Kmg lesson capture

Skill: kmg-lesson-capture

Purpose: Auto-capture lessons when user solves complex bugs, makes breakthroughs, identifies patterns, or completes debugging sessions.

Trigger Keywords (detect any of these):

• "figured it out"
• "the fix was"
• "turns out"
• "solved it"
• "the problem was"
• "I found out why"
• "learned that"
• "pattern here is"
• "this is a gap"
• "the issue is" (when describing a process or UX failure, not a git issue)
• "noticed that"
• "this doesn't work because"
• Bug resolved, workaround found, first-time setup complete, pattern identified
• Completion of debugging/troubleshooting context
• UX or process failure observed and named during active use

Behavior: When triggered:

1. Pre-structure context from the conversation:

   • Extract the problem — what was wrong or what was being figured out
   • Extract the solution — what actually fixed it
   • Extract the pattern — any generalizable lesson or warning worth noting
   • Identify relevant tags from conversation (technology names, domain, etc.)

2. Note: When dispatching to lesson capture, the lesson-capture-agent includes a snapshot gate that offers to preserve session context first. This is presented to the user inside the agent flow — do not add a separate snapshot prompt here.

   ECC Compatibility: The slash command syntax (/kmgraph:kmg-capture-lesson) is Claude Code–specific. On other ECC platforms, the lesson-capture-agent is dispatched directly via agent invocation without a command namespace.

3. Dispatch to lesson-capture-agent with the pre-structured context as a named payload:

   context_provided: true
   problem: "[extracted problem]"
   solution: "[extracted solution]"
   pattern: "[extracted generalizable lesson]"
   tags: ["[tag1]", "[tag2]"]
   suggested_category: "[architecture|debugging|patterns|process]"
   

   The agent uses context_provided: true to skip its interactive wizard and go directly to draft generation.

4. Use friendly, user-addressed language — don't volunteer agent mechanics unprompted:

   • ✅ "Looks like you just solved something worth keeping — here's what I'd save:"
   • ✅ "I noticed you figured something out — want me to document that?"
   • ❌ "Dispatching to lesson-capture-agent..."
   • ❌ "Triggering agent with context..."

Do NOT trigger on:

• Behavioral process directives without a solved bug or learned pattern ("always X", "from now on X", "don't do X" as a standing rule) → handled by rules-capture skill. If both a lesson and a rule apply to the same turn, both skills fire independently.

Example Trigger:

User: "Figured it out! The issue was the config being cached in memory.
We needed to invalidate on every write."

Skill response:
"Looks like you just solved something worth keeping. Here's what I'd capture:

- Problem: Config being cached in memory between writes
- Solution: Invalidate cache on every write operation
- Pattern: Cache invalidation timing matters in multi-state systems
- Tags: [caching, memory, invalidation]

Want me to document this as a lesson?"