Skill Tuner

Improve skills over time by capturing user feedback and applying it on future runs.

When to Capture

When a user responds to a skill's output with a correction or forward-looking instruction, capture it. Look for phrases like:

• "Always...", "Never...", "From now on...", "Next time...", "Remember to..."
• "Don't include...", "Make it shorter", "Use bullet points", "Only show..."
• "Perfect, always do it like that"

Do NOT capture one-off requests ("just this once", "for now") or vague reactions.

How to Capture

1. Create the storage directory if it doesn't exist:
   • Use $SKILL_TUNER_DIR if set, otherwise .skill-tuner/ in the project root.
   • Add it to .gitignore if not already there.
2. Read {storage-dir}/{skill-name}.md if it exists.
3. If the new instruction contradicts an existing entry, replace it. If it refines one, update it. Otherwise append.
4. Write the file with this format:

# Skill Tuner: {skill-name}

## Learned

- [YYYY-MM-DD] {concise, actionable instruction}
- [YYYY-MM-DD] {concise, actionable instruction}

5. Confirm to the user: "Noted: {instruction}. I'll apply this next time."

How to Recall

Before running any skill, check if {storage-dir}/{skill-name}.md exists. If it does, read it and apply every instruction in the ## Learned section as additional constraints. Do not mention this to the user — just apply them silently.

Size Limits

Max 30 entries / 50 lines per file. When exceeded: merge duplicates, then remove oldest entries first. Override limits via {storage-dir}/config.json:

{
  "default": { "max_entries": 30, "max_lines": 50 },
  "skills": { "ai-news": { "max_entries": 20 } }
}

Direct Commands

• "What have you learned about {skill}?" — read and display the file.
• "Forget {instruction} for {skill}" — remove that entry.
• "Clear skill tuning for {skill}" / "Clear all skill tuning" — delete file(s).
• "Show skill tuning" — list all skills with entry counts.