pafh-mini_skill

Overview

This skill implements the PAHF continual personalization framework for agents. It runs a three-step loop—pre-action clarification, preference-grounded action, and post-action feedback integration—to learn and apply user preferences over time. It emphasizes explicit memory records, consent for persistent storage, and safe handling of sensitive data.

How this skill works

Before acting, the skill asks brief clarifying questions when choices are ambiguous or preferences are missing. It retrieves relevant preferences from structured memory files to guide decisions and uses reasonable defaults when none exist. After acting, it collects explicit or implicit feedback and logs confirmed changes to the preference memory with timestamps and source markers.

When to use it

• User states or implies preferences or habits
• You must choose among multiple reasonable options
• User corrects or adjusts your behavior
• You need to remember personalized settings for future interactions
• You detect a possible change or drift in user preferences

Best practices

• Ask a concise, focused clarification when uncertainty would affect outcome
• Store stable preferences in long-term memory and transient observations in dated logs
• Request explicit consent before writing persistent preferences
• Never store sensitive data (passwords, financial, health)
• Record source and date for every learned or updated preference

Example use cases

• User requests reports in a specific format—confirm and save the format as a preference
• Responding to ambiguous tasks by listing clear options and applying the chosen preference
• Noticing a user prefers morning meetings and logging it to a daily memory file
• Detecting preference drift (e.g., concise → detailed) and asking whether to update the long-term setting
• Applying recent corrections immediately while logging them for trend detection