auto-save-tips

Automatically save session tips when user provides important feedback or corrections

$ Installieren

git clone https://github.com/lambdalisue/dotfiles /tmp/dotfiles && cp -r /tmp/dotfiles/home/claude/skills/auto-save-tips ~/.claude/skills/dotfiles

// tip: Run this command in your terminal to install the skill


name: auto-save-tips description: Automatically save session tips when user provides important feedback or corrections

Auto-Save Tips Skill

This skill detects when the user provides valuable feedback and automatically saves it to .claude/TIPS.md.

Trigger Conditions

Activate this skill when the user provides:

  1. Corrections or Preferences:

    • "いや" / じゃăȘくど" / "ではăȘく" / "No" / "instead" / "actually"
    • "æŹĄă‹ă‚‰" / "ä»ŠćŸŒăŻ" / "remember"
    • "æ°—ă‚’ă€ă‘ăŠ" / "æłšæ„" / "ćż˜ă‚Œăšă«"
    • "prefer" / "should be" / "should not"
  2. Important Decisions:

    • Explicit choices between multiple approaches
    • Architectural or design decisions
    • Pattern or convention establishment
  3. Explicit Instructions:

    • "always do X" / "never do Y"
    • "from now on" / "going forward"
    • Statements about future behavior

Workflow

When triggered:

  1. Acknowledge the feedback:

    • Confirm understanding of the user's point
    • Briefly explain what you'll change
  2. Invoke /tips:save:

    • Use the Skill tool to execute /tips:save
    • This will capture the feedback and organize TIPS.md
  3. Continue the conversation:

    • Apply the feedback immediately
    • Proceed with the original task

Examples

Example 1: User Correction

User: じゃăȘくど、ă‚čăƒ©ăƒƒă‚·ăƒ„ă‚łăƒžăƒłăƒ‰ă«ă—ăŸă»ă†ăŒă„ă„ă‹

Claude: ăȘă‚‹ă»ă©ïŒăƒ•ăƒƒă‚Żă§ăŻăȘくă‚čăƒ©ăƒƒă‚·ăƒ„ă‚łăƒžăƒłăƒ‰ăźæ–čăŒè‰Żă„ăšă„ă†ă“ăšă§ă™ă­ă€‚
        [Uses Skill tool to invoke /tips:save]
        æ‰żçŸ„ă—ăŸă—ăŸă€‚ăă‚Œă§ăŻă‚čăƒ©ăƒƒă‚·ăƒ„ă‚łăƒžăƒłăƒ‰æ–čćŒă§ćźŸèŁ…ă—ăŸă™...

Example 2: User Preference

User: I prefer LLM-generated summaries over pattern matching

Claude: Understood! You prefer having the LLM review and write summaries
        rather than mechanical pattern matching.
        [Uses Skill tool to invoke /tips:save]
        I'll implement it that way...

Important Notes

  • Do NOT over-trigger: Only activate for genuinely valuable feedback
  • Do NOT interrupt flow: Save tips smoothly without disrupting the conversation
  • Be selective: Not every user comment needs to be saved
  • Common responses like "ok" or "thanks" should NOT trigger this skill