI behavior
• Database: SQL, indexes, schema design, ORM behavior
• DevOps: Docker, Kubernetes, CI/CD, deployment, cloud
• Algorithms: data structures, algorithms, LeetCode, complexity
• System Design: architecture, design patterns, scaling, tradeoffs
• Programming Language: syntax, runtime behavior, async, decorators, type systems
• AI Engineering: prompts, agents, RAG, tool use, model integration

Workflow

1. Read knowledge/knowledge.json if it exists. If it does not exist, initialize the schema above with empty profession and domains.
2. Inspect the latest 1-3 meaningful user/assistant exchanges in the current conversation.
3. Apply the filtering rules. If nothing should be saved, tell the user no valuable technical knowledge point was detected.
4. Check for near-duplicates in the existing knowledge base by comparing domain, subdomain, topic, question, and tags.
5. Add one knowledge point under domains[domain][subdomain].
6. Read templates/graph.html, replace {{KNOWLEDGE_DATA}} with the complete JSON object, and write knowledge/index.html.
7. Tell the user what was saved, where it was categorized, and the path to knowledge/index.html.

Implementation Notes

• Preserve existing knowledge records.
• Keep knowledge/knowledge.json valid JSON with two-space indentation.
• Do not overwrite templates/graph.html.
• If templates/graph.html is missing, still update knowledge/knowledge.json and report that visualization generation was skipped.
• Prefer concise summaries over copying full long responses, but keep enough conversation context to reconstruct the learning point.